Keywords: Technology for Humanitarian Purposes, Industry 4.0, Innovation and Engineering Process for Bridging the Gender Gap

Keywords: Emerging memory technologies, Electronic devices, materials and fabrication process, Materials and Structures

Keywords: Entrepreneurship, Sustainable manufacturing process, Industry 4.0

Keywords: Business Intelligence, Crowd Sourcing & Social Intelligence

Keywords: Future-oriented and Personalized Educational Concepts, Digital Transformation

Keywords: Digital Transformation, Student-centered Learning Environments, Future-oriented and Personalized Educational Concepts
Abstract: The objective of this research was to enhance content recognition and application among 23 first-year master's degree international students in the Strategic Planning and Evaluation for Integrated Communications Course at the Faculty of Communication Arts. Utilizing tools like myCourseVille Learning Management System, multimedia slides created with Spark, and teleconferencing via Zoom, the study explored the impact of interactive media on short-term memory retention and creative content application. Drawing on the Cognitive Load Theory, the findings suggest that repetitive quizzes and tests stimulate short-term memory, enhancing content recognition. However, the study emphasizes the need for questions beyond mere memory testing to foster long-term retention and deeper understanding. Learner-teacher interaction through Zoom and practical learning activities were found to foster content application. The use of diverse interactive media, tailored to the learners' level, boosted participation but also revealed potential drawbacks such as student fatigue and decreased enthusiasm for excessive memory testing. Recommendations include reducing the frequency of memory-focused tests, incorporating more diverse question types, and utilizing interactive learning exchanges that facilitate conversational knowledge exchange and feedback on student work.

Keywords: Engaging Undergraduate Students in Research
Abstract: The objective of this study was to develop the oral pathology mobile application (PathoPal) to support dental students' learning process and to identify satisfaction of dental students in using the mobile application. Participants were the third- and fifth- year dental students (193 students) at Faculty of Dentistry, Chulalongkorn University. Students were given one week to use the application then their satisfaction in using the application were gathered via a questionnaire survey. Of 193 participants, 60 percent were satisfied with the application. When compared with textbooks, the most significant advantage was the convenience of use and the speed of data processing in identifying pathological information. Students reported that the application was more accessible than textbooks resulting in their learning process became more effective. In conclusion, the PathoPal application can be helpful for inexperienced dental students as it provides tentative differential diagnoses of all possible oral pathologic diseases supporting learning and clinical reasoning skills.

Keywords: Future-oriented and Personalized Educational Concepts, Digital Transformation, Student-centered Learning Environments
Abstract: Artificial intelligence (AI) technologies continually improve and become more pervasive in many facets of our lives. ChatGPT is a chatbot created by OpenAI with a conversational artificial intelligence interface. Academic institutions could routinely use artificial intelligence (AI) and language models like ChatGPT, with an increasing range of applications and ramifications. This study investigates the adoption of ChatGPT in academia which include applications, challenges and opportunities using the lenses of educational transformation, response service quality, usefulness privacy concerns. The article first examine diverse applications of ChatGPT, including automation, sentiment analysis and natural language processing. Second, it addresses the challenges and limitations that come with using these technologies, like regulatory compliance algorithmic prejudice, and ethical issues. Third, the study emphasize the opportunities brought about by the implementation of AI and ChatGPT, such as improved research capacities, individualized learning experiences, and new career pathways. To promote an efficient and responsible adoption and deployment of ChatGPT, the study's findings offer several research directions and implications in academia.

Keywords: Multidisciplinary and Transdisciplinary Education
Abstract: A collaborative course called Global Awareness for Technology Implementation was conducted by the Faculty of Engineering, Chulalongkorn University, Thailand, and the Tokyo Institute of Technology, Japan. The objective of this course was to facilitate the exchange of knowledge, perspectives, and experiences between Thai and Japanese students through project work, using an engineering design approach and the design thinking process to create innovation relevant to the Sustainable Development Goals (SDGs). During the class, experts shared their knowledge and experiences with the students. The students then worked on their respective projects to develop solution concepts using the design thinking process. Finally, this study concluded with a summary of the results and findings on the empathy process in the chosen SDGs, and of the commonalities and differences between Thailand and Japan in terms of ideation.

Keywords: Crowd Sourcing & Social Intelligence, Neural Networks and Deep Learning, Machine Learning
Abstract: The advent of social media has transformed the way we communicate and connect, enabling individuals worldwide to instantly and openly interact with friends, family, and colleagues on a frequent basis. People utilize social media platforms as a means to express their opinions, share personal experiences, narratives, and challenges. Nevertheless, concerns have arisen due to the growing prevalence of suicidal content on social media platforms, where discussions of hardship, thoughts of death, and self-harm are widespread, particularly among younger generations. Consequently, harnessing the power of social media to detect and identify suicidal behavior, including the presence of suicidal thoughts, becomes essential in offering appropriate interventions that discourage self-harm and suicide, as well as in preventing the spread of suicidal ideations throughout these platforms. This paper presents suicidal content detection using two deep learning architectures, LSTM, and DistilBERT with the latter showing better performance in respectively. We conclude by drawing implications for deep learning architectures in detecting suicidal content on social media and initial deployment of the models using Telegram bot which detects the message containing suicidal content and sends a motivational message in response and also informs their friends and relative through alerts.

Keywords: Machine Learning, Neural Networks and Deep Learning
Abstract: A rapid increase in users on social media has given rise to a vast amount of user-generated content, including hate speech and offensive language. Such content can have serious negative consequences, ranging from psychological harm to inciting violence and discrimination. Existing studies have explored different deep learning and Natural language processing (NLP) methods to perform hate speech detection, and these solutions have yielded significant performance. Most existing solutions are limited to detecting hate speech only in English with less focus on content generated in other languages, particularly in low-resource or regional languages. The goal of this paper is to address this challenge of hate speech detection for low-resource languages and propose a tool that could provide a real-time prediction for social media posts. In this study, the main focus was on English, Hindi, Hinglish, Bengali, and Marathi languages which are commonly used in social media platforms in India. A meta-learning-based model was employed to perform hate speech detection in these languages. The proposed method helps overcome the limitation of data scarcity and provides fast adaptation to an unseen target language. Extensive experiments were conducted on datasets comprised of different regional languages spoken in India. Accuracy, Precision, recall, and F1-score metrics are used to evaluate the model’s performance. The results show that when the dataset size is small, meta-learning-based models

Keywords: Machine Learning, Neural Networks and Deep Learning
Abstract: Video surveillance is a tiresome task to perform by a human. Anomaly events are factored by body gestures and head movements of a person. Body gestures of a crowd or a person are only considered for video surveillance when they are in a detectable range from the camera and head movements are considered when a person’s face is close enough to the video surveillance, so the necessary facial features are examined. In this paper, we aim to detect abnormal activities by first calculating the facial distance to trigger the necessary modules based on the generated focal length through the camera. The abnormal activity detection module is triggered when the threshold distance is crossed and the head movement module is called when the face detected is close enough to the camera. The abnormal event notification is promptly sent to the entitled person so that necessary actions are taken.

Keywords: Machine Learning, Neural Networks and Deep Learning
Abstract: Large language models (LLMs) have attracted considerable interest in the fields of natural language understanding (NLU) and natural language generation (NLG) since their introduction. In contrast, the legacy of Capsule Neural Networks (CapsNet) appears to have been largely forgotten amidst all of this excitement. This project's objective is to reignite interest in CapsNet by reopening the previously closed studies and conducting a new research into CapsNet's potential. We present a study where CapsNet is used to classify toxic text by leveraging pre-trained BERT embeddings (bert-base-uncased) on a large multilingual dataset. In this experiment, CapsNet was tasked with categorizing toxic text. By comparing the performance of CapsNet to that of other architectures, such as DistilBERT, Vanilla Neural Networks (VNN), and Convolutional Neural Networks (CNN), we were able to achieve an accuracy of 90.44%. This result highlights the benefits of CapsNet over text data and suggests new ways to enhance their performance so that it is comparable to DistilBERT and other reduced architectures.

Keywords: Neural Networks and Deep Learning, Human Computer Interface, Machine Learning
Abstract: Affective computing aims to facilitate effective communication between humans and machines. Many affective computing systems use machine learning models trained on labeled data, like images and videos, to recognize emotions. Among these, the Generative Adversarial Network (GAN)-based expression GAN (ExprGAN) stands out as it can generate faces displaying various expressions. However, the generated faces often lack clarity in crucial facial features, such as the eyes and lips, which are essential for defining facial expressions. To address this issue, a novel feature extraction block is proposed. This module incorporates two parallel channels with multi-scale dilated convolution to mimic the human visual system and extract multi-scale facial features from the facial images. Additionally, an unsharp masking filter is integrated into the pre-processing stage to enhance the quality of facial expression features, making them sharper and clearer. The proposed model can generate faces with six distinct expressions: Anger, Disgust, Fear, Happiness, Sadness, and Surprise. To evaluate, a Convolutional Neural Network (CNN)-based expression classifier, a Principal Component Analysis (PCA)-based face classifier, and the FID (Frechet Inception Distance) score are used for comparison. The results demonstrate that the proposed model outperforms the existing ExprGAN, providing better quality and more expressive faces, which can be beneficial for improving human-machine interaction.

Keywords: Machine Learning, Data Mining, Neural Networks and Deep Learning
Abstract: Short Message Service (SMS) is a generally used communication method due to its convenience and affordability. SMS spam message is an unauthorized text message that contains a variety of content types such as advertisements, fraudulent texts, and promotions. These messages can pose a serious threat to mobile phone users as they may contain security threats, malicious activities, and other concerning issues. These can lead to identity theft, financial loss, and other types of fraud. To deal with the problem of spamming, various machine-learning models are applied to develop an optimized model that effectively, reliably, and precisely identifies and filter out spam or junk message from a genuine SMS text. The dataset used is a combination of self-acquired data and internet collected dataset with 60-40 ham to spam partitions. With regards to the accuracy of the model, the Bernoulli Naive Bayes achieved the highest performance with 96.63% accuracy upon optimization.

Keywords: Machine Learning, Data Mining
Abstract: Federated Learning (FL) in Machine Learning (ML) has become very popular nowadays because it trains classifiers across multiple decentralized devices without transferring data to a central server. It is a decentralized learning approach that amalgamates several different nodes into one. In 2016, Google introduced the concept of FL when the use and misuse of personal data were gaining global attention. In this paper, we have proposed a privacy-preserving approach for mining big personalized data employing scalable decision tree induction with FL. The concept of the RainForest framework refers to addressing big data challenges via a Decision Tree (DT) classifier. The proposed method does not share the data or personal information with the central server. It only transfers the local models' parameter values to the central server. Each individual device or node trains its own local DT classifier and shares the prior and conditional probability values with the central server. We have tried to simulate the proposed concept using five benchmark datasets. The results of the evaluation indicate that the model exhibits exceptional performance and accuracy.

Keywords: Machine Learning, Human Computer Interface, Neural Networks and Deep Learning
Abstract: This work proposes a cloud-based attendance system that uses face recognition technology to authorize identity. The system uses the Amazon Web Services (AWS) Rekognition service and a serverless architecture. The proposed system provides a reliable and tamperproof solution to track attendance, eliminating the need for manual record keeping and minimizing human involvement. It also offers potential benefits, such as improved security and transparency in attendance management

Keywords: Machine Learning, Neural Networks and Deep Learning
Abstract: The limited physical constraints of nanosatellites due to their size, hinders their ability to transmit large amounts of image data. Because of this, the use of machine learning methods to filter data onboard has become more prominent to increase the bandwidth efficiency of these devices. By having an AI-based classification system for the images, the bandwidth necessary to transmit all these images and the tradeoff when it comes to storage, can potentially be offloaded through having a system which generates metadata that can indicate the data samples which offer the most usability, thus freeing up more space and bandwidth for these more important samples.

This study explores the task of land cover classification, by utilizing one of the more prominent image segmentation models, U-Net. The model is implemented and evaluated using Pytorch using the DeepGlobe 2018 land cover classification dataset, achieving an average class IoU score of 0.68. This study seeks to support the viability of such a solution and is intended to support any future work which seeks to implement a fully automated data prefiltering system for satellite imagery.

Keywords: Machine Learning, Neural Networks and Deep Learning
Abstract: The development of machine learning methods for onboard satellite processing is important in order to facilitate the filtering of collected data samples to maximize the use of the device’s limited resources. Land cover classification can be used to focus the collected data on certain terrain types by utilizing classification methods to determine the class probabilities of individual pixels in a collected satellite image. The importance of the accuracy of the segmentation model used for such a task is important in order to avoid the trashing of data samples that offer significant information and the prioritization of data samples which offer less in terms of usable information, which in the case of land cover classification is determined by which terrain features may be prioritized over others.

This study focuses on the U-Net segmentation architecture and performs an experimental study on the effects on two aspects on the training of a segmentation model for increased performance. This includes the division of the images in the dataset into smaller patches and the replacement of the CNN encoder of the segmentation architecture. The changes made to the baseline model introduced an increase in the IoU score from 0.68 to 0.7273.

Keywords: Ad-hoc, Mesh and Sensor Networks, Sustainable Communications, Wireless Communications and Networks
Abstract: In the present era, networking has become an essential requirement as individuals seek interconnectivity and efficient means to share data rapidly. When discussing the swift exchange of information, a network characterized by high speed and minimal congestion is favored by all. In networking, Switches play a very important role to transfer packets of information from source to destination, as routers do in any network topology. For switches to exchange data from the source port of one switch to the destination port of another switch, it requires a special arrangement where every switch is connected to another using chassis topology. When packets transfer, in certain cases, it leads to congestion in the output port due to heavy traffic which triggers a notification called QCN (Quantized Congestion Notification) to the source port. As the packets are received from different sources, the notification needs to be replicated to all the source ports sending to the congested target. This paper proposes the analysis of QCNs based on different cases using an advanced replication architecture i.e., BIER (Bit Index Explicit Replication) to study its effect to achieve high throughput and low latency of the whole system and proposes a novel approach called periodic QCN for switches in chassis topology

Keywords: Ad-hoc, Mesh and Sensor Networks
Abstract: MQTT is a simple message protocol with the publish/subscribe model, which enables loosely coupled communication. In large-scale systems, load balancing is required because of the problem of concentrated load on brokers and subscrebers. However, existing load balancing methods have difficulty in distributing the load according to the load status of broker and the sending frequency of publisher, and there is a problem of a single point of failure.

In this study, we propose a load balancing method using server redirection specified in MQTT v5.0 to improve the problems of existing methods. In order to verify the effectiveness of the proposed method, we conducted comparison experiments with DNS round-robin, an existing method, and measured the throughput and CPU utilization of each broker as evaluation indices.

Keywords: Vehicular Networks, Wireless Communications and Networks
Abstract: In response to the explosive growth of data computation in vehicular terminals, computation offloading has emerged as a viable solution to mitigate the limitations of vehicle resources. Efficient offloading decisions not only meet the demanding requirements of complex vehicular tasks in terms of time, energy consumption, and computational performance but also minimize competition and resource consumption in the network. Research on vehicular network task offloading is extensive. However, the existing body of work often exhibits certain limitations, such as incomplete consideration of relevant factors or suboptimal utilization of available resources. This research presents the construction of a three-layer vehicular network environment, which is based on cloud and edge computing paradigms. The design entails the formulation of real-time vehicle location tracking and task priority metrics, while also considering the challenges posed by time-varying channels and signal blockage prevalent in vehicular network environments. And we propose a novel variant of the Deep Deterministic Policy Gradient (DDPG) algorithm Deep Deterministic Policy Gradient (NLDDPG) to iteratively train the model, aiming to optimize a weighted objective function. And results demonstrate the algorithm's efficacy in improving the overall effectiveness and precision of decision-making processes. Furthermore, the application of NLDDPG successfully reduces the average task utility of the current policy.

Keywords: Wireless Communications and Networks, Cellular Networks
Abstract: Network traffic has changed significantly since the introduction of 5G technology. The massive flow of connected devices as well as new applications are causing a variety of traffic patterns, quality of service (QoS) requirements and scalability challenges. The solution to these problems is network slicing which enables operators to distribute resources and set network characteristics per slice by constructing multiple virtual network slices within the shared 5G infrastructure. Resources are reserved for each slice for some time period. Traffic prediction with high accuracy is of great importance in the dynamic environment of 5G networks for resource planning and scheduling as well as for reliable and effective transmission of network data. In this paper, We generate user requests in a slice having arrival and departure rates using Poisson distribution. We did a comparative analysis of traffic prediction and resource allocation using deep learning models such as Long Short Term Memory(LSTM), Bidirectional LSTM(BiLSTM), Stacked LSTM and Gated Recurrent Unit(GRU).

Keywords: Wireless Communications and Networks
Abstract: The Cloud Radio Access Network (C-RAN) is an innovative technology with great promise for minimizing wireless network deployment and maintenance costs. In this study, our main goal is to reduce the costs associated with functionally placing the RAN while accounting for the computational expense and the front-haul bandwidth usage among various users. To achieve this, we propose to apply Particle Swarm Optimization (PSO) to achieve effective allocation of computational resources and the front-haul bandwidth, ensuring an efficient and cost-effective C-RAN design. Experimental results on different traffic have shown that the proposed PSO can provide cost-effective design of the C-RAN as compared to the optimal solution from the integer linear programing (ILP) approach.

Keywords: Biomedical Imaging and Video analytics, Biomedical and Health Informatics
Abstract: The degree of retinal blood vessels’ tortuosity may indicate the progression of various diseases. Therefore, observing the changes in tortuosity levels in retinal fundus images can be an effective biomarker for diagnosing diseases. Several studies have proposed parameters or methods to measure blood vessel tortuosity, but these methods still have some drawbacks that can be improved to better describe tortuosity both in retinal arteries and veins. This study aims to propose tortuosity measurement parameters that can increase the correlation between tortuosity assessments from ophthalmologists and automatic tortuosity measurement results. This research led to the development of a combined parameter incorporating some features that are included in previous studies, such as the comparison of arc length and chord length, as well as the angle values at the critical points of the curve structure. The proposed parameter has been applied to a retinal blood vessels dataset that consists of 60 fundus images (30 artery images and 30 veins images with similar length and caliber). The automatic measurement of each image’s tortuosity level are compiled, ranked, and compared to the ground truth tortuosity ranks of each image provided in the dataset. The proposed parameter for retinal blood vessels tortuosity measurement achieved a Spearman rank correlation coefficient of 0.886 for arteries and 0.884 for veins.

Keywords: Biomedical Imaging and Video analytics
Abstract: This paper presents an automated measurement of retinal artery and vein blood vessels using the state-of-the-art deep learning architectures. The measurement of the artery-vein ratio plays a vital role in predicting intracranial pressure (ICP) in traumatic patients. In the proposed method, the artery-vein and optic cup-to-disc (OCD) information are extracted from the retinal fundus imaging using the deep CNN (D-CNN). The process involves the preprocessing of the retinal fundus image to highlight the vessels information and OCD more clearly. Further, the feature extraction of the vessels and OCD is performed using the base architecture of D-CNN. The extracted vessels and OCD determined the artery-vein ratio. The performance of the segmented artery-vein and OCD is evaluated and analyzed. Validation of measured artery-vein and OCD has been done by comparing these values with the ground truth values. The accuracy of the segmented artery-vein is determined to be 95.21 for the HRF dataset and the segmented optic cup and disc found to be 94.70 and 92.36 respectively for Drishti dataset. The extracted feature of artery-vein and OCD determines the artery-vein ratio using the connected component analysis. The algorithm generated measured value is compared with the manually generated value by the two observer for the artery-vein ratio. The average error for the INSPIRE-AV dataset on total 40 images is found to be 0.15.

Keywords: Biomedical Imaging and Video analytics
Abstract: Diabetic retinopathy, one of the main causes of blindness in the world, can be diagnosed early by utilizing microaneurysms (MA). The existence of microaneurysms can be detected by performing semantic segmentation on the retinal fundus image. In this paper, deep learning method testing was carried out using the GlobalNet model at various input image resolution values to see the effect of resolution alteration on MA detection in retinal fundus images. The experiment provided AUPR values of 0.391 ± 0.026, 0.387 ± 0.035, and 0.394 ± 0.050, along with F1 scores of 0.361 ± 0.034, 0.361 ± 0.022, and 0.360 ± 0.022. Evaluation done at pixel and lesion levels shows that the difference in resolution of the input images does not cause a significant change in the AUPR and F1 scores. However, the resolution alteration does affect the total number of false positive lesions in outputs. In addition, a combination method is developed, providing a better trade-off in sensitivity and precision compared to the GlobalNet segmentation model.

Keywords: Biomedical Imaging and Video analytics
Abstract: Glaucoma is a vision-threatening condition resulting from increased intraocular pressure, leading to optic nerve damage and potential vision loss. Early detection is crucial, but in Indonesia, over half of glaucoma cases are diagnosed at severe stages. One of the approaches for early detection is by using characteristics visible in a fundus image as indicators. With this approach, the segmentation of optic disc and optic cup plays an important role to extract features such as cup-to-disc ratios or rim-to-disc ratios as an indication for glaucoma. Almustofa (2021) proposed an automated glaucoma detection method using these segmentations but only tested it on limited datasets, namely Drishti-GS and REFUGE Training Set. This study presents an optimized method by incorporating the REFUGE Validation and Test sets. Optic disc segmentation attains F-Scores of 0.979 ± 0.005 for Drishti-GS and 0.942 ± 0.026 for REFUGE. Optic cup segmentation achieves F-Scores of 0.948 ± 0.020 in Drishti-GS and 0.843 ± 0.068 in REFUGE datasets. These results demonstrate the improved performance of the optimized method for glaucoma detection.

Keywords: RF and Microwaves in Medicine and Biology, Antennas, Propagation and Computational EM, RF/Millimeter-wave Circuits and Systems
Abstract: A compact dual-band implantable multi-slot patch antenna for biotelemetry applications is proposed here. The proposed dual-band antenna operates at both WMTS frequency band of 1395-1400 MHz as well as the 5.8 GHz ISM band. The size of the dual band antenna is greatly reduced by inserting multi-slots into patch and ground plane in an asymmetric patterns. The proposed antenna occupies a compact size of 12 X 12 X 0.64 mm3 . The antenna achieves gains of -20.29 dBi and -8.52 dBi at dual-band of operation. The proposed antenna has the potential to be employed in different implantable biomedical devices for real-time monitoring of physiological data due to its small size and dual-band characteristics. Specific Absorption Rate (SAR) is also analysed to ensure human safety conditions.

Keywords: Audio / Speech / Spoken Language Processing, Digital & Multirate Signal Processing, Signal Processing Algorithms and Architectures
Abstract: Accent Variation Classification is the technique of detecting an accent or dialect of a human speech based on speech patterns and features from speech. This is useful in developing speech recognition systems, language learning systems, dialect preservation systems, sociolinguistic studies, voice assistance, improving speech synthesis and voiceover systems. It can be used in conducting forensic analysis on audio data to determine regional origin or specific accent traits. Furthermore, it is a useful tool in criminal investigations and judicial actions. Deep Neural Networks (DNNs) are utilized for speech recognition tasks because they can successfully learn complex variables of speech input such as patterns, intensity, rhythm, and temporal information. In this study, we propose Zero Crossing Rate (ZCR), Mel Frequency Cepstral Coefficients (MFCC), Root Mean Square (RMS), Mel-Spectrogram based feature extraction and DNN based Bangla Speaker Accent Variation Classification model to classify the speaker’s variation from Bangla Speech data. We train our model with 7443 audios from 9303 audios (Formal, Dhaka, Khulna, Barisal, Rajshahi, Sylhet, Chittagong, Mymensingh and Noakhali) and our model achieves 94% accuracy from unseen or new data. We compare its accuracy and performance with other neural networks where LSTM, Stacked LSTM and DCNN achieve accuracy of 67%, 71% and 85% respectively.

Keywords: Image / Video / Multimedia Signal Processing, Pattern Recognition and Object Tracking, Digital & Multirate Signal Processing
Abstract: In recent years, human skeleton tracking technology has attracted a lot of attention in the fields of virtual reality, human-computer interaction and medical rehabilitation. Human skeleton tracking technology is the basis for building human models in virtual reality scenarios. Among them, Kinect camera is widely used as a motion tracking sensor for virtual reality human-computer interaction. However, many current studies on skeleton point tracking are limited to single or dual camera systems, which leads to problems such as occlusion, missing skeleton data and errors. To solve the problems of limited capture range and data occlusion of a single Kinect camera, this paper proposes a skeleton point tracking method based on multiple Kinect cameras. The method uses multiple Kinect cameras to track the 3D coordinates of 32 body joints simultaneously, and unifies the joint coordinates captured by multiple Kinect cameras in different viewpoints into the same world coordinate system through coordinate transformation. The BP (Back Propagation) neural network is used to train the skeleton data from multiple viewpoints, thus generating a reliable user skeleton position in real time. By this method, the problems of the existing methods for obtaining skeleton points in a single camera view are solved.

Keywords: Image / Video / Multimedia Signal Processing, Pattern Recognition and Object Tracking, Signal Processing Algorithms and Architectures
Abstract: As we are aware that thousands of fires break out every day around the world, which results in high numbers of casualties and serious threat to property safety and forest vegetation. Hence, it becomes particularly important to detect the fire at its early stage, because once the fire has spread in an area, it gets cataclysmic and difficult to control. In particular, the early detection of fire is associated with rising smoke. Therefore, the smoke can be considered as a good indicator to predict fire. In the presented work, smoke detection is performed with the help of its dynamical features. The dynamical features are considered in the form of optical flow color map. The motivation of this work is to use fractional order optical flow instead of images to provide the precise location and rate of growth. The estimation of optical flow is carried out using a non-local weighted fractional order variational model, which is capable in preserving dynamical discontinuities in the optical flow. Optical flow helps to find the active region of the images (video). This non-local weight also incorporates the robustness against noise. Further, the optical flow field is converted into a color map using an RGB color wheel. These color maps are used in different deep learning models for training and testing. The experiments are conducted on a dataset consisting of 18 smoke and 17 non-smoke videos. Different accuracy metrics are used for performance evaluation and comparison study.

Keywords: Image / Video / Multimedia Signal Processing, Pattern Recognition and Object Tracking, Signal Processing Algorithms and Architectures
Abstract: As we are aware that the world witnesses a huge number of fire breakouts everyday, which results in high numbers of hazardous events and severe losses to property and forest vegetation. Therefore, early stage fire detection is of vital importance, for once it spreads it becomes unmanageable and disastrous. The early detection of fire can be performed with the help of vision based deep learning techniques. The novelty of the work lies in performing the fire detection using the static and dynamic features of fire. The static fire features are taken as shape, texture, and color, while the dynamic feature accounts for its flickering motion. For this purpose, the fire motion is estimated in terms of optical flow from videos (image sequences) by using a motion edge preserving level set segmentation based fractional order variational model. Level sets provide nicely segmented flow fields, while fractional order derivatives are capable to deal with discontinuities in the motion field. The estimated optical flow field is used to derive four fire features, which are constituted as 4D vectors. These 4D vectors reduce the data dimensionality and mitigates over-fitting problem. Finally, the fire detection is carried out by implementing a mixed data CNN-LSTM model. The mixed data presented in the work is composed of a reference image frame and the corresponding 4D vector sequence. Also, the significance of the model is manifested through an ablation study.

Keywords: Image / Video / Multimedia Signal Processing, Pattern Recognition and Object Tracking
Abstract: Vision sensors are widely used to perceive the environment for Advanced driver assistance systems in which computer vision algorithms play a vital role. The performance of such algorithms depends on features such as edges, sharpness and contrast of the captured images. Different imagers are tuned subjectively for the image quality and hence, there is a need for developing reliable Image Quality Assessment (IQA) metrics to retain the performance of computer vision algorithms irrespective of imagers. In this work, an overall image quality score is proposed as a linear combination of five individual image quality metrics. Images from open source datasets such as KITTI and nuScenes are considered as captures from two imagers. Spatial correlation and mean opinion scores are used to evaluate the accuracy of the proposed metric. The performance of an object detection algorithm is also evaluated using KITTI and nuScenes images resulting in 81% and 51% precision respectively. It is observed that a combination of VQA, BRISQUE, and JNB IQs provides the highest correlation with MOS with correlation coefficients of 0.9433 & 0.921 respectively for KITTI & nuScenes imagers. Enhancing the sharpness and blur of nuScenes images resulted in increased precision and correlation of 89% and 0.9524. This shows that proposed IQ metric evaluates image quality effectively. Further, defining a IQ metric will be beneficial to retain the performance of CV algorithms irrespective of imagers.

Keywords: Future-oriented and Personalized Educational Concepts, Digital Transformation
Abstract: Personalized learning paths are created to optimize learning time and improve student learning performance by providing an appropriate learning sequence for each student based on their different characteristics. A widely used method for building personalized learning paths is based on the student's knowledge but ignores the student's interest in the material topic. This study uses a deep learning and rule-based approach to recommend appropriate material based on the topic's difficulty level, student interest and knowledge level. The topic's difficulty level is predicted using deep learning. The level of student interest is collected using a questionnaire and then processed using a rule-based to create a learning path. A dynamic learning path is modelled by measuring student knowledge level in every topic and updating the learning path. Evaluation is measured by comparing the learning outcomes of students who used conventional e-learning with those who followed a personalized learning path. The results showed that students scored 29%, or 15.06 points, better than those who used conventional e-learning

Keywords: Virtual and Remote Labs and Classrooms, Digital Transformation, Student-centered Learning Environments
Abstract: This paper investigates the impact of virtual reality (VR) on student engagement in engineering education and their potential in enhancing the student learning experience through technology-led learning. The study evaluated the efficacy of VR technology as an immersive learning tool. The survey collects data from a diverse sample of students who experienced the use of VR-based flight simulator as a part of their continuous assessment, providing valuable insights. Conditional results indicate that by using VR headsets 70% of students reported improved learning outcomes for the module, while 100% students agreed that VR technology offered a more immersive learning experience. The survey results indicate and emphasize the potential benefits that integrating VR headsets into engineering education could bring to the student learning experience through enhanced student engagement and the acquisition of practical skills in simulated immersive environments. The paper also makes recommendations for further research and implementation of VR technology in other fields of STEM education.

Keywords: Student-centered Learning Environments, Engaging Undergraduate Students in Research, Digital Transformation
Abstract: The digital revolution has made access to information very easy. The onset of the COVID-19 pandemic also called for further digitization. Every organization; be it an office, an educational institute or a government entity, was forced to shift to an all virtual mode of operation. This led to the conduct of online examinations with very little time for formulating an anti-cheating examination pattern. Audio and video proctoring tools are considered helpful but are very expensive and do not provide a method to detect plagiarism in the handwritten text. This is a serious problem for academic enterprises and institutes where there is a need for plagiarism detection in the submitted assignments, answer-scripts against the information available on the Internet as well as against other submissions. This paper presents a plagiarism detection system for handwritten text in English. The proposed system uses authentication tools/services, cloud storage, and optical character recognition (OCR) services to automate the process of checking plagiarism between two handwritten documents, as well as plagiarism with respect to all information available online.

Keywords: Student-centered Learning Environments, Future-oriented and Personalized Educational Concepts, Game-based Learning and Gamification
Abstract: This paper presents the design and implementation of a pipelined Lessen Extrinsic Garrulity (LEGv8) architecture simulator, subset of ARMv8 architecture. The simulator is developed as a web application that can simulate the assembly of instructions in the LEGv8 assembly language, the execution of the instructions, and the visualisation of data path during execution. The simulator supports both single cycle and pipelined execution, with the option to select the control and datahazard handling methods to use. Users will be able to analyse the changes in the registers and memory, branching behaviour, hazard detection and elimination, as well as visualise data flow when stepping through instructions. This gives users the freedom to comprehend computer architecture more easily at their own pace by making use of the user-friendly and interactive educational simulator to enhance their understanding beyond what can be taught in the classroom.

Keywords: Digital Transformation, Multidisciplinary and Transdisciplinary Education, Future-oriented and Personalized Educational Concepts
Abstract: The rapid advancements in AI have transformed industries, increasing productivity. However, they have also widened the gap between developed and developing countries, necessitating digital skills to bridge the divide and ensure equal access to opportunities. This research examines the distribution of digital skills of the daily internet users in Myanmar, revealing that participants demonstrated Intermediate level of digital skills in various areas, with lower scores in communication, collaboration, and digital content creation. Age, education, and gender weakly correlate with digital skills. Older individuals have slightly lower proficiency, while high school graduates score lower on average but perform well in information/data literacy and digital content creation. Males score higher in data literacy, problem-solving, and digital safety, while females excel in digital content creation, showcasing their creativity. On the other hand, digital access, particularly Wi-Fi availability at home, and financial capability, strongly correlated with higher scores. Addressing these disparities requires prioritizing access to digital resources and implementing educational initiatives for marginalized groups, focusing on digital content creation and communication skills.

Keywords: Neural Networks and Deep Learning
Abstract: Human action identification is a practical area of study with broad applicability in various domains, such as medical care, sport science, and manufacturing management. In logistics, it is essential to identify and examine individual actions, enabling machines to perceive and comprehend human motions for non-verbal interaction. This study specifically focuses on efficiently classifying working activities in the logistics industry using wearable sensors, particularly in the context of human activity recognition. To achieve the research objective, a deep residual neural network was introduced, integrating convolutional layers, shortcut connections, and aggregated transformation for human activity recognition in logistics. The authors evaluated the effectiveness of their proposed deep learning model using the publicly accessible LARa dataset. The LARa dataset comprises a diverse range of human actions in the logistics domain, including standing, walking, cart handling, and synchronization. The details of activity were captured using wearable sensors affixed to different anatomical sites of the study participants. The experimental findings indicate that the model achieved a maximum F-measure of 85.30%.

Keywords: Neural Networks and Deep Learning, Machine Learning
Abstract: Understanding pedestrian behaviour when crossing the road is an important key to the development of autonomous vehicles. Because pedestrians are considered Vulnerable Road Users (VRUs), they are likely to be killed if they are involved in an accident. To ensure their safety, it is then necessary to predict the pedestrian’s intention based on their behaviour. In this experiment, we propose head pose observation for predicting their intention, by observing pedestrians’ head pose data, we can then predict their intention to cross the road. To achieve this purpose, we use human head detector and head pose extraction feature, and the resulting yaw, pitch and roll as three head pose features. To select the most optimal feature is important for predicting pedestrian intention, then we make 7 combination scenarios based on these three features and compare it with the same model. Based on this scenario, it is proved that all these three data are optimal to observe pedestrian intention. There are three behavioural annotation that have been used, there are crossing, not crossing and will crossing. We derive will crossing from the annotation looking and not crossing while waiting at the roadside. Prediction of pedestrian behaviour is done by using LSTM model, and the resulting precision on crossing and not crossing with 0.98, while will crossing is 0.94.

Keywords: Neural Networks and Deep Learning
Abstract: Research findings have unveiled that facial expressions possess the ability to convey a variety of intense emotions. Hence, in this study, a deep-learning based approach, 2-Dimensional Convolutional Neural Network (2D CNN) for facial emotion recognition is proposed. The proposed network is running at least 47.28 times lesser number of parameters at 542,136, compared to the state-of-the-art (SOTA) network from RAVDESS dataset. The saving from reduced parameters is expected to translate into faster execution in real time. The proposed network scored accuracy of 92% and 94% that outperformed majority of the SOTA networks trained on RAVDESS and SAVEE dataset respectively, except one LSTM network from RAVDESS dataset that scored 98.90% in accuracy but with 116.5x higher number of parameters.

Keywords: Neural Networks and Deep Learning, Machine Learning
Abstract: The recent advancement of fake face creation and fake face generation motivates the development of an excellent fake face detection method that can effectively detect the difference between fake and real. Various fake detection methods are available with adequate performance, but the limitation of those available methods is they are not performing well with highly compressed images with degraded quality. Manipulation of face images is getting advanced, and becoming difficult to trust the content over the media, and generating and detection should go parallelly to balance society. Therefore we are proposing a novel approach to solve this problem which uses the hybrid model HF-Detect, which combines the advantage of the Xception network along with the F3 Net.

Keywords: Neural Networks and Deep Learning, Machine Learning
Abstract: Public transportation is a crucial aspect of passenger transportation, with buses playing a vital role in the transportation service. Passenger counting is an essential tool for organizing and managing transportation services. However, manual counting is a tedious and time-consuming task, which is why computer vision algorithms are being utilized to make the process more efficient. In this study, different object detection algorithms combining with object tracking are investigated to compare passenger counting performance. The system employs the EfficientDet algorithm, which has demonstrated balanced performance in terms of speed and accuracy. Our results show that the EfficientDet can accurately count passengers in varying conditions with an accuracy of 94%.

Keywords: Neural Networks and Deep Learning, Machine Learning
Abstract: This study explores the application of Attentive Cross-Domain Few-Shot Learning (ACDFSL) in Hyperspectral Image (HSI) Classification, specifically addressing challenges associated with environments possessing limited labeled data. Our approach applies the Squeeze-and-Excitation (SE) attention and Residual elements within a deep learning architecture of four convolution blocks. This innovative strategy of integrating attention mechanisms into few-shot learning models represents a significant departure from traditional practices. After rigorous assessment, the ACDFSL model showcased outstanding results, revealing performance rates of 92.14%, 96.23%, and 91.27% in OA, AA, and Kappa, respectively, on the Salinas dataset. Additionally, the model attained rates of 85.67%, 89.66%, and 85.4% on the University of Pavia (PU) dataset. These results indicate an edge over existing state-of-the-art techniques such as SVM, 3D-CNN, SSRN, and other DFSL variants. This considerable progress emphasizes the potential and applicability of the ACDFSL approach in real-world HSI Classification scenarios, especially where labeled data is sparse, and paves the way for future research in this sphere.

Keywords: Neural Networks and Deep Learning
Abstract: Image denoising holds significant importance in the realm of image processing due to the potential distortions caused by environmental factors and technical problems. Consequently, it is logical to consider image denoising as a critical research domain as it aids in addressing various other image processing challenges. Although numerous techniques for image denoising have emerged in recent years, a majority of them primarily focus on restoring images afflicted by a single source of noise. In this study, the effectiveness of UNET and its variant in denoising facial images with mixed noises is examined. Furthermore, traditional filtering techniques are investigated for the purpose of comparison. The experimental results indicate the insufficiency of conventional filtering techniques in effectively mitigating mixed noise in facial images. Conversely, employing UNET-based architectures yields promising outcomes, characterized by facial images exhibiting commendable values of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Furthermore, the denoised images produced by employing the proposed residual attention UNET exhibit notable enhancements in terms of clarity and intricate details.

Keywords: Neural Networks and Deep Learning
Abstract: Quality video services have already gained high technical and commercial importance. The published work so far in this domain proposed mathematically and computationally complex algorithms, followed by the recent training-greedy deep learning-based denoising algorithms. This work proposes a video-denoising algorithm based on multiple UNet networks. The proposed video-denoising algorithm uses multiple encoder-decoder networks for video noise residual frame estimation, unlike the single encoder-decoder used by the published denoising algorithms. Using multiple skip connection UNets, we increase the residual noise modelling accuracy while restricting the signal features, which helps to improve denoising performance. The proposed network is trained end-to-end without motion compensation to reduce its complexity. The proposed network outperforms all the video denoising algorithms in terms of SSIM metric, while it yields comparable performance in terms of PSNR.

Keywords: Neural Networks and Deep Learning
Abstract: Deep neural networks (DNNs) are widely used algorithms in machine learning. Even though most of the deep learning applications are driven by software solutions, there has been significant research and development aimed at optimizing these algorithms over the years. However, when considering hardware implementation applications, it becomes essential to optimize the design not only in software but also in hardware. In this paper, we present a straightforward yet effective Convolutional Neural Network architecture that is meticulously optimized both in hardware and software for character recognition applications. The implemented accelerator was realized on a Xilinx Zynq XC7Z020CLG484 FPGA using a high-level synthesis tool. To enhance performance, the accelerator employs an optimized fixed-point data type and applies loop parallelization techniques combining 2D convolution and 2D max pooling operations. The hardware efficiency of the proposed DNN is compared with some of the existing architectures in terms of hardware utilization.

Keywords: Neural Networks and Deep Learning
Abstract: Wearable smart devices, such as smartphones and smartwatches, offer great potential as platforms for automated human action identification. However, accurately monitoring complex human actions on these devices poses a challenge due to the presence of similarities in patterns across different actions. This occurs when distinct human actions exhibit comparable signal patterns or characteristics. The placement of motion sensors on the body plays a crucial role in detecting human behavior. Typically, wearable sensors placed at the trouser pocket or a similar location are used for this purpose. However, this positioning is not suitable for identifying actions involving manual gestures, such as smoking, eating, drinking coffee, or giving a speech. To address this, wrist-worn motion sensors are employed to detect these specific behaviors. This study aims to investigate the effectiveness of deep learning models in accurately categorizing complex human actions using sensor data from wrist-worn devices. Nine deep learning models utilizing convolutional neural networks and recurrent neural networks were examined for their identification capabilities. The models were evaluated using the WHARF dataset, a publicly available benchmark dataset for human activity recognition. The investigation revealed that the CNN-BiGRU model outperformed other deep learning models, achieving an accuracy rate of 87.20% and an F1-score of 84.46%.

Keywords: Ad-hoc, Mesh and Sensor Networks, Sustainable Communications
Abstract: Advancement in technology introduced us to various platforms that changed our way of communicating and collaborating. With the sudden demand for technological dependence, the COVID-19 situation exposed the limited internet infrastructure and costly internet service of countries like the Philippines. In this study, the researchers explored the option of real-time offline collaboration in an ad hoc network as a solution to unimpeded collaboration in low bandwidth and poor connectivity environments. The researchers developed on top of Conclave, an open-source collaborative text editor that utilizes WebRTC. Offline functionality was enabled in this collaborative platform along with the implementation of offline peer discovery using Websockets and a peer handling technique inspired by Scalable Content Addressable Networks (CAN). These modifications were evaluated on an ad hoc network using different metrics such as packet delay, jitter, data rate, and peer disconnection resolution time. Results of the testing were favorable to the modified version of Conclave in an ad hoc network. With these results, the project has successfully enabled offline use and continued operation in the event of a disconnected peer, and has successfully implemented a peer handling technique that has improved the P2P operation of the native application.

Keywords: Sustainable Communications
Abstract: In this paper, we exploit caches on intermediate nodes for QoE enhancement of multi-view video and audio transmission over ICN/CCN by controlling the content request start timing of consumers. We assume the selected single viewpoint transmission method; a consumer receives video and audio streams of a requested viewpoint. We perform a simple experiment with two consumers. When the consumers play video and audio with the time difference, we assess the effect of cached content by the former consumer's request on the output quality of the latter consumer. We deal with two types of viewpoint change strategies for the former consumer, which affect the efficiency of cache utilization. From the assessment results, we see that cache utilization has an important factor in enhancing QoE.

Keywords: Ad-hoc, Mesh and Sensor Networks, Wireless Communications and Networks
Abstract: The Low-Power and Lossy Network (LLN) is the most important building block in the Internet of Things (IoT), comprising numerous tiny sensor nodes connected together. The Routing Protocol for Low-Power and Lossy Networks (RPL) is an IPv6-based protocol developed by the Internet Engineering Task Force (IETF) to facilitate routing for LLN devices. The Destination Advertisement Objects (DAOs) are transmitted from RPL nodes in the network toward the root node to construct downward routes. The malicious node exploits the DAO transmission mechanism to replay the DAO with a fixed time interval in the network in order to launch the DAO Insider attack. The DAO Insider attack causes a large number of DAO, which contributes to network congestion; as a result, data packets are delayed, and network performance is degraded. This paper proposes a defense solution that monitors DAO timestamps between child and parent nodes, flagging suspicious nodes that exceed a threshold within a time interval, blacklisting, and discarding DAOs from identified malicious nodes. Moreover, it limits the number of DAO transmitted by a child node within a specified time interval to mitigate the impact of an attack. The experiments show that the DAO insider attack has a negative impact on network performance (packet delivery ratio, average end-to-end delay, and throughput) at various DAO replay intervals. The proposed defense solution restores optimal network performance with a high detection rate.

Keywords: Wireless Communications and Networks, Ad-hoc, Mesh and Sensor Networks
Abstract: In n-party Computation, data from different sources are integrated to achieve a common goal, which can be beneficial for different applications. However, this goal may be at risk due to the fact that this corresponding data can be the target of attacks from outside attackers as well as inside ones. On the other hand, in Secure n-party Computation, the data can be processed by preserving security and privacy in such a way that no party can know the data of the others. Several schemes have been proposed to address the issues regarding n-party computation. However, some of them are not efficient in terms of computation and communication, or some of them have used a Trusted Initializer (TI), a third party, which is considered as trusted, for communicating between different parties. Therefore, to address these issues, we propose an n-party secure computation protocol without using any Trusted Initializer (TI). We formally simulate our protocol using Automated Verification of Internet Security Protocols and Applications (AVISPA) as well as we elaborate on the formal and informal security analysis of our proposed protocol. Performance analysis of our proposed protocol is also carried out and it is observed our proposed protocol is secure and protects the privacy of the system.

Keywords: Wireless Communications and Networks, Vehicular Networks
Abstract: The Unmanned Aerial Vehicles' (UAVs) capability to respond to people's needs accounts for their pervasiveness. UAVs with extended functions and capabilities when supplied with communication equipment can be deployed to appropriate places in the field to supplement the networks operate more efficiently and in vital missions such as infrastructure monitoring operations. To be effective, an UAV must interact securely with its network's entities, such as ground control stations, other UAVs, air traffic control systems, and navigation satellite systems. UAVs are exposed to a dangerous and costly world of cyber dangers as a result of cyber technology and connections. The UAV and the Ground Control Station (GCS) exchange information using communication lines, which are vulnerable to cyber attacks. The Micro Air Vehicle Link (MAVLink) protocol is a widely used lightweight communication protocol for enabling communication between UAVs and GCSs. It carries information about the UAV's condition as well as commands for control from the GCS. Although widely used, the MAVLink protocol lacks sufficient security measures and is susceptible to various types of attacks. In the current study, a new stream cipher method with low duty cycles is proposed for protecting data in UAVs and is compared with existing security algorithms on basis of various factors.

Keywords: Biomedical Imaging and Video analytics
Abstract: The examination of peripheral blood smear images for acute lymphoblastic cells represents a diagnostic approach for leukemia. The utilization of semantic segmentation of acute lymphoblastic cells can be employed in the development of a computer-aided analysis system. In the realm of peripheral blood smear analysis, deep learning methods, particularly convolutional neural networks, are commonly utilized. Currently, transformer-based models have emerged as the state-of-the-art approach for semantic segmentation tasks. In this study, SegFormer, a transformer-based model for semantic segmentation, was utilized to segment and classify acute lymphoblastic cells using four distinct training strategies. The optimal outcome was achieved with a mean intersection-over-union (IoU) of 0.821 and a mean accuracy of 0.917.

Keywords: Biomedical Imaging and Video analytics
Abstract: Being a safer modality, ultrasound is used in many imaging applications for diagnostic purposes. While Conventional Focused Beam (CFB) is an established technique and available commercially, newer synthetic aperture (SA) techniques have shown promise to overcome several limitations of CFB. However, a significant limitation of these SA techniques is that the imaging depth is much lesser than CFB. Although some works employing SA with increased transmit voltage have been shown to overcome the imaging depth challenge, not much work is reported on their thermal and mechanical indices (TI and MI), which is of prime importance to estimate the safety of ultrasound exposure. This work attempts to analyze and compare the image quality and imaging depth improvements for the different SA schemes for increased input acoustic energy while keeping the MI and TI below that of CFB. Specifically, Synthetic Transmit Aperture (STA) and Diverging Beam with Synthetic Aperture Technique (DB-SAT) are compared with CFB. Contrast ratio (CR) and Contrast to Noise Ratio (CNR) parameters are taken for image quality analysis. The experiments using needle hydrophone were done. The result suggests that an additional increase in input voltage to STA and DB-SAT schemes yielded an improvement in contrast ratio and imaging depth, without crossing the safety threshold.

Keywords: Biomedical Imaging and Video analytics
Abstract: Quasi-static elastography (QSE) is a well-known technique, where the ultrasound data acquired before and after a small tissue compression is analyzed to form an image, which is related to the stiffness parameter of the underlying tissue. Most studies have reported the effect of Conventional Focused Beamforming (CFB) transmit scheme on image quality of ultrasound elastography. Recently, Diverging Beam Synthetic Aperture technique (DBSAT) has been shown to improve the image quality of ultrasound images, however, very little is explored of this scheme for elastography application. This research paper presents a novel strategy to obtain better-quality elastography images using the DBSAT transmit scheme. The widely used CFB technique is a line-by-line scanning technique that produces one elastogram by processing one pre-compression and one post-compression frame, however, DBSAT being a synthetic aperture approach produces several low-resolution images before summing to obtain one high-resolution image. In this

experimental work on a tissue-mimicking phantom, we demonstrate that estimating 128 low-resolution elastograms and summing them to produce one high-resolution elastogram is an effective way to reduce noise and increase the elastographic image quality. More than 15dB improvements were found when comparing the image quality metrics of signal-to-noise ratio (SNR) and Contrast-to-noise ratio (CNR) from elastography images obtained from DBSAT vs. CFB.

Keywords: Translational Engineering in Healthcare, Biomedical Imaging and Video analytics
Abstract: Breast cancer is the most prevalent cancer among females, often leading to complications like lymphedema. Current detection methods for lymphedema following breast cancer surgery are imprecise and time-consuming. We present a cost-effective and user-friendly contactless arm volumeter for accurate lymphedema detection. Our volumeter utilizes six depth cameras to capture a 3D arm representation. By combining the recorded depth points from multiple angles, we measure arm circumference and volume accurately. In a study with 25 volunteers, we compared volumeter measurements with standard methods using a tape measure and water displacement. The intraclass correlation coefficient (ICC) assessed agreement and reliability. Results showed strong consistency agreement (ICC: 0.981 for volume, 0.868 for upper arm circumference, and 0.933 for lower arm circumference) and absolute agreement (ICC: 0.975 for volume, 0.866 for upper arm circumference, and 0.895 for lower arm circumference). Reliability was high (ICC: 0.993 for right arm volume, 0.975 for left arm volume, 0.988 for right arm circumference, 0.975 for left arm circumference (upper arm), and 0.948 for right arm circumference, 0.933 for left arm circumference (lower arm)). Our contactless arm volumeter is a reliable and cost-effective solution for detecting lymphedema after breast cancer surgery. Its ease of use and accuracy enable timely detection and treatment, potentially improving patient outcomes.

Keywords: RF/Millimeter-wave Circuits and Systems, THz, mmWave and RF Systems for Communications
Abstract: The focus of this work is the development of a sub-6 GHz (2-6 GHz) low noise amplifier (LNA) for 5G applications, using a 65 nm CMOS process. A novel two stage common source (CS) cascode source degeneration LNA topology by incorporating a contemporary series parallel LC network and two stage LC network for input and output matching respectively is proposed. The circuit implementation, simulations and evaluation of the LNA’s performance are done utilizing the RF Spectre Cadence Virtuoso. According to the evaluation results, the LNA dissipates a total power of 19.6 mW at the supply voltage of 0.7 V. It offers an operational wide bandwidth (BW) of 3.2 GHz which ranges from 2.8 GHz to 6 GHz. The LNA has a peak gain of 36 dB and minimum noise figure (NF) of 1.1 dB across the sub-6 GHz spectrum. The proposed LNA also performs well in terms of stability and linearity measures. The layout of the proposed LNA occupies an area of 0.182mm2.

Keywords: Advanced CMOS devices and process, Analog and mixed signal ICs, Beyond CMOS device technology
Abstract: This paper introduces a Schmitt trigger with adjustable hysteresis, specifically developed for low power performance and fabricated using 22nm FDSOI technology. The proposed design incorporates a dynamic threshold voltage generation technique to achieve both reduced power consumption and high noise immunity. The voltage-controlled hysteresis is adjustable to enable the design to be customized for different applications. The Schmitt trigger is implemented using a standard CMOS process and the circuit performance is analyzed using Cadence Virtuoso simulation. The outcome presents that the suggested design achieves a low power consumption ranging from nW to pW and a high noise immunity with an adjustable hysteresis of 0 V to 0.4V to both gate voltages. The suggested design is well-suited for low power requirements in various digital circuit applications, including memory, microprocessors, and sensors.

Keywords: Advanced CMOS devices and process, Analog and mixed signal ICs, Beyond CMOS device technology
Abstract: This paper presents a novel approach to voltage reference design, harnessing the bulk biasing technique in 22nm Fully Depleted Silicon-on-Insulator (FDSOI) technology. The proposed architecture exhibits ultra-low power consumption of 95.28 nW while having a dissipating supply current of 157 nA and a high Power Supply Rejection Ratio (PSRR) of more than -100 dB, realized through an all-MOSFET construction where TC compensation of PTAT and CTAT voltage generators is adopted. A temperature coefficient of 22.22ppm/◦C is achieved over wide temperature range from -45◦C to 100◦C with an output voltage VREF of 351 mV. With the growing demand for efficient, low-power, and compact solutions in Internet of Things (IoT) power management, this development contributes a significant step forward in the domain.

Keywords: Advanced CMOS devices and process, Emerging memory technologies
Abstract: Internet of Things (IoT) devices have only limited computing resources, which means we need to reduce the scale of operation circuits and energy consumption to build a neural network (NN). The binarized neural network (BNN) and computing-in-memory (CiM) have been proposed to fulfill these requirements, and recently, magnetic random access memory (MRAM), next-generation memory for CiM-based architectures has attracted interest. In this study, we utilize a CiM architecture based on an MRAM array to build a BNN on the edge side. We also implement an XNOR gate on our MRAM array using voltage-controlled magnetic anisotropy (VCMA)-based magnetization switching to reduce the scale of the multiply-and-accumulate (MAC) operation circuits by half. Further, we propose a BNN training algorithm utilizing ternary gradients to enable both training and inference on the edge side using only binary weights and ternary gradients. Experiments on the MNIST dataset showed that our MRAM array can achieve an accuracy of around 80%.

Keywords: Machine Learning, Neural Networks and Deep Learning, Meta heuristic algorithms
Abstract: In the modern world, where health concerns necessitate continual diet monitoring, the challenge of food image identification is crucial. Many machine learning models are available to automate the identification procedure. This is done predominantly with Convolutional Neural Networks (CNN) that help extract features for food images with different textures. But this comes with certain limitations such as diversity in food items, variation in the appearance of images, overfitting, and the inability to capture long-distance connections, which can result in inadequate feature representations. This paper attempts to explore Vision Transformers (ViTs) in an effort to overcome these limitations. ViTs are known for their attention mechanism, increased interpretability, better generalization, and robustness to adversarial cases. In this study, VOLO (Vision Outlooker for Visual Recognition), a contemporary vision transformer, improves learning by encoding fine-level information into the token representations. Also, a traditional flat classifier ceases to perform well because there are so many different cuisines and unique food items. Prediction systems with hierarchical classifiers were also developed to address this. Thus, the proposed method uses VOLO to accomplish hierarchical food classification. The experimental results support the proposed method's performance and contribution to an overall improvement in prediction accuracy.

Keywords: Analog and mixed signal ICs, Advanced CMOS devices and process, MEMS and semiconductor sensors
Abstract: Voltage references are crucial for various mixed-signal and RF systems in the analog market, including IoT, wireless technologies, and body-area networks. In recent technological advancements, the increasing demand for low-power, low-cost, and energy-efficient solutions has driven the need for sub-1V voltage reference circuits. This paper presents the design and development of a sub 1-V voltage reference circuit for ultra-low power applications. The MOSFET-only circuit proposed here is compatible with CMOS fabrication in the 180 nm node. The voltage reference circuit achieves a stable 0.406V reference with a low line sensitivity (LS) of only 0.2%/V over a wide source voltage range from 648mV to 3.6V. It exhibits a temperature coefficient (TC) of 32 ppm/◦C within the temperature range of -40◦C to 105◦C with a power consumption of 9.9nW at room temperature. The performance analysis includes a comprehensive assessment considering all process corners and statistical analysis through Monte Carlo simulations. Furthermore, a comparison with state-of-the-art references validates the effectiveness of the proposed circuit.

Keywords: Engaging Undergraduate Students in Research, Multidisciplinary and Transdisciplinary Education, Non-traditional Lab concepts
Abstract: This paper explores the intersection of education, synesthetic interaction design, and the aesthetics of engineering and technology. The first section delves into the concept of synesthetic interaction design. It examines the merging of sensory modalities to create meaningful and holistic user experiences. By integrating multiple senses, synesthetic interaction design aims to engage users on a deeper emotional and cognitive level, enhancing their overall experience and satisfaction. The second section explores the aesthetic elements in emerging technology and the importance of aesthetics in engineering, examining how sensory integration contribute to the overall experience. The final section of the essay focuses on the implications for education in synesthetic design with emerging engineering and technology. It discusses the need for interdisciplinary approaches in educational curricula, highlighting the significance of integrating design principles, engineering fundamentals, and aesthetics. The author argues that a comprehensive education in these fields should encompass not only technical skills but also a deep understanding of sensory perception, and emotional engagement. By nurturing a holistic and multidisciplinary mindset, future professionals can bridge the gap between engineering and aesthetics, paving the way for innovative.

Keywords: Student-centered Learning Environments, Multidisciplinary and Transdisciplinary Education, Engaging Undergraduate Students in Research
Abstract: This work describes the design of an iterative learning approach to teach basic concepts in autonomous and control systems to novice engineering students. The Learning Continuum approach builds on three theoretical principles: Zone of Proximal Development, Selective Exposure, and Concrete before Abstract. The developed curriculum unit splits the lesson into two iterations. The first focuses more on the high-level concepts using tools that avoid low-level complexities of typical microcontrollers. The second iteration then uses a more traditional microcontroller setup to explore the learned ideas. We tested this framework with 217 novice engineering students against a control group consisting of 111 students. Results from a range of data collected show that this approach helps students during the learning process. Instructors in the study have adopted our learning continuum for regular use in their course.

Keywords: Student-centered Learning Environments, Multidisciplinary and Transdisciplinary Education, Virtual and Remote Labs and Classrooms
Abstract: Autism Spectrum Disorder (ASD) affects how people communicate, learn, behave, and socially interact. Early intervention and effective educational practices can greatly improve the condition. Finding solutions for autism should be done from the early stages of diagnosis since helping a person’s development in the child stage has a positive impact. There are limited opportunities for autistic students in the Information Technology space, so it is necessary to provide applications or technical assistance to help them. Developing an application for autistic children requires more experimental attention. In this research we are suggesting a solution to enhance the soft skills and learning skills of autistic kids based on the identification of emotions and behaviors over a constant average period of time.

Keywords: Multidisciplinary and Transdisciplinary Education
Abstract: Environmental issues and sustainable development are becoming increasingly vital topics in today's global context, necessitating a thorough understanding of public perceptions and attitudes towards these critical concerns. This conference paper presents a comprehensive survey report and analysis that investigates students' perceptions of environmental issues and sustainable development. The study aimed to assess the awareness, knowledge, and attitudes of students, who are pivotal stakeholders in fostering sustainable practices and shaping the future. The survey for data collection comprised a range of questions covering multiple dimensions, including environmental awareness, knowledge of sustainable development principles and personal behaviors. The results of the survey revealed valuable insights into student views on environmental and sustainability concerns. The findings indicated a moderate to high level of awareness among the surveyed students, with varying degrees of knowledge and attitudes toward sustainable practices. Moreover, the analysis shed light on the factors influencing the perspectives of students, including educational background, environmental education and awareness provided in the curriculum, personal experiences, locus of control and environmental responsibility and exposure to environmental initiatives. This study aims to contribute to the broader goal of creating environmentally conscious and responsible citizens.

Keywords: Others
Abstract: This study examines the funding strategies of Chulalongkorn University's Learning Innovation Center (LIC) to support every type of education research in every level, initiated in 2011. We analyzed 267 research projects from 21 faculties and institutes, categorized based on the funding they received. The objective was to learn from past funding experiences to improve future decisions. Our analysis revealed that while many projects were successful within their classrooms, their impact was limited beyond those settings. These findings underline the urgent need to reevaluate and adjust our funding policies, calling for the incorporation of state-of-the-art knowledge in educational practices. By pinpointing past shortcomings, we aim to shape an improved funding strategy that broadens the impact of research projects, ensures equitable support across all faculties and institutes, and better aligns with the university's lifelong learning goals. Ultimately, this research seeks to augment the efficacy of LIC's initiatives, fostering a sustainable and innovative learning environment that aligns with national development strategies.

Keywords: Big Data Analytics, Data Centric Programming, Data Modeling & Semantic Engineering
Abstract: Simulating multivariate random variables is essential in data analytics as it allows for more accurate modeling, improved decision-making, and a better understanding of complex systems and processes. The NORTA algorithm, a widely used method, can accomplish this task. However, it requires an initial correlation matrix to produce multivariate random variables with the desired correlation matrix, and both matrices usually differ. This paper presents an efficient simulation-based algorithm for determining the initial correlation matrix, leveraging quasi-Monte Carlo integration with the Halton sequence, an adaptive sum of squared errors, and some probability distributions. The proposed algorithm is tested to generate many cases of multivariate random variables with different distributions and different correlation matrices. The results show that this algorithm can substantially reduce time compared to the traditional simulation-based method.

Keywords: Knowledge Engineering
Abstract: Natural language processing (NLP) is mandatory in working with text. There are many tools and applications that are based on it. However, most of those tools are often operated in, or only, the English language. In recent years, there has been a continuing development for NLP tools to support other languages or creating a specific tool for certain language for simple NLP tasks, but for some of the advanced tasks, the advancement is still behind that of the English language, Thai language is also one of them. So, in this research, the capabilities of the currently existing Thai NLP tools are explored and evaluated, with the tasks of extracting text from the Thai government dataset (eMENSCR) and creating the knowledge graph from it to improve data interpretability and gain more insight from the data, by utilizing queries that are exclusive, or less complex to execute, when the data is stored in the graph database such as performing a path traversal or relationship counting on the data. Natural language processing’s part of speech tagging and named entity tagging is used to perform entity and relation extraction after filtering the unneeded data fields. Then the extracted information will be formulated into the format of “triple”, which is in the form of (head, relation, tail). After the process of triple construction is finished, The triples are evaluated by Precision, recall, and F1 in order to measure the pipeline’s performance and import to the Neo4j for query testing

Keywords: Knowledge Engineering, Data Modeling & Semantic Engineering, Domain Specific Data Management
Abstract: Resource Description Framework (RDF) is widely used in semantic extraction, unified organization, and intelligent processing of large amounts of data because of its machine intelligibility. For example, knowledge graph based on RDF is commonly used in intelligent search, recommendation system, and smart medical treatment. And RDF is used to express the relationship between entities and process the semantics of data. Many efforts have been made to convert various data (such as relational database, XML, and JSON) into RDF. Yet, the effective generation of usable RDF data is still an urgent problem to be solved. With the wide use of NoSQL database, massive data is stored in NoSQL database, but the research on generating RDF from NoSQL database is not emphasized. We put forward a formal definition of MongoDB, and according to this definition, we propose a method of automatically extracting data from MongoDB and building corresponding RDF. Based on this method, we have also implemented a prototype system named M2R to validate method performance. The experimental results show that our approach is feasible and efficient.

Keywords: Data, Text, Web Mining, & Visualization, Big Data Analytics
Abstract: In Japan, The declining birthrate and aging population have become a social problem, and it is predicted that the number of elderly people will reach about 40 million in 2040. As a result, the number of dementia patients is expected to　increase. There are four major dementias, and appropriate care methods are different for each. In addition, early detection is important to suppress symptoms. Therefore, there is a need for a means to easily determine the type of dementia. In this study, we propose an automatic classification method for four major dementias. Using crowdsourcing, we extract text data of four major dementias symptoms from Twitter. By fine-tuning BERT with them, We obtain a classifier. Experimental result shows that the proposed method provides higher accuracy than random classification.

Keywords: Data, Text, Web Mining, & Visualization, Business Intelligence, Machine Learning
Abstract: The development of digital technology 4.0 has significantly impacted online shopping behavior. The primary research problem revolves around the crucial aspect of maintaining consumer trust in online stores. This study aims to conduct direct observations of customers and model the findings using PLS SEM. Additionally, it involves collecting data on both negative and positive customer comments to assess customer satisfaction through comparative analysis using the Naive Bayes algorithm method. The ultimate goal is to achieve optimal results and gain valuable insights into the factors that influence customer satisfaction in the context of online purchases. This research directly benefits buyers as it helps them understand the factors that influence customer trust and satisfaction, enabling them to make informed purchasing decisions.The study's findings indicate that the gaussian naive Bayes algorithm initially achieved a low accuracy score of 0.37. However, after implementing hyperparameter tuning, the accuracy significantly improved to 0.62. Additionally, the results of the questionnaire analysis using smart pls showed a normed fit index (NFI) of 0.707, slightly below the acceptable threshold of 0.90. The standardized root mean square residual (SRMR) was 0.071, which is lower than the acceptable value of 0.08, suggesting a good fit of the model. However, the rms theta value of 0.24 exceeded the threshold of 0.102

Keywords: Business Intelligence, Crowd Sourcing & Social Intelligence, Machine Learning
Abstract: Analyzing the crucial factors which help predict startups’ funding amounts is important for senior executives of these firms to formulate effective business strategies, which leads to the ultimate startup success. In this research, we crawled real-world startup funding data from the well-known “AngelList” platform which disseminates information about the company profiles of startups, the specific business sectors, and potential investors. Potential investors browse the information posted on AngelList, which may in turn influence their decisions in funding certain startups. Our work aims to evaluate the bundle of factors (e.g., sentiments and emotions embedded in startup profile descriptions, startups’ fundamentals, etc.) that may influence startups’ funding successes. Moreover, we have examined a variety of state-of-the-art machine learning-based prediction models. In particular, we applied TextCNN, a well-known deep learning method to extract sentimental and emotional features from company profile text to enhance the startup funding prediction task. Our experimental results show that the emotion feature can significantly boost startup funding prediction performance by 12% in terms of F-score, and it is also among the key factors that influence startup funding amounts. To our best knowledge, this work represents the first successful research on examining the relationship emotions captured in company profile text and startup funding success.

Keywords: Business Intelligence, Neural Networks and Deep Learning, Machine Learning
Abstract: This report introduces an innovative stock market portfolio management approach, utilizing reinforcement learning and sentiment analysis techniques. Unlike traditional methods, which rely on time-consuming fundamental and technical analysis susceptible to human biases, our proposed approach leverages machine learning advancements such as DQN, DDQN, and Dueling DQN algorithms, combined with sentiment analysis to automate and enhance portfolio decisions. This study introduces a novel framework that combines reinforcement learning with multiple methods and sentiment analysis for stock market portfolio management. We have used Sensex NSE stock data comprising of historical data of open ,low , high , close prices from 2010 to 2023 . Experimental results demonstrate that our approach outperforms traditional portfolio management methods regarding returns and risk management.

Keywords: Neural Networks and Deep Learning, Machine Learning, Data Mining
Abstract: Real estate is a critical sector in Thailand's economy, which has led to a growing demand for a more accurate land price prediction approach. Traditional methods of land price prediction, such as the weighted quality score (WQS), are limited due to their reliance on subjective criteria and their lack of consideration for spatial variables. In this study, we utilize aerial or satellite imageries from Google Map API to enhance land price prediction models from the dataset provided by Kasikorn Business-Technology Group (KBTG). We propose a similarity-based asset valuation model that uses a Siamese-inspired Neural Network with pretrained EfficientNet architecture to assess the similarity between pairs of lands. By ensembling deep learning and tree-based models, we achieve an area under the ROC curve (AUC) of approximately 0.81, outperforming the baseline model that used only tabular data. The appraisal prices of nearby lands with similarity scores higher than a predefined threshold were used for weighted averaging to predict the reasonable price of the land in question. At 20% mean absolute percentage error (MAPE), we improve the recall from 59.26% to 69.55%, indicating a more accurate and reliable approach to predicting land prices. Our model, which is empowered by a more comprehensive view of land use and environmental factors from aerial or satellite imageries, provides a more precise, data-driven, and adaptive approach for land valuation in Thailand.

Keywords: Neural Networks and Deep Learning, Machine Learning, Human Computer Interface
Abstract: Large Language Models with autoregression generative capabilities like ChatGPT have garnered lots of attention from its launch. However, the cyber security community is also wary of the threats that it poses with cybercriminal and cyber security threat related activities. It could generate highly deceptive phishing and social engineering attacks that could evade human detection and render existing phishing or social engineering detection tools useless. Inspired by the approach used to develop Foundation Model that resulted with amazing capabilities from the contemporary model constructs like ChatGPT, our research endeavour demonstrates a model construct developed using Foundation model approach could yield potential as defensive tool to detect GPT generated text. Preliminary evaluation results show promising results.

Keywords: Neural Networks and Deep Learning, Machine Learning, Human Computer Interface
Abstract: Railway track detection is crucial in railway infrastructure maintenance, safety, and operational efficiency. This paper proposes a railway track detection method based on the SegNet deep learning architecture. The SegNet model is a convolutional neural network (CNN) designed explicitly for semantic segmentation tasks. By training the SegNet model on annotated railway track images, we enable it to accurately classify each pixel in the input images as either track or non-track. The proposed method leverages the rich feature representation capabilities of deep learning to achieve robust and precise track detection, even in complex and challenging scenarios. We evaluate the performance of our approach on a benchmark dataset, considering metrics such as accuracy, intersection over union (IoU), and mean BF score. The experimental results demonstrate that our method outperforms existing track detection methods regarding accuracy and efficiency. The proposed railway track detection based on SegNet deep learning has the potential to significantly improve railway maintenance practices and enhance overall safety and operational effectiveness.

Keywords: Mobile robotics, Control system modeling
Abstract: In recent years, location-based services (LBS) have attracted much attention because of increasing smartphone users. The technology mainly used for estimating position of smartphone users is the global navigation satellite system (GNSS). However, the satellite positioning has demerit it cannot provide accurate position information in environments such as indoors or near buildings. Therefore, we focus on the pedestrian dead reckoning (PDR), which estimates the pedestrian position using the sensors in the terminal. The positioning by PDR needs to take into account the walking characteristics of the pedestrian individually. This paper proposes a method to adaptively estimate the parameters for pedestrians to improve the accuracy of positioning by PDR.

Keywords: Mobile robotics, Intelligent control, Neuro-control, Fuzzy control and their applications
Abstract: It is a prevalent trend nowadays in most places that the elderly population keeps increasing. Also, there is a community that has physical disabilities, especially mobility. Due to the lack of trustworthy caretakers, the busy lives of family members, and psychological issues attributed to loneliness, intelligent service robotic devices have been developed. Vocal navigational commands consist of uncertain terms. There are methods introduced to enhance these uncertainties using spatial parameters such as obstacle distances, previous robot movements, robot localization, etc. However, they have only considered the vocal information provided by the user and deduced the rest of the information from external spatial information. In reality, users include other partial information from complementary modalities such as hand gestures and there is a significant probability that the partial information carried by the hand gestures might change the interpretation completely. Therefore, this paper presents an intelligent system that would enhance the understanding of distance-related uncertainties of vocal navigational commands using multimodal fusion of hand gesture information. Partial gesture information extracted from hand gestures has been used to interpret the distance-related uncertainties using a fuzzy logic-based approach. Experiments were conducted to validate the intelligent system and the user ratings given were used to validate the system.

Keywords: Networked control systems, Control system modeling, Industrial automation
Abstract: This paper investigates the leader-follower synchronization problem of delayed complex dynamical network. The proposed methodology analytically derives the synchronisation condition for complex delayed dynamical network with a power leader such that all follower systems (delayed or non-delayed) exponentially synchronize their states to power leader through local interactions. For developing the explicit conditions for synchronisation, partial contraction theory along with properties of coupling matrices are exploited. Simulation results are used to demonstrate the effectiveness of the proposed control algorithm.

Keywords: Networked control systems, Control system modeling, Intelligent control, Neuro-control, Fuzzy control and their applications
Abstract: The expansion of urbanization has resulted in a reduction of available land for agricultural purposes. As a response, aquaponics has emerged as an environmentally friendly solution for local food production. By integrating aquaculture (fish farming) and hydroponics (soilless farming), aquaponics provides an opportunity for individuals without access to land to engage in farming and aquaculture activities. Traditional aquaponic systems typically require substantial space. However, these systems can be upgraded through adequate financial resources and offer enhanced flexibility. This study presents a compact aquaponics system that operates efficiently using solar panels as a sustainable power source. Unlike the traditional aquaponic system, which does not offer any automated monitoring features, this system incorporates automatic monitoring features facilitated by Arduino microcontrollers and sensors, achieving a remarkable accuracy rate of over 90% in maintaining optimal conditions. In conclusion, aquaponics is a viable solution for urban farming in the face of limited available land. The compact aquaponics system proposed in this study, powered by solar panels and automated monitoring capabilities, exemplifies a scientifically rigorous and sustainable approach to agricultural practices.

Keywords: Biomedical Imaging and Video analytics, Biomedical and Health Informatics
Abstract: Breast cancer is a disease where breast cells grow out of control and leads to cancer. Various methodologies have been developed to identify breast cancer. In this paper, we have developed an approach to classify breast cancer from histopathology images. The approach makes use of deep learning based architectures by setting same parameters for all while training and testing on them. Thereafter, all the architectures are compared to see which one is most suited for the classification of breast cancer. Previous works on AlexNet, VGG, ResNet have already been published, and here we have tried to see the performance of those models which have less number of trainable parameters, namely DenseNet121, DenseNet169, DenseNet201, EfficientNetB0, EfficientNetB5, EfficientNetV2B0 and EfficientNetV2S. Here, all the experiments are conducted on BreakHis histopathology dataset by utilizing all the images of resolutions 40X, 100X, 200X and 400X of benign and malignant cancer.

Keywords: Biomedical Imaging and Video analytics, Biomedical Signal Processing and Instrumentation, Computational Systems, Modeling and Simulation in Medicine
Abstract: Breast density is an important bio-marker for predicting the risk of breast dancer. Studies have showed that women with dense breast have higher probability of breast cancer. The density, assessed using mammogram images, is a measure of amount of fibro-glandular tissues in a breast and the appearance of it can mask the lesions leading to lesser sensitivity in breast cancer detection. Hence the current clinical workflow for breast detection incorporates a density classification stage which suggest detailed analysis of dense breast using additional imaging modalities like Digital Breast Tomosynthesis (DBT). In this work a breast density classifier using deep neural network architecture employing transfer learning is proposed for the binary classification of breast density to assist in the clinical workflows. The results are evaluated on publicly available databases namely DDSM, MIAS and InBreast. The proposed model outperforms the existing works in the literature giving on an average 96% and 94% accuracy respectively, when tested on DDSM and MIAS database.

Keywords: Biomedical Imaging and Video analytics, Biomedical Signal Processing and Instrumentation
Abstract: Melanocytes, which produce melanin, the pigment that gives your skin its colour, are the source of carcinoma, one of the most life-threatening types of skin cancer. Although there is no confirmed cause for all tubercles, exposure to ultraviolet (UV) rays from the sun, tanning lamps or prolonged exposure to sunlight increases the risk of developing the condition. If cancer is not treated at early stages of development, there is a high probability of mortality. The probability ratios of the case surviving can be improved with a timely and precise opinion. Initial identification of skin cancer can save the lives of those affected. Because of this, it is vital to develop a computer-based support system for the detection of carcinoma. In order to determine whether the specimen skin lesions are benign or malignant, this paper discusses various novel deep transfer learning methods for early melanoma diagnosis. Deep convolutional neural networks are used to determine if these particular skin lesions are malignant or benign. The use of various datasets to evaluate the viability of the deep learning architecture is discussed as well. According to the findings of the experiments, the deep learning strategy performs better than many of the traditional deep learning algorithms in terms of computational efficiency and precision.

Keywords: Biomedical Imaging and Video analytics
Abstract: The discernment of the gender of the fetus during obstetric ultrasound has played a major in sex-selective abortions. Researchers have shown that in India, there are approximately 50,000 to 100,000 abortions per year. Though there are laws on female feticide, poor awareness and difficulty in implementing technological solutions has made it difficult to effectively dissuade such practice. In the current scenario, B-scans are displayed on monitor screens in real time, which can lead to the unnecessary display of frames that reveal the gender of the fetus. This workflow adds undue burden on the operator to take further precautions to prevent leak of the gender information. Our work not only focuses on automatically detecting ultrasound frames that contain gender but also identify the specific region of the frame that indicated the gender, so that it can be obscured during real-time display.

Keywords: Analog and mixed signal ICs, Device modeling & characterization
Abstract: The ternary logic approach was proposed in an asynchronous digital system to eliminate overhead wires for communication signals in preference to the binary logic approach. Spacer detectors (SDs) are crucial in ternary logic to determine whether or not the input voltage is equivalent to empty. The Internet of Things (IoT) devices leverage asynchronous multi-value logic systems for machine learning inference to minimize power consumption. These IoT devices are also expected to operate at ultra-low operating voltages (sub- or near-threshold voltages). An ultra-low supply voltage restricts the operation of the SD circuit because of the need to determine the space value with a logical intermediate value. This study presented SD circuits was applied a pseudo-differential amplifier instead of the traditional element-based inverter circuit to design H-element and L-element. Hence, our proposed SD circuits can operate at low supply voltage without body bias using a bulk-controlled method. The proposed circuit was simulated in 65 nm UMC LL process technology in the Cadence Virtuoso analogue design environment with power consumption compared with the difference in supply voltage from 0.35 to 0.9 volts.

Keywords: Analog and mixed signal ICs, Electronic devices, materials and fabrication process
Abstract: Agricultural fertilization plays an important part in maintaining and increasing the yield capacity and growth of plants.  Fertilizer application must be accompanied by the correct fertilizer recommendation.  This study aims to develop a system that can measure the NPK contents of soil and provide fertilizer recommendations for Bok Choy and determine if there will be a significant difference in the plant’s average leaf length and weight when compared to a method that uses a complete fertilizer application. Two different setups were done in the study to grow the plant. Setup A used fertilizer recommendations based on the system while Setup B used only complete fertilizer. Data shows that in terms of leaf length and weight, Setup A produces on average, longer leaves and heavier yields, respectively. However, based on the t-test analysis, both the p-value for the average length of leaves and weight were greater than the significance level, indicating that there was no significant difference in the results when using Setup A compared to Setup B. In conclusion, the use of an NPK recommendation and fertilizer recommendation system does not significantly affect the average length of leaves and weight of Bok Choy plant when compared with the use of a complete fertilizer setup.

Keywords: Analog and mixed signal ICs, MEMS and semiconductor sensors, Device modeling & characterization
Abstract: Design of a low-power CMOS image sensor (CIS) with a duplicated comparator for pixel-signal prediction is discussed in this paper. The proposed design features a pixel-signal-prediction-based scheme comprising a main comparator and a duplicated comparator. The comparator generates a prediction signal based on the difference of auto-zeroed voltage derived from static current differences of two comparators, finally resulting in the reduction of the number of counter toggles. In addition, for further reduction of power consumption, the second-stage amplifier in the main comparator utilizes the proposed positive-feedback bias-sampling technique to cut off the current path after the comparison. The proposed CIS with a 11-bit single-slope ADC(SS –ADC) is implemented using a 110-nm CMOS process, has a resolution of 640 × 480, and operates at a frame rate of 299 frames per second. The experimental results demonstrate that the reduction of the power consumption of SS-ADC with the proposed comparator is about 65%. In addition, we obtained the total power consumption of the proposed CIS per column of 13.1 μW and a figure of merit of 44.6 fJ/conv.-step.

Keywords: Analog and mixed signal ICs
Abstract: This article presents grounded RC series/parallel impedance simulator circuits employing a single voltage differencing buffered amplifier (VDBA) as an active element. The proposed RC series impedance simulator consists of one capacitor and one resistor, while the proposed grounded RC parallel type requires only a single capacitor as a passive element. The simulated equivalent resistance and capacitance values can be tuned electronically through the transconductance gain of VDBA device which can be adjusted by means of the external biasing current. The performance validation of the proposed circuits including the resonant circuit application is proved by simulation results via PSPICE software based on TSMC 0.25-µm CMOS technology.

Keywords: Beyond CMOS device technology
Abstract: The goal of this work is to exclusively investigate the effective current conduction path in the channel of a Triple Gate (TG) Silicon-ON-Insulator (SOI) Junctionless Fin Field Effect Transistor (JLFinFET). It is observed that various structural parameters play a key role in deciding the location of the effective current conduction path both in full depletion mode and partial depletion mode in TG SOI JLFinFET. Considering the present day technology requirements 20 nm was chosen as the gate length. Simulations performed using 3-D TCAD namely ATLAS by Silvaco Inc. reveal that the conducting path from source to drain starts from nearer to the centre of the channel (i.e, at half the fin height and half the fin width) when the transistor switches from the OFF state to the ON state. It is also observed that when the triple gate transistor scales down in size the capacitive coupling between the top gate and side gates is a crucial factor in determining the location of the effective current conduction path.