Computing and Computational Intelligence (CCI) 1
Track C7F1 CCI 7.1: Computing & Computational Intelligence (CCI) 7.1
Room: F1. Sipadan I (Level 4)
Chair: Sanjaya Kumar Panda (National Institute of Technology Warangal & NITW Techsammelan Private Limited, India)
11:30 Adaptive Feature Aggregation Enhanced by Using DenseNet for Robust Breast Cancer Histopathology Image Classification
Zaka Ur Rehman (Malaysia); Mohammad Faizal Ahmad Fauzi, Wan Noorshahida Mohd-Isa and Meriem Touhami (Multimedia University, Malaysia); Arbab Sufyan Wadood (Multimedia University, Malaysia & BUITEMS, Pakistan); Muhammad Kashif Jabbar (Shenzen University, China)
Accurate classification of breast cancer from histopathology images is critical for early diagnosis and effective treatment planning. While deep learning techniques-particularly convolutional neural networks (CNNs)-have achieved substantial success in medical image analysis, existing models often struggle with issues such as overfitting, channel redundancy, and insufficient focus on clinically salient features. To address these limitations, this paper introduces a deep learning framework that integrates an Adaptive Feature Aggregation (AFA) block into the DenseNet121 architecture. The proposed AFA module learns to emphasize important feature channels by modeling inter-channel dependencies through a lightweight attention mechanism, thereby improving the network's ability to distinguish between benign and malignant tissue patterns. Extensive experiments were conducted on the BreakHis 400x histopathology dataset. The proposed model achieved 98% accuracy, 98% F1-score, and an AUC of 0.99, outperforming the baseline DenseNet model. Evaluation metrics such as precision, recall, ROC-AUC, and confusion matrix analysis confirm the robustness and effectiveness of the proposed method for breast cancer classification.
11:45 Enhancing Mental Health Disorder Classification: a Decision Tree-Based Approach with Optimized Feature Engineering, Data Balancing, and Hyperparameter Tuning
Md. Arifur Rahman Akib, Jannatul Ferdous and Fariah Mahzabeen (North South University, Bangladesh); Riasat Khan (North South University, Bangladesh & New Mexico State University, USA)
Recent advancements in natural language processing (NLP) have employed transformer-based models to classify mental health conditions using social media data. Nevertheless, these models often face challenges with complex feature dependencies, require substantial computational power, and may not always yield optimal results. In contrast, conventional machine learning models can attain similar or even better performance by utilizing effective feature engineering, balancing data, and optimizing hyperparameters. This study addresses an important gap by demonstrating that a properly optimized Decision Tree model can greatly improve classification accuracy, surpassing deep learning methods such as BERT and BiLSTM. By utilizing Term Frequency-Inverse Document Frequency (TF-IDF), extracting N-gram features, and applying the Synthetic Minority Over-sampling Technique (SMOTE) for balancing classes, we improve the accuracy of predictions. Our Decision Tree model shows a 16% increase in accuracy compared to transformer-based models, demonstrating that a well-tuned conventional machine learning method can compete with or exceed deep learning methods which have more computational cost.
12:00 M-MedNeRF: 3D Modeling and Novel View Synthesis from Single-View X-Rays Using Mamba-Accelerated Neural Radiance Fields
Lohith Saradhi Kandukuri and Jiji Victor Charangatt (Shiv Nadar University Chennai, India)
Reconstructing 3D medical images from single-view X-rays offers a low-risk, low-cost alternative to traditional 3D reconstruction methods that use multiple CT or MRI scans. But current methods often face limitations in computational efficiency and visual fidelity. In this paper, we extend the state-of-the-art GAN-based novel view generation model, MedNeRF, by proposing M-MedNeRF, which integrates the Mamba architecture-a novel state-space model optimized for long-range sequence modeling with linear time complexity -into the volumetric rendering pipeline. The generator of the model, samples points in 3D space to train a Deep Neural Network to predict view for a given Camera position, which is passed on to a discriminator to evaluate the generator's performance. The proposed M-MedNeRF model captures spatial dependencies more effectively, modeling ray sequences as a whole rather than independently. We demonstrate that this architecture outperforms the baseline in SSIM and LPIPS metrics, with notable qualitative improvements in anatomical reconstruction. These results highlight the potential of fast and efficient sequence modeling in advancing single-view 3D medical image reconstruction.
12:15 Visualization of the Contributions of Frequency-Domain Features to Person Identification in Motor Imagery EEG
Yuki Arai, Tadanori Fukami and Chako Takahashi (Yamagata University, Japan)
When classifying a person's state or emotions from brain activity, individual differences in electroencephalography (EEG) signals pose a key issue. If those features of motor imagery EEG that contribute to improved accuracy in person identification can be identified, they may provide insights into the nature of individual variability in EEG data. In this study, we used the frequency-domain feature attribution method proposed by Tachikawa et al. to calculate and visualize the contribution of EEG frequency bands to person identification performance. We trained a person identification classifier on motor imagery EEG datasets, which achieved approximately 80% accuracy. For this classifier, we calculated the contributions of frequency-domain features. We observed that both the amplitude and phase exhibited highly contributive frequency bands in the low-frequency range. Furthermore, in the case of amplitude, the high-contribution bands were distributed over a broader frequency range, suggesting that individual differences may be reflected differently in EEG amplitude and phase information.
12:30 Explainable AI for Breast Cancer Diagnosis Using EfficientNetB3 with Attention Mechanism
Md Serajun Nabi, Mohammad Faizal Ahmad Fauzi and Hezerul Abdul Karim (Multimedia University, Malaysia); Tong Boon Tang (Universiti Teknologi PETRONAS, Malaysia); Normy Abdul Razak (Universiti Tenaga Nasional, Malaysia); Hasanul Bannah (Multimedia University, Malaysia)
Accurate classification of HER2 immunohistochemistry (IHC) scores is essential for determining effective breast cancer treatment, yet it remains challenging due to subjective manual interpretation, especially for borderline scores (1+ and 2+). This study proposes an interpretable deep learning framework that combines EfficientNetB3 with a Convolutional Block Attention Module (CBAM) to strengthen feature extraction and attention to regions of interest that are diagnostically significant. To facilitate clinical trust, explainable AI (XAI) is performed using Gradient-weighted Class Activation Mapping (Grad-CAM). Evaluated on a HER2-IHC-40x-WSI dataset of 10,997 image patches distributed over four HER2 classes (0, 1+, 2+, 3+), the proposed model achieved an overall accuracy of 0.96% and a macro-averaged F1-score of 0.93%, demonstrating strong performance, particularly in borderline cases. The system also demonstrates robustness across varied samples, highlighting its generalization capability. These results highlight the potential of applying attention mechanisms with explainable AI for stable and interpretable HER2 IHC scoring in digital pathology.
12:45 BitRelation: Exploring Bit-Level Dependencies in Neural Cryptanalysis
Yue-Tian Goi and Shu-Min Leong (Monash University, Malaysia Campus, Malaysia); Raphael C.-W. Phan (Monash University, Malaysia); Ana Sălăgean (Loughborough, United Kingdom (Great Britain)); Shangqi Lai (CSIRO Data61, Australia); Wei Chuen Yau (Xiamen University Malaysia, Malaysia)
This paper applies Explainable Artificial Intelligence (XAI) to improve the interpretability of neural differential cryptanalysis on the SPECK cipher. We use Local Interpretable Model-agnostic Explanations (LIME) to analyse and visualise feature importance in neural distinguishers, giving signed contributions and absolute rankings. Signed contributions show whether, and how strongly, specific bit positions influence the model's decision, while absolute rankings reflect their importance regardless of sign. To study interactions beyond single bits, we introduce a Systematic Masking Approach to reveal relations among bits by testing if chosen combinations of masked bits alter classification accuracy. On Gohr's 8-round SPECK32/64 distinguisher, masking up to four-bit combinations shows that decisions involve multi-bit interactions rather than isolated single-bit effects. Although LIME highlights strong single-bit signals, masking reveals interaction patterns consistent with differential cryptanalysis. These findings clarify model behaviour in neural cryptanalysis and show XAI's value for exposing and visualising interaction structure in ciphertext features and decisions.
Electronics, Circuits & Devices (ECD) 1
Track C7F2 ECD 7.1: Electronics, Circuits & Devices (ECD) 7.1
Room: F2. 501 Kadamaian (Level 5)
Chair: Ismail Saad (Universiti Malaysia Sabah, Malaysia)
11:30 Development of a Strap-Based Active Back-Support Exoskeleton for Static Postural Sway Correction
Jun Han Lee and Yu Zheng Chong (Universiti Tunku Abdul Rahman, Malaysia); Siow Cheng Chan (University Tunku Abdul Rahman, Malaysia)
Postural sway control has traditionally been targeted through lower-limb or ankle-based interventions, especially critical for individuals with chronic low back pain (CLBP), older adults, and those with neurological disorders. This paper introduces a strap-based soft active back-support exoskeleton that applies corrective torque directly across the trunk and shoulders using pneumatic actuation. Trunk sway is detected via an inertial measurement unit (IMU) mounted on the sternum via nylon straps, with corrective action triggered through a lightweight ESP32-based threshold control algorithm. Fifteen healthy adults were evaluated across four static stance conditions, with the Tandem Stance Eyes Closed (TSEC) posture representing the greatest challenge to balance. When the exosuit was active, the centre of pressure (CoP) pathlength decreased by 51.5%, while RMS sEMG activity was reduced by 51.2% in the rectus abdominis, 38.3% in the external oblique, and 41.8% in the erector spinae. These results highlight the feasibility of a trunk-focused active exosuit for improving static balance control, offering a complementary alternative to conventional ankle or limb-based corrective strategies. These findings suggest the exosuit's potential as a practical tool for assisting static balance.
11:45 An FPGA Based Accelerated Perception Sub-System
Sreehari N, Naghulram S P, Sahil Ahamed S, Shreyas Srinivas A and Yamuna B (Amrita Vishwa Vidyapeetham, India); Karthi Balasubramanian (Amrita Vishwa Vidyapeetham & Amrita School of Engineering Coimbatore, India)
This paper outlines an FPGA-accelerated perception sub-system for real-time object detection employing the You Only Look Once (YOLO) v5 model. The system uses the Deep Learning Processing Unit (DPU) B4096 implemented on the Xilinx Kria KV260 platform for efficient deep learning inference. The implementation shows a 7.5× speedup over traditional CPU solutions, achieving up to 59.94 Frames Per Second (FPS) for the YOLOv5 Nano model and 17.4 FPS for the YOLOv5 Large model. The system further shows high detection accuracy with a mean Average Precision (mAP) of 0.76 and an average Intersection over Union (IoU) of 0.72 for the intelligent traffic perception system. The implementation incorporates optimized data pipelines, quantized models, and high throughput inference methodologies within the Kria Docker environment, utilizing OpenCV, Xilinx Runtime(XRT), and Vitis AI Runtime (VART). The results highlight that an effective hardware-software co-design approach, combining high throughput, optimal resource utilization, and reliable inference accuracy, significantly enhances the performance of FPGA-based AI (Artificial intelligence) workloads in applications such as surveillance, industrial automation, and production line safety.
12:00 A Novel 5-Bit Ascon s-Box with Multiplexer for FPGA Implementation
Muhammad Hafiz Abu Hassan (Universiti Malaysia Perlis (UniMAP), Malaysia); Rizalafande Che Ismail and Siti Zarina Md Naziri (Universiti Malaysia Perlis, Malaysia)
The explosive growth of the Internet of Things (IoT) has ushered in an era of smart, interconnected devices yet with it comes an urgent need for robust and efficient security. IoT devices often operate under severe constraints, including limited processing power, memory, and energy resources making traditional cryptographic solutions impractical. This has fueled the rise of lightweight cryptography (LWC), a new frontier of algorithms tailored to deliver strong security without compromising performance. Among the standout candidates, Ascon has emerged as the primary recommendation for lightweight authenticated encryption in the final round of the prestigious CAESAR competition (Competition for Authenticated Encryption: Security, Applicability, and Robustness). At the heart of Ascon lies a complex permutation process involving round constants, substitution boxes (S-boxes), and a linear diffusion layer components that demand optimized hardware design. This paper introduces novel S-box implementation techniques for FPGA platforms, pushing the boundaries of existing 5-bit hardware designs. Our approach not only improves efficiency but also secure, high-performance cryptographic systems in the age of pervasive IoT.
12:15 An Energy-Efficient 8T SRAM Cell with Self-Regulating Intramural Loop for Leakage Suppression
Shaik Lal John Basha (VIT-AP University, Amaravati, Andhra Pradesh, India); Avishkar Kant (VIT-AP University Amaravati Andhra Pradesh, India); Atul Shankar Mani Tripathi (VIT-AP University, India)
In contemporary VLSI design, memory blocks have emerged as dominant contributors to both silicon area and overall power consumption. As a result, the development of energy constrained memory architectures is critical to meet the demands of energy-efficient design. Among the available memory technologies, Static Random-Access Memory (SRAM) continues to be the preferred choice for System-on-Chip (SoC) implementations, primarily due to its high-speed operation and compatibility with traditional CMOS integration. However, aggressive scaling of MOSFET devices introduces significant challenges, particularly in the form of elevated leakage currents. These arise from factors such as reduced channel lengths, thinner gate oxides, and lower threshold voltages. These factors adversely impact the reliability and energy efficiency of contemporary systems. To address these challenges, this paper proposes a 8-transistor (8T) memory architecture that incorporates a novel Intramural Loop mechanism. This innovative approach introduces an internal, self adaptive and regulating feedback structure dynamically adjusts the biasing of internal nodes based on stored logic state. The technique operates autonomously, without requiring additional control lines or incurring area overhead, and is specifically tailored to suppress subthreshold, gate, and junction leakage currents. According to simulation studies using 45nm CMOS technology, the proposed configuration achieves a leakage power reduction of up to 99.84% at subthreshold voltages. On average, the cell demonstrates more than 90% leakage reduction across a wide voltage range. These results affirm the potential of the proposed memory architecture for ultra-low power applications such as wearable electronics, biomedical implants, and Internet of Things (IoT) devices.
12:30 Impedance Matching Technique for Dual-Band Radio Frequency Energy Harvesting Unit Utilizing 900MHz (UHF) and 2.45GHz (Wi-Fi) Frequency Band
Jefferson A. Hora (MSU-Iligan Institute of Technology, Philippines); Xi Zhu (University of Technology Sydney, Australia); Marjie Anne Thezza S. Teleron (Mindanao State University - Iligan Institute of Technology, Philippines)
This study presents a dual-band impedance matching technique designed for wireless power transfer (WPT) applications operating at 900 MHz (UHF/GSM) and 2.45 GHz (Wi-Fi/ISM) frequencies. The proposed method employs a bridged-T coil (BTC) network integrated with a single-stub tuning approach to simultaneously match the input impedance at both frequency bands, ensuring efficient power delivery. Fabricated using 65nm CMOS technology, the rectifier circuit demonstrates promising performance metrics. At an input power of -15 dBm, the power conversion efficiency (PCE) reaches 58.01% at 900 MHz and 50.52% at 2.45 GHz. Under higher input power of -3 dBm, the peak PCE improves to 68.57% and 57.44% for 900 MHz and 2.45 GHz, respectively. The design also achieves a dynamic range of 19.04 dB at 2.45 GHz, indicating robust performance across varying input levels. Furthermore, excellent return loss (S11) values are observed, with -34.48 dB and -13.27 dB at 900 MHz, and -11.28 dB and -16.27 dB at 2.45 GHz for GSM and Wi-Fi bands, respectively.
12:45 Multiplexed Biomarker Detection with Dual-Gated Organic Electrochemical Transistors: Toward Prostate Cancer Diagnosis
Ke Meng (The University of Electronic Science and Technology of China, China); Ruyou Zhang (The University of Electronic Science and Technology of China, Malaysia); Jia Zhu and Yuan Lin (University of Electronic Science and Technology of China, China)
Organic electrochemical transistor (OECT) is an ideal sensing platform due to its advantages of effective signal amplification, low working voltage, and ultra-high sensitivity, which has shown great potential in disease diagnosis and monitoring. This study developed an OECT-based aptamer sensor with a dual-gate configuration for real-time biomarker monitoring in biofluids. This innovative biosensor not only has high sensitivity but also the capability of simultaneously detecting two key biomarkers relevant to prostate cancer: prostate-specific antigen (PSA) and vascular endothelial growth factor (VEGF). The dual-gate design effectively promotes device miniaturization and provides multi-analyte detection capability, significantly enhancing the diagnostic accuracy for prostate cancer. The OECT-based aptamer sensor with the dual-gate configuration exhibits an ultra-low detection limit (as low as 100 fg/mL) and a wide linear range (100 fg/mL to 1 μg/mL) towards biomarkers. This work lays a solid foundation for the development of next-generation portable and multifunctional diagnostic tools towards prostate cancer.
Power, Energy & Electrical Systems (PES)
Track C7F3 PES 7: Power, Energy & Electrical Systems (PES) 7
Room: F3. 502 Mesilau (Level 5)
Chair: Chai Chang Yii (Universiti Malaysia Sabah, Malaysia)
11:30 Multi-Timescale Hierarchical Coordination of Market-Based Day-Ahead Dispatch: Practical Applications and Prospects in CSG
Xin Yin (South China University of Technology, China); Haoyong Chen, Prof. Chen (South China University of Technology, China & Universiti Tunku Abdul Rahman, Malaysia); Yiping Chen, He Huang and Yuefeng Lu (China Southern Power Grid, China); Xin Zeng (South China University of Technology, China)
A high proportion of grid-connected renewable energy in power systems represents an economically viable and reliable pathway toward achieving clean and low-carbon power systems. However, the safe and efficient integration of unstable and uncontrollable renewable energy sources remains a significant challenge for power grid operators. Moreover, indiscriminate investment in flexible resources may result in suboptimal resource utilization. Drawing on operational experiences from the power dispatching and controlling center of China Southern Grid (PDCC-CSG), this paper demonstrates how PDCC-CSG mobilizes flexible resources through dispatch mechanisms such as hierarchical coordination, multi-timescale coordination, and "power market plus dispatch" coordination approaches to enable the efficient utilization and consumption of renewable energy in scenarios where new energy accounts for more than 20% of the total power generation. Furthermore, in order to cope with the scenarios with an even higher share of renewable energy, the future application of advanced cooperative and intelligent dispatch technologies is envisioned.
11:45 Development of a Single Phase Shunt Active Power Filter Using Synchronous Reference Frame and Self-Charging Algorithms
Muhammad Ammirrul Atiqi Mohd Zainuri (Universiti Kebangsaan Malaysia, Malaysia); Yushaizad Yusof (Jabatan Kejuruteraan Elektrik, Elektronik & Sistem, Fakulti Kejuruteraan & Alam Bina, UKM, Malaysia); Ahmad Asrul Ibrahim, Nor Azwan Mohamed Kamari, Mohd Hairi Mohd Zaman and Mohd Asyraf Zulkifley (Universiti Kebangsaan Malaysia, Malaysia)
The increasing use of non-linear loads in modern electrical power systems has led to significant power quality concerns, primarily due to the generation of harmonic currents. This study explores the effectiveness of a single-phase Shunt Active Power Filter (SAPF) in mitigating these harmonics. The primary objective is to design a SAPF using advanced harmonic extraction methods and DC-Link capacitor voltage control algorithms to maintain Total Harmonic Distortion (THD) below 5%, in compliance with IEEE standard 519-1992. The system's performance is assessed through simulations in MATLAB/Simulink, under both steady-state and dynamic conditions. A comparative analysis is conducted between the proposed method and the conventional Proportional-Integral (PI) controller. The study highlights the benefits of using the synchronous reference frame (SRF) technique combined with a self-charging mechanism and fuzzy logic controller (FLC). Results show that this approach delivers improved harmonic mitigation, achieving lower THD, faster response time, and reduced overshoot and undershoot compared to traditional methods. These findings suggest that the proposed SAPF configuration offers a promising solution for enhancing power quality in systems affected by non-linear loads, making it suitable for modern industrial and residential applications requiring reliable and efficient harmonic compensation.
12:00 Geospatial Analysis of Biomass Energy Systems: Site Suitability, Energy Yield, and Validation
King Harold A Recto (Ateneo de Manila University, Philippines)
This study investigates the potential of biomass energy as a renewable power source through geospatial analysis and system evaluation. Using Geographic Information Systems (GIS), it assesses the suitability of sites for biomass energy development based on resource availability, land use, and proximity to existing energy infrastructure. The research includes an evaluation of the current energy supply landscape, quantification of annual biomass energy potential from agricultural sources (rice, corn, sugarcane, and coconut), and estimation of projected energy generation. A cost-benefit and sustainability analysis was also performed to examine the environmental, social, and economic implications of deploying biomass energy systems. The validation process ensured data reliability and alignment with practical deployment considerations. Results indicate strong potential for integrating biomass energy into regional power systems, particularly in areas with abundant agricultural waste. However, high capital costs remain a key challenge. The study recommends further investigation into the optimal operating point ("Q-point") for biomass energy within a hybrid renewable energy framework to enhance sustainability, accessibility, and energy security.
12:15 Performance of Transformer Insulation Paper in the Presence of Multi Walled Carbon Nanotube (MWCNT) in Palm Oil Methyl Ester (POME)
Nurul Izzati Hashim (Universiti Malaysia Sarawak, Malaysia); Shirley Anak Rufus (Universiti Malaysia Sarawak (UNIMAS) & Universiti Teknologi Malaysia (UTM), Malaysia); Nazreen Junaidi (Universiti Malaysia Sarawak, Malaysia); Sharifah Masniah Wan Masra (Universiti Malaysia Sarawak (UNIMAS), Malaysia); Yanuar Arief (UNIMAS, Malaysia); Nur Eryshazana Amirulzaki Hassan (Universiti Malaysia Sarawak, Malaysia)
The reliability of power transformers depends greatly on their insulation systems, which traditionally use Mineral Oil (MO) and Kraft paper. Due to environmental concerns with MO, Palm Oil Methyl Ester (POME) has emerged as a sustainable alternative, though it requires performance enhancement for high-voltage use. This study investigates the effect of adding 0.02 g/L Multi-Walled Carbon Nanotubes (MWCNTs) to POME on the mechanical and dielectric properties of Kraft paper, under both unaged and thermally aged conditions. Tensile strength (TS) was evaluated in both machine (MD) and cross directions (CD), while dielectric strength was assessed using AC Breakdown Voltage (ACBDV) tests per IEC standards. Tensile testing confirmed the anisotropic behavior of Kraft paper, with MWCNT addition improving strength by 13 % in the cross direction and showing only a 1.4 % decrease after aging. In the machine direction, strength remained high with minimal changes. All samples met IEC 60641-3-2 standards. AC breakdown voltage improved by 5.8 % in unaged and 3.0 % in aged MWCNT-treated samples, while aging effects were less severe compared to pure POME samples.
12:30 Customer-Centric Power Reliability Assessment of Selected Cebu Distribution Utilities via Real-Time Localized Intelligent Power Monitoring System
Wilen Melsedec O. Narvios, Jayson C Jueco, Rafran P de Villa, Ferdinand F. Batayola, Gilbert Silagpo and Maria Gemel B Palconit (Cebu Technological University, Philippines)
There is restricted access to reliable data from distribution utilities due to privacy concerns, and inadequate infrastructure results in frequent outages and hinders analysis of power reliability issues in regional areas in the Philippines. The paper evaluated power distribution reliability by calculating the Customer Average Interruption Duration Index (CAIDI) using data from a real-time intelligent monitoring system across multiple sites served by local utilities. The intelligent monitoring system utilized an ETL model to collect and manage data via cloud infrastructure and integrate AI models to detect anomalies in the system. VECO and CEBECO I exceeded DOE CAIDI limits with values of 185 and 140 minutes, respectively, while CEBECO II, III, and MECO demonstrated strong reliability with zero interruptions in key areas. To address these gaps, the paper recommends deploying reclosers, advanced outage management systems, and integrating distributed energy resources to reduce outage durations by up to 60% and enhance grid resilience.
12:45 Solar PV Power Generation Forecasting Employing Feedforward Artificial Neural Network for Virtual Power Player Setup
Muhammad Haiqal Mohd Aminuddin, Madihah Md Rasid and Syed Norazizul Syed Nasir (Universiti Teknologi Malaysia, Malaysia)
Integrating renewable energy-based distributed generation (RE-DG) into modern electrical grids is reshaping power system operations. However, this transition brings notable challenges for grid operators, especially in managing variable generation and maintaining reliable supply. A major concern is the need for accurate forecasting to plan generation schedules effectively and avoid excessive reliance on standby sources. This work focuses on improving PV output prediction to support distributed generation scheduling in a Virtual Power Player (VPP) setup. Therefore, this study developed an Artificial Neural Network (ANN) model trained on historical irradiance and weather data from the Solcast Historical Time Series Service. The model was tested on a 39 MWp PV plant located at 1°32'45.6"N, 103°40'12.5"E. The forecasting results show that the approach can deliver high prediction accuracy with 8.3766×8.376×〖10〗^(-6)at epoch 7 mean square error (MSE), and 0.256% range on the scatter plot beyond 30MW. This model may be employed to help operators allocate renewable resources more effectively and reduce backup generation needs across the distribution network.
Electronics, Circuits & Devices (ECD) 2
Track C7F4 ECD 7.2: Electronics, Circuits & Devices (ECD) 7.2
Room: F4. 503 Dinawan (Level 5)
Chair: Somesh Kumar (ABV IIITM Gwalior India, India)
11:30 Electroencephalogram-Based Feature Classification During Swallowing Using Deep Learning
Rui Takahashi, Shuya Shida and Kyoko Yamazaki (Toyo University, Japan); Motoki Arakawa (Biomedical Engineering, Japan); Kaoru Miyano and Yutaka Suzuki (Toyo University, Japan)
Food texture, which includes properties such as viscosity and elasticity, considerably affects swallowing ease and sensory perception. This study aims to objectively evaluate the physiological responses when swallowing jelly drinks with varying physical properties by analyzing electroencephalogram (EEG) signals. EEG data were collected while participants swallowed four different commercially available jelly drinks. Spectrograms obtained after preprocessing and time-frequency conversion using the short-time Fourier transform were input into the EfficientNetB7 deep learning model for classifying the jelly types based on the EEG patterns. This model achieved high classification accuracy on the training data; however, its performance on validation data was notably lower, suggesting potential overfitting. Jelly1 and Jelly3 were frequently misclassified because of their similar textures, while Jelly4 showed relatively higher classification accuracy, which can be attributed to its distinct physical characteristics. These findings suggest that EEG signals recorded during swallowing contain texture-related neural signatures, and deep learning models can be used to partially capture these differences. This study contributes to the development of neurophysiological methods for food texture evaluation and lays the foundation for applications in dysphagia assessment and food engineering.
11:45 Core-Shell Nanostructures for Dynamic Color Control in Electrochromic Plasmonic Nanopixels
Kawshik Nath (Bangladesh University of Engineering and Technology, Bangladesh & Chittagon University of Engineering and Technology, Bangladesh); Md. Shariful Islam (Bangladesh University of Engineering and Technology & Bangladesh Telecommunications Company Limited, Bangladesh); Bibekananda Nath (Bangladesh University of Engineering and Technology, Bangladesh & Chittagong University of Engineering and Technology, Bangladesh); Ahmed Zubair (Bangladesh University of Engineering and Technology, Bangladesh & Rice University, USA)
Electrochromic nanoparticle on mirror (eNPoM) facilitates voltage-controlled color changes through the adjustment of optical resonances at nanoscale. This study focused on designing eNPOMs integrating plasmonic cores made of Au, AZO, GZO, and ITO with an electrochromic shell made of PANI for three different configurations: cylindrical core-shell, cylindrical core-shell structure with hollow center, and pyramidal core-shell designs. We utilized finite-difference time-domain (FDTD) solver to investigate the scattering cross section and electric field distributions in oxidized, semi-oxidized, and reduced states. Moreover, chromaticity coordinates for different redox states were quantified through the CIE 1931 diagram and color differences were numerically assessed according to the CIEDE2000 standard. Remarkably, the structures comprised of GZO and Au showed CIEDE2000 color differences exceeding 50, whereas the ITO-based configuration demonstrated >56 chromatic contrast with distinct plasmonic resonances. The electric field distribution observed in this study indicated strong field confinement in oxidized states. Electron delocalization was through reduction, aligned with our calculated spectral trends. Moreover, a comparative analysis of the calculated results with WO3 based eNPoM was performed. However, PANI-based systems demonstrated relatively higher color contrast and modulation depth. Our findings will significantly enhance the development of tunable eNPOM platforms such as high-resolution nano-displays, adaptive optics, and responsive metasurfaces.
12:00 Development of a Generator Condition Monitoring Device for Predictive Maintenance Applications Considering Temperature and Vibration Analysis
Vann Anthony Viray (Mindanao State University - Iligan Institute of Technology, Philippines); Marven E Jabian (MSU - Iligan Institute of Technology, Philippines & Mindanao State University - Iligan Institute of Technology, Philippines)
A condition monitoring device was developed to support predictive maintenance strategies for industrial generators by acquiring vibration and temperature data-two common indicators of mechanical and thermal faults. The system integrated a digital accelerometer and a non-contact infrared sensor, with an ESP32 microcontroller serving as the main processing unit to monitor real-time operational parameters. Data was collected at fixed intervals and transmitted wirelessly using an NRF24L01 module. Additional features included battery monitoring, LED indicators, and an alarm system. All components were integrated onto a custom-printed circuit board and enclosed within a 3D-printed housing. The device was designed to operate non-intrusively, powered by either an external source or a rechargeable battery, with power-saving capabilities via light sleep mode for extended deployment. Initial tests demonstrated accurate sensor readings and effective wireless transmission, validating the device's capability for fault detection. Although currently limited to data acquisition and transmission, the system is capable of future integration with data logging, analytics, or machine learning models for advanced predictive maintenance. This solution offers a cost-effective and practical approach to real-time generator condition monitoring.
12:15 Development of a High-SNR, Feature-Based Machine Learning Model for Accurate Multiclass Lung Sound Classification
Jose Antonio J. Loren, Andre Joaquin D. Adiz, Al Fred C. Picana, Arvy C. Santos, J.R. S. Templanza, Seigfred V. Prado and Wally Enrico M Ingco (University of Santo Tomas, Philippines)
Lung diseases are one of the leading causes of global morbidity and mortality, accounting for more than four million deaths annually according to the World Health Organization. Although machine learning has shown promise in automating lung sound classification, most existing models are either focused on binary lung sound classification or have a narrow subset of abnormal sounds, hindering their diagnostic utility. This study addresses these limitations by developing a high-SNR, multiclass, and accurate classification model capable of identifying normal lung sounds and five abnormal types: wheezes, crackles, stridor, rhonchi, and pleural rubs. Various feature extraction techniques were compared, including the standard Mel-Frequency Cepstral Coefficients (MFCC) model, an enhanced-MFCC (eMFCC) model, and a hybrid Discrete Wavelet Transform-Short-Time Fourier Transform (DWT-STFT). The input-output Signal-to-Noise Ratio (SNR) analysis confirmed a significant improvement in signal quality, with median SNR increasing from 16 dB (MFCC) to 35 dB (eMFCC), validating the effectiveness of the enhancement techniques. Classification was performed using Support Vector Machine (SVM) and K-Nearest Neighbors (KNN), with the highest accuracy of 97.69% and 98.34%, respectively, achieved using eMFCC mean features. The results demonstrate the robustness of the proposed high-SNR feature-based model for accurate multiclass lung sound classification.
12:30 Multilayer All-Oxide Polarization-Independent Narrowband Emitter: A Step Towards the Future Thermophotovoltaic Applications
Bibekananda Nath (Bangladesh University of Engineering and Technology, Bangladesh & Chittagong University of Engineering and Technology, Bangladesh); Kawshik Nath (Bangladesh University of Engineering and Technology, Bangladesh & Chittagon University of Engineering and Technology, Bangladesh); Ahmed Zubair (Bangladesh University of Engineering and Technology, Bangladesh & Rice University, USA)
The emitter is an indispensable part of a thermophotovoltaic (TPV) energy conversion system. However, conventional metal-dielectric emitters suffer greatly from the oxidation of metal layers at high temperatures. Emitters based on all-oxide structures can be a possible solution to this problem. Here, we present a polarization and incident angle-insensitive multilayer emitter structure based on MgO/ITO composite layers for a conventional TPV system operating at 1450 to 1500 K temperature. The emission mechanism of the proposed structure was assessed using the finite-difference time-domain (FDTD) method, and the structural dimensions were optimized using a brute-force design approach. The optical simulation of the optimized structure provides a peak emission of around 98.8% at the wavelength of 1928 nm, which coincides perfectly with the spectral response of the In₀.₇₄Ga₀.₂₆As cell and the blackbody radiation of the 1450 to 1500 K heat sources. Moreover, our designed structure was polarization-independent and insensitive to the incident angle of radiation up to 70◦ for both TM and TE polarized light. This study will have an immense impact on high-temperature applications, such as thermophotovoltaic systems, photodetectors, and sensors.
12:45 Design and Comparison of Enhanced P&O and FOCV-Based MPPT for Indoor Light Energy Harvesting in 65nm CMOS Process
Rochelle M Sabarillo, Luisa Mae M Mamburao, Quezza Phola S Patulin and Winzil Khaye V Pitogo (Mindanao State University - Iligan Institute of Technology, Philippines)
This paper presents a comparative study of two maximum power point tracking (MPPT) algorithms Enhanced Perturb and Observe (P&O) and Fractional Open-Circuit Voltage (FOCV)-targeted for indoor photovoltaic (PV) energy harvesting applications. Both designs were implemented using the 65nm CMOS process in Cadence Virtuoso and integrated with a boost converter to ensure efficient energy transfer under low-light conditions. The evaluation focuses on key performance metrics including output voltage, settling time, ripple voltage, boost accuracy, and power consumption. Simulation results show that both algorithms achieve comparable performance in terms of output voltage stability and ripple control, with post-simulation ripple voltages of 0.622% for Enhanced P&O and 0.641% for FOCV, and boost accuracies above 99% for both. However, Enhanced P&O clearly outperforms FOCV in terms of settling time, achieving 370 µs compared to 810 µs, making it more suitable for applications requiring dynamic adaptation and fast-tracking response. On the other hand, while FOCV demonstrates slightly higher power consumption and slower response, it offers lower circuit complexity and simplified control implementation due to the absence of current sensing and a feedback loop. These trade-offs highlight the strengths of each technique depending on application-specific constraints.
Engineering Technologies & Society (ETS) 1
Track C7F5 ETS 7.1: Engineering Technologies & Society (ETS) 7.1
Room: F5. 504 Madai (Level 5)
Chair: Heng Jin Tham (Universiti Malaysia Sabah, Malaysia)
11:30 Sustainability in the Urban Context: Evaluating Metro Manila Residents' Intention to Accept and Adopt Vertical Farming
Shantelle Magsino, Vanessa Mae Malabuyoc and Ritchie Ybañez (National University, Philippines); Ma. Kathleen Duran (National University, Philippines & Mapua University, Philippines); Ferly Ann Revilloza (National University, Philippines)
Rapid urbanization in Metro Manila has intensified challenges in food security, environmental sustainability, and resource efficiency. This study investigates the intention of Metro Manila residents to accept and adopt vertical farming as a viable urban agricultural solution. Anchored on the Theory of Planned Behavior, Sustainable Development Theory, and Diffusion of Innovation Theory, the research examines how economic, societal, and environmental factors influence behavioral intention. A structured questionnaire was administered to 100 residents using stratified random sampling. Data were analyzed using Partial Least Squares-Structural Equation Modeling (PLS-SEM) to identify the relationships among variables. Results revealed that economic and environmental factors significantly affect residents' adoption intention, while societal factors and attitudes toward sustainability integration showed limited influence. The study also highlights the role of perceived behavioral control and subjective norms in shaping adoption behavior. Findings suggest that enhancing public awareness, policy support, and cost-accessibility could facilitate the integration of vertical farming into urban sustainability initiatives. This research provides evidence-based insights for policymakers, urban planners, and agricultural innovators aiming to promote vertical farming in densely populated cities like Metro Manila.
11:45 Informatics Analysis of Glycosylation Pathway by Network Flow Algorithm and Machine Learning
Taisei Matsuo and Kento Totsuka (Soka University, Japan); Akira Togayachi and Kiyohiko Angata (Glycan and Life System Integration Center (GaLSIC), Japan); Shinomiya Norihiko (Soka University, Japan)
Glycans, often termed the "third life chain," are fundamental biomolecules exhibiting unparalleled structural complexity, crucial for diverse biological processes including cancer and cellular communication. Their intricate, branched architectures and non-template driven biosynthesis pose significant challenges for traditional biochemical analysis, necessitating advanced computational approaches. This research addresses the urgent need for in silico modeling of glycan dynamics by focusing on directed glycosylation pathways, which represent the sequential biosynthesis of glycans. Leveraging insights from recent advancements in graph neural networks for glycomics, this study pioneers the application of informatics to glycosylation pathway analysis. Our primary objective is to develop a robust, informatics-based tool that enables researchers without specialized biological expertise to accurately analyze glycan data. Specifically, we propose constructing directed graphs from biochemical glycosylation pathway data and applying the Edmonds Karp algorithm to quantify glycan production and Node2Vec to compare structural properties of pathways. This framework will facilitate the comparison of glycosylation maps across different cancer stages, an area with rich biological data but limited computational evaluation methods.
12:00 Investigation of the Mechanical Properties and Thermal Conductivity of Lightweight Concrete Hollow Blocks with Crushed Coconut Shell as Partial Replacement of Fine Aggregates
Brian Ivan Atienza, John Aeron Humphrey Padilla and Rosendo Jr De Guzman (Philippines); Joseph Carlo Labampa, Bryan De Guzman and Kaycee T. Alcantara (National University, Philippines)
This study examines the use of crushed coconut shells (CCS) as a partial substitute for fine aggregates in lightweight concrete hollow blocks (CHBs). The research evaluates how CCS affects key properties, including compressive strength, thermal performance, and water absorption. Three replacement levels were tested: 5%, 10%, and 15% CCS. Results showed that CCS reduced the density of CHBs by 3.5% to 5.4% compared to conventional lightweight concrete. However, compressive strength decreased by 23.9% to 33% as CCS content increased, making the blocks suitable only for non-load-bearing applications. On the other hand, CCS improved thermal insulation, with lower thermal conductivity observed at higher replacement levels. Water absorption rates increased with CCS content due to its porous structure, but stability improved at 10% to 15% replacement. The findings suggest that 10% CCS replacement offers the best balance between thermal benefits and structural performance. This makes CCS a viable sustainable material for non-structural construction, particularly in tropical climates where thermal efficiency is important. Further research should focus on optimizing mix designs, assessing long-term durability, and exploring large-scale applications to promote wider adoption in the construction industry.
12:15 Implementation of Virtual Reality Based Home Automation System
Harinatha Reddy Chennam and Raghuram Reddy Gajula (G Pulla Reddy Engineering College, India); Vuyyuru Lakshmi (JNTU, India); Pradeep Kumar Allagadda, Bramhananda Reddy Teegala and Siva Reddy Y V (G Pulla Reddy Engineering College, India)
In today's fast growing economies, Home Automation has become a crucial part. There is a rapid change in it's technology and various automating strategies are implemented. The, and New ways of producing super sensor systems is growing up as the conceptual understanding for automation has been changed. Home automation is the only way to manage everything. In Indian houses almost all the people make mistakes while using normal electrical circuit combination. They make mistake of not switching off the electrical gadgets when not in use. They even forget when go out or completely forget about it. The result of this is the wastage of energy when it is not in use. In order to avoid these drawbacks, automation techniques can be implemented. This paper presents how to fill the gap between the company and client in imagination of Home Automation system before the installation in real world. In this paper a virtual home automation system that reflects the real world is presented. Client can experience his desired home automation system even before the installation in real world. The Virtual Home Automation is developed using Unreal Engine Software
12:30 SpeechPal: Specialized Speech Aid Device for Therapists Assisting Children with Repaired Cleft Lip and Palate
Jafeth C Estocado, Jan Carlo A Magpantay, Joyce Ann P Precilla and Prince Earl M Salva (FAITH Colleges, Philippines); Marco A Burdeos (First Asia Institute of Technology and Humanities, Philippines)
Children with repaired cleft lip and palate (CLAP) often face challenges in speech clarity and communication, necessitating tailored therapeutic support. In response, the researchers designed and implemented SpeechPal, an innovative assistive technology intended to aid speech therapists in providing effective therapy. The system integrates three key features: text-to-speech and speech-to-text conversion, nasal airflow detection, and augmentative and alternative communication (AAC) functionality. The project implementation involved Intel Core i7-9700 (9th Gen) with MSI GeForce GTX 1050 Ti Dual Fan OC, Ypa 4016 Headset Microphone, HXV710 Nasal Air Flow Sensor, ANMITE 14" HDR IPS FHD Portable LED Gaming Touch Monitor, HuBERT Model, Google Speech Recognition, and PyQt6-Based Graphical User Interface for AAC. The results showed that the researchers successfully implemented the system and achieved the objectives of the study. This research highlights the potential of SpeechPal to address speech therapy challenges for children with repaired CLAP, offering a reliable, efficient, and userfriendly solution that promotes inclusivity and improved communication.
12:45 Early Flexural and Thermal Behavior of Recycled PET Fiber-Reinforced Fiber Cement Boards for Sustainable Applications
Ahl Paz, Jeremiah Lara and Chester Cortez (Philippines); Jomar Llanto and Kaycee T. Alcantara (National University, Philippines); Ma. Kathleen Duran (National University, Philippines & Mapua University, Philippines)
This study examines the flexural and thermal performance of fiber cement boards (FCBs) reinforced with recycled polyethylene terephthalate (PET) fibers as a sustainable alternative to conventional materials. Mixes with 5%, 10%, 15%, and 20% PET content by weight were prepared and evaluated against a control. Flexural strength was measured after seven days using a three-point bending setup, while thermal behavior was assessed under simulated tropical conditions using a temperature-controlled environment. The 15% PET mix showed the highest flexural strength, although it remained below the commercial board benchmark, indicating the need for further optimization. The 5% PET mix demonstrated the most consistent thermal insulation, likely due to improved pore structure at lower fiber content. While PET fiber inclusion enhanced post-crack behavior and thermal resistance, early-age mechanical performance remained limited and did not meet industry thresholds. Nonetheless, incorporating recycled PET supports sustainable construction by reducing plastic waste and encouraging circular material utilization. Future research should include cost analysis, hybrid fiber development, and field-based durability evaluation.
Computing & Computational Intelligence (CCI) 2
Track C7F6 CCI 7.2: Computing & Computational Intelligence (CCI) 7.2
Room: F6. 505 Sepilok (Level 5)
Chair: Pei Yee Chin (Universiti Malaysia Sabah, Malaysia)
11:30 iTRAC: Intelligent Tracking and Real-Time Analysis on Cloud Using RFID and MQTT
Dr. Jayanthi Ganapathy, Sr (Sri Ramachandra Institute of Higher Education and Research, India); Nachiappan N (National University of SIngapore, Singapore); Sreevijnya M (Sri Ramachandra Institute of Higher Education and Research, India & Agilisium Consulting, India); Purushothaman Ramachandran (Sri Ramachandra Institute of Higher Education and Research, India)
This research explores real-time monitoring of diverse assets using AIoT, leveraging edge-cloud collaboration for enhanced responsiveness and intelligence. RFID sensors are employed to validate edge computing functionality in tracking and anomaly detection. RFID tag reads are processed on the edge using Python scripts, which collect temporal data such as tag frequency and signal strength with precise timestamps to identify irregularities. A compound AI system integrates an LSTM model for data encoding and normalization, and an Isolation Forest for outlier detection on critical features. These models are containerized using Docker to enable consistent deployment across edge devices and cloud platforms. Processed RFID data is transmitted from a Raspberry Pi to a private OpenStack cloud via MQTT using QoS level 1, ensuring reliable, low-latency anomaly reporting. A lightweight dashboard interface enables real-time visualization and monitoring of anomalies. The system is scalable, secure, and adaptable for applications in logistics, healthcare, and industrial asset tracking.
11:45 Unlocking Battery Health: Real-Time State of Health Estimation Using Deep Learning on Partial Charging Data Segments
Hasanur Zaman Anonto (American International University-Bangladesh, Bangladesh); Md Ismail Hossain (American International University Bangladesh, Bangladesh); Isha Das (Chittagong University of Engineering and Technology, Bangladesh); Afrin Tanzila Rabbani and Samiul Ahasan Sajid (American International University-Bangladesh, Bangladesh); Md Aktar Hossain and Minul Khan Rahat (Lamar University, USA); Abu Shufian (American International University-Bangladesh, Bangladesh)
Partially and randomly based charging information on electric vehicles and energy storage systems is essential in the proper selection of charging strategies to guarantee effectiveness and safety, and this depends heavily on the accurate online estimation of battery state of health (SOH). This paper presents a comprehensive end-to-end evaluation protocol that uniformly samples a variety of different aspects of partial charging and a comparison of three kinds of neural network architecture feed-forward (FNN), convolutional (CNN), and long short-term memory (LSTM)) under direct and transfer learning conditions. The latency and error rates of inference on each of these models are profiled to determine viability to be deployed on-board. Explainability methods detect key charging intervals that work on the predictions. Observations demonstrate that rather than just lightweight implementation, FNN with moderate accuracy, when compared to CNN, leads to enhanced performance through the extraction of local patterns and the best precision when compared to LSTM which models sequences. All the architectures are always improved with transfer learning, with average errors roughly decreasing by 0.05-0.1% point and error bands getting narrower. As far as model compression is concerned, it is posited that the application of efficient LSTM variants would accommodate embedded hardware constraints. The development of this work will offer usable information on how to pick and tune deep learning models to perform well, maintain real-time SOH monitoring in realistic charging situations.
12:00 A Single-Shot Multi-Box Detector with MobileViT Backbone for Metallic Surface Defect Detection
Adelson Lok Thien Chee, Saaveethya Sivakumar, King Hann Lim, Ing Ming Chew and Chye Ing Lim (Curtin University Malaysia, Malaysia); Siew Eng Fui (Press Metal, Malaysia)
Defect detection is an essential step to ensure the quality of manufactured metal products. Industrially, detecting metallic surface defects can be challenging due to a resource constrained environment in terms of hardware computational availability. In this study, we explore a lightweight defect detection model, specifically the single-shot multi-box detector variant called SSDLite. In this study, the SSDLite is paired with varying backbone base networks, MobileViT, MobileNetv2 and MobileNetv3. Transfer learning is employed to enhance SSDLite learning by enabling a relation between previous tasks and the targeted task, which is a domain-specific task. The same implementation details are applied across all the SSDLite models (MobileViT, MobileNetv2 and MobileNetv3) being trained on the PASCAL VOCdataset, and then the prior knowledge in the form of pre-trained weights is used to fine-tune the model on a domain specific metallic surface defect dataset. The metallic surface defect in this study is based on a hot-rolled steel strip surface defect dataset called the NEU-DET dataset. This study lays the groundwork for future investigations into alternative backbone architectures with SSD for enhanced detection accuracy and efficiency in real-world manufacturing applications.
12:15 ContractIQ: A Multimodal RAG-Based Agentic System for Intelligent Contract Understanding
Abhay A Rao, Abhay Raghavendra Revankar, Nikita Nair, Shreya Mittal, Uma D and Ujjwal Mohan Kumar (PES University, India)
When organizations in industries like finance, health-care, real estate, and technology deal with intricate, high-risk contracts, legal contract analysis cannot be avoided. Accurate and proper analysis assists in the interpretation of minute details, evaluation of risks, and adherence in contracts. Smart contractual automation is needed since manual checking is cumbersome and susceptible to errors. By applying Retrieval-Augmented Generation (RAG), an agentic eleven-coordinated agent, and Chain-of-Thought (CoT) prompting to interpret legal analysis, the study proposes a legal contract analysis system based on Google's Gemini large language model (LLM). The system includes Clause extraction, definition detection, risk analysis, compliance checks, QnA and summarization. Gemini can be finely adapted to be used in the legal field through LoRA finetuning. Testing achieves strong performance: ROUGE-1 of 0.42, ROUGE-2 of0.38, ROUGEL of 0.40, F1 score of 0.70, and BLEU of 0.45, indicating high-quality legal text generation. This solution significantly reduces manual labor but with enhanced legal precision and compliance.
12:30 Ganoderma Detection in Oil Palm Plantations Using UAV Hyperspectral Imaging and AI
Chee Seng Kwang and Siti Fatimah Abdul Razak (Multimedia University, Malaysia); Sumendra Yogarayan (Multimedia University (MMU), Malaysia); Abdul Mateen Montree Bin Muhammad and Ai Ling Choo (iRadar Sdn Bhd, Malaysia); Shahrul Azman Bakar (FGV R&D Sdn Bhd, Malaysia); Haryati Abidin (University Putra Malaysia & FGV R&D, Malaysia)
The imperative for early and accurate detection of Ganoderma Boninense infections in oil palms is paramount to mitigating the devastating impact of basal stem rot. This disease poses a significant threat to palm oil production and the economic stability of affected regions. Conventional detection methods rely heavily on visual inspection or destructive laboratory analysis, which are time-consuming, labour-intensive, and cost-inefficient for large plantations. To address these limitations, this paper proposes a novel method for Ganoderma detection using frame-based hyperspectral imaging captured by an unmanned aerial vehicle (UAV). The approach incorporates feature-based band registration to correct spectral misalignments, followed by dimensionality reduction using principal component analysis (PCA). Support vector machines (SVMs) were evaluated for classification alongside a fine-tuned ResNet50 model. The results demonstrated that the TF-ResNet50 model achieved an accuracy of 84%, with a sensitivity of 74% for early infected trees and a specificity of 86% for healthy trees, underscoring the potential of UAV-based hyperspectral imaging for scalable, non-destructive disease monitoring in oil palm plantations.
12:45 Automated Nuclei Segmentation in PR-IHC Breast Cancer Images Using the Cellpose Deep Learning Model
Hasanul Bannah and Mohammad Faizal Ahmad Fauzi (Multimedia University, Malaysia); Sarina Mansor (MMU, Malaysia); Md. Shoukhin Khan, Wan Siti Halimatul Munirah Wan Ahmad and Md Serajun Nabi (Multimedia University, Malaysia); Seow-Fan Chiew, Phaik Leng Cheah and Lai Meng Looi (University Malaya Medical Center, Malaysia)
In digital pathology, precise nuclei segmentation in immunohistochemical-stained tissue sections is essential for clinical decision-making and subsequent quantification of biomarkers. This task is particularly important for the analysis of hormone receptors in breast cancer, where the status of the progesterone receptor (PR) plays a key role in determining the response to treatment. However, because of differences in nuclear morphology, staining intensity, and overlapping structures, nucleus segmentation in PR-IHC images is still difficult. In order to separate PR-expressing nuclei from high-resolution breast cancer histopathology images, we present an automated instance segmentation framework in this work that is based on the Cellpose deep learning model. supported by an entirely novel ground truth (GT) dataset produced by a hybrid pipeline. In order to create the GT, 250 high-resolution PR-IHC images with trustworthy binary nuclei masks were combined with automated segmentation (StarDist), extensive manual corrections, and multi-round pathological validation. On the test set, our Cellpose-based approach consistently performs properly, achieving an average F1 score of 0.8535, precision of 0.8882, recall of 0.8215, and IoU of 0.7445. Strong segmentation of extracted and overlapping nuclei is confirmed by visual results. This study offers a useful resource for future research in hormone receptor quantification and computational pathology, as well as the first automated segmentation benchmark for the PR-IHC dataset.
Engineering Technologies & Society (ETS) 2
Track C7F7 ETS 7.2: Engineering Technologies & Society (ETS) 7.2
Room: F7. 506 Selingan (Level 5)
Chair: Siti Nurfadilah Binti Jaini Status (Universiti Malaysia Sabah (UMS), Malaysia)
11:30 Surface EMG - Based Quantitative Assessment of Muscle Coordination and Complexity in Post-Anterior Cruciate Ligament Injury Individuals
Arunthathi S (Anna University, India); S Saranya (Sri Sivasubramaniya Nadar College of Engineering, India); Rakshana R (SSN College of Engineering, India)
Surface Electromyogram (sEMG) signals can be used as inevitable tool to assess the muscle coordination and complex activation muscles, particularly strength and order of muscle recruitment during walking. The muscle coordination feature like Co-Activation Index (CAI) is an important parameter to examine the contribution of specific antagonist and agonist muscle pairs. In the proposed study, features which exhibits complex behavior of muscles like kurtosis, Correlation Dimension (CD), Sample Entropy (SE) and Skewness were used to compare the variations among post Anterior Cruciate Ligament (ACL) injury and normal subjects. The dataset consists of surface EMG signals obtained from Medial Hamstrings (MH), Biceps Femoris (BF), Vastus Lateralis (VL), Gastrocnemius Medialis (GM), Gastrocnemius Lateralis (GL) from Control Groups (CG) and individuals undergoing post-ACL injury Reconstruction (PAIR) between 18 and 30 years of age. CAI was calculated between VL & BF muscle pairs. CAI (p=0.84) and kurtosis measures fail to show the significant differences when comparing PAIR with CG while other features like CD (PGM=0.04), SE (PMH=0.04) and Skewness (PGL=0.03) provide the significant difference. Therefore, the proposed study contributes to the understanding of the dynamics of EMG features mentioned and its importance in quantitative assessment of post-ACL injury reconstruction.
11:45 Future Trip Profile Nomination (FTPN): A Framework for Proactive EV Routing
Farhaan P Alawiya and Abdul Aziz G. Mabaning (Mindanao State University - Iligan Institute of Technology, Philippines)
Future Trip Profile Nomination (FTPN) is proposed as a proactive framework for electric vehicle (EV) routing and traffic coordination. Built upon the broader concept of Future Behavior Nomination (FBN), FTPN allows EV users to voluntarily submit anticipated trip information-including origin, destination, departure time, and optional waypoints. These submissions are aggregated to support optimized routing decisions that minimize traffic congestion and ensure sufficient battery charge levels, taking into account each vehicle's range and the availability of charging infrastructure. Foundational models are introduced to integrate these behavioral nominations with conventional forecast data, incorporating time-dependent accuracy and varying user participation rates. A simulation-based validation demonstrates that the proposed models can significantly improve trip prediction accuracy by fusing user nominations with conventional forecasts. By shifting from reactive, demand-driven routing to a behavior-informed paradigm, FTPN supports the development of Proactive Vehicle Traffic Management (PVTM) systems. This framework offers a scalable and intelligent approach to mobility planning that aligns with United Nations (UN) Sustainable Development Goals (SDG), particularly SDG 11 on sustainable cities and communities, and SDG 13 on climate action through improved energy efficiency and reduced transportation emissions.
12:00 A Method for Generating Panoramic Borehole Images via 3D Analysis Using Gaussian Splatting
Kyuhei Honda (National Institute of Technology, Oita College, Japan)
This study investigates a method for converting borehole camera images, used for evaluating the integrity of geological formations and rock masses, into three-dimensional point cloud data for high-precision analysis of borehole wall geometry. The procedure is as follows. First, a camera-equipped probe is moved vertically within the borehole to acquire continuous video images. Then, the borehole wall is reconstructed in 3D using a technique called 3D Gaussian Splatting. This technique enables rapid and high-quality representation and shape estimation of the 3D point cloud. The wall surface irregularities are represented by the point cloud, while wall textures are derived based on Gaussian distributions. The acquired 3D data often contains noise, such as points located away from the actual borehole wall. To address this, the borehole wall is approximated as a cylinder, and noise is removed based on the estimated cylinder axis and radius. Unfolded panoramic images are generated using the cylinder axis and the color information associated with each point. This approach significantly enhances the quantitative assessment and operational efficiency of borehole investigations, contributing to preventive maintenance against infrastructure deterioration and geological hazards.
12:15 Projection Net: A CNN Framework for Segmention of Teeth from Panoramic X-Ray Images
Nagaraj Yamanakkanavar (Central University of Karnataka & CUK Karnataka, India); Sibasankar Padhy (Indian Institute of Information Technology, India); D Chaitra (Indian Institute of Information Technology, Dharwad, India); Sameena Begum (Central University of Karnataka, India); Santosh Uppinal (ESIC Hospital, India)
The segmentation of X-rays and computed tomography (CT) images is crucial for identifying and separating tooth characteristics. This process plays a vital role in various clinical applications such as dental diagnostics, treatment planning, and surgical procedures. Moreover, a detailed analysis of tooth structures from segmented X-ray images enables accurate diagnosis of specific oral health conditions. Among these, deep learning techniques have gained significant attention due to their ability to produce effective results on large datasets. Consequently, deep learning is increasingly favored over traditional machine learning approaches. In this paper, we aim to explore current deep learning-based segmentation algorithms used for quantitative tooth analysis in diagnosing oral health issues. The proposed projection module, integrated with group convolutions, enhances feature aggregation, leading to improved accuracy while also reducing complexity (number of trainable parameters). An evaluation of the proposed method was carried out using images from the State University of Southwestern Bahia's diagnostic imaging center (UFBA-UESC). In this facility, a dental dataset is available with 1500 images that can be used to segment teeth based on panoramic X-rays. As a result of applying the proposed method to the UFBA-UESC dataset, a mean accuracy of 0.97, precision of 0.92, recall of 0.91, and F1-score of 0.92 were achieved, surpassing the performance of existing methods for segmenting teeth images.
12:30 LIFE-MAP: an Indigenous Biosignature Detection Protocol Using Multi-Biomolecule Based Analysis
Md. Ehsanur Rahman, Tyseer Ninad, Md Anwarul Islam Aion, Md Rafin Haque and Tasin Ahmed (Aviation and Aerospace University Bangladesh, Bangladesh); Md. Samin Rahman (American International University-Bangladesh (AIUB), Bangladesh)
This paper introduces LIFE-MAP (Life Investigation Framework for Extraterrestrial Mapping and Analysis Protocol), a modular system developed to accurately identify biosignatures in extraterrestrial environments. Combining advanced mechanical, electronic, and sensor technologies, LIFE-MAP integrates biomolecular assays, microbial volatile organic compound (mVOC) analysis, and environmental monitoring to classify samples as Extant, Extinct, or NPL (No Presence of Life). With specialized sensors designed to detect proteins, carbohydrates, chlorophyll, and microbial byproducts like ethanol and formaldehyde. The system provides precise astrobiological detection using multi-biomolecule based soil analysis and CNN based rock analysis. The modular design of LIFE-MAP enables seamless integration with robotic platforms, including Mars rovers. This system marks a significant advancement in life-detection technologies, drawing inspiration from exploration missions to extreme environments where life adapts to harsh conditions, including nuclear disaster sites. The system's ability to assess all parameters positions it as a vital tool for future missions, offering a comprehensive and efficient approach to detect life across a range of extreme environments.
12:45 Characterization of Physiological, Psychological, and Physical Stress Responses from Wearable Technology
Joyce Anne A Bernardino (University of Santo Tomas, Philippines & BRAIN Lab, Philippines); Danna Francheska F. De Regla, Florenz TJ DG Galvez, Sean Archie D Gregorio, Anne Margarita C. Yu Ekey, Wally Enrico M Ingco and Seigfred V. Prado (University of Santo Tomas, Philippines)
Stress affects health and well-being, with heart rate (HR) and heart rate variability (HRV) serving as the key indicators closely related to the autonomic nervous system (ANS), which are parameters controlling stress responses. Albeit most studies focused on one to two parameters, this study explored the interplay of three, namely, physiological, psychological, and physical stress responses from wearable technology data. With the employment of the Gaussian Mixture Model (GMM), Uniform Manifold Approximation and Projection (UMAP) Manifold Learning, and statistical analysis, the research aimed to uncover hidden patterns of stress responses, assessing stress severity and binary classifications. Significant overlaps in moderate stress levels were observed between the mild and severe levels, confirming challenges in distinguishing stress levels. This led to the adoption of binary classifications, enhancing robustness and simplifying the outputs. Statistical validation confirmed strong correlations between UMAP components and stress severity. Physiological markers, including HR, electrodermal activity (EDA), and accelerometer data (ACC), consistently exhibited high correlations with severe stress levels. While skin temperature (TEMP) and interbeat intervals (IBI) contributed to moderate stress differentiation. Finally, mild stress has a notable connection to blood volume pulse (BVP), IBI, and HR.
Computing & Computational Intelligence (CCI) 3
Track C7F8 CCI 7.3: Computing & Computational Intelligence (CCI) 7.3
Room: F8. 507 Monsopiad (Level 5)
Chair: Kit Guan Lim (Universiti Malaysia Sabah, Malaysia)
11:30 Element Based User Interaction with Design Semantics of Mobile Apps and Usability Assessment
Gundala Shanmukhi Rama, Sangharatna Godboley and Ravichandra Sadam (National Institute of Technology Warangal, India)
Designing interactive UI templates is a challenging task. Designers often struggle to determine the best design choices, and even experienced professionals spend significant time evaluating layouts. Moreover, assessing whether a design is good or bad for users remains difficult. This study aims to develop a framework for predicting and scoring the placement of interactive UI elements. By providing usability scores for element placement, our goal is to help trace the users interaction and make it more efficient and assist designers in optimizing their layouts. We employ the YOLO model to detect interactive elements in UI screenshots and assess their placement on a usability scale. The model predicts element positions and usability scores, enabling designers to refine their layouts based on data-driven insights. The model evaluates UI designs by identifying interactive elements and assigning usability scores. Designers can use these scores to assess the effectiveness of their layouts and make informed improvements without direct position suggestions. Our approach enhances UI usability, helping designers create more effective interfaces aligned with current design trends.
11:45 Design and Development of Deep Learning Framework for Glaucoma Detection via Retinal Scan Analysis
Siddharth Ranganatha (R V College of Engineering, Bengaluru, India); Sujatha Badiger (RV College of Engineering, India)
A deep learning framework has been developed to automatically detect glaucoma through the analysis and segmentation of retinal scans, leveraging the U-Net architecture for precise identification of key features. The U-Net's distinctive design facilitates the effective extraction and segmentation of anatomical structures critical for glaucoma diagnosis, including the Cup-to-Disc Ratio (CDR), Rim-to-Disc Ratio (RDR), Disc Damage Likelihood Scale (DDLS), and the Inferior-Superior-Nasal-Temporal (ISNT) rule. This framework's robustness is further improved by its advanced preprocessing capabilities, which ensure accurate identification of biomarkers through techniques such as noise reduction, contrast enhancement, and data augmentation. These preprocessing steps are crucial in enhancing image quality and enabling the model to focus on relevant features. Evaluations conducted on retinal image datasets have validated the framework's effectiveness in distinguishing anatomical structures pertinent to glaucoma diagnosis. This capability supports early risk assessment and contributes to the development of scalable, cost-effective tools for glaucoma detection in clinical settings. By facilitating early intervention, this framework holds promise for improving clinical outcomes and advancing the accessibility of glaucoma screening.
12:00 A Stacking-Based Multi-View Class-Level Refactoring Prediction Framework
Hardik Hardik (NIT Kurukshetra, India); Lov Kumar and Vikram Singh (National Institute of Technology, Kurukshetra, India)
Refactoring prediction plays a crucial role in software maintenance by identifying structural improvements in code without altering external behavior. However, class imbalance and the inherent complexity of real-world codebases pose significant challenges for traditional detection approaches. In this study, we present a multi-view stacking-based framework that integrates structural features from object-oriented (CK) metrics with semantic representations derived from CodeBERT embeddings. To address data imbalance, SMOTE is applied exclusively to the combined feature representation. Each feature modality is processed by dedicated base classifiers, and their outputs are aggregated via a meta-learner in a stacking ensemble. The proposed method is evaluated on two open-source Java projects-ANTLR4 and Titan-using 22 diverse classifiers within a stacking pipeline. Results show that combining inputs with SMOTE significantly improves classification performance. For example, LightGBM and Extra Trees achieved AUC scores of 0.92 and 0.91, respectively, on ANTLR4, while Gradient Boosting and LightGBM exceeded 0.90 AUC on Titan. The stacking approach consistently outperformed single-view baselines, with notable improvements in F1-score (up to 15%) and F-measure (over 12%) across configurations. These findings validate the effectiveness of multi-view stacking with class balancing for robust and scalable refactoring prediction.
12:15 Analysis of Speech Features in Identifying Client's Change Talk in Motivational Interviewing
Shareef Babu Kalluri (University of Petroleum and Energy Studies, India); Deepu Vijayasenan (NITK, India)
Motivational Interviewing (MI) is a frequently used and effective psychotherapy approach for treating behavioral problems. MI is a collaborative interaction for understanding the client's own reasoning for a change in behavior. In this study, we analyzed an MI corpus in the nutrition and fitness domains, in which counselor and client utterances were categorized using Motivational Interviewing Skill Code (MISC). During the interaction when the client expresses the need or willingness to change is the Change Talk (CT). We aimed to analyze the speech features and proposed a BiLSTM multimodal neural network model for detecting the change talk or not a change talk. Our approach using speech and language information in detecting the change talk is par with other multimodal approaches (language and facial information) with an F1-score of 0.573 for CT. The proposed set of speech features shows the statistical significance in identifying the change talk or not a change talk.
12:30 Bridging the Gap Between Natural Language and CLI: An Intelligent Assistant Approach
Arya Ajay Gupta, Uchit N M, Lakshmi Kamath, Mahima NR and Uma D (PES University, India)
There is a large gap between natural language and command-line interfaces (CLI) because of the syntactic stiffness, non-tolerance to errors, and steep learning curve for terminal interaction. This paper introduces an AI-driven command-line assistant that interprets natural language queries into executable Unix shell commands based on a Retrieval-Augmented Generation (RAG) framework coupled with a multi-agent architecture. The platform utilizes the Gemini Large Language Model (LLM) to interpret user intent and return human-readable responses, and FAISS-based vector indexing to allow for speedy, semantically close command retrieval. The Command Optimizer module suggests sequences based on functional compatibility, usage patterns, and graph-based relationships between commands. The multi-agent pipeline breaks down user queries into dedicated stages allowing for modular reasoning and successful task execution. Experimental assessment proves contextual precision, response appropriateness, and flexibility improvements. This research validates the value of agent-based RAG models in covering the gap between natural language interaction and CLI-based system management.
12:45 Integrating Generative AI Tools with Design Thinking: a Facilitator's Manual Approach for Rapid Innovation
Ahmad Nizar Harun (Mimos Berhad & Universiti Teknologi Malaysia, Malaysia); Azman Bin Hussin, Saidatul Farrah Binti Muhammad Johar, Fatin Khairunnisa Binti Mohd Adha and Tuan Amera Binti Tuan Kamaluddin (MIMOS, Malaysia)
This paper examines a novel approach to integrating generative Artificial Intelligence (AI) tools within Design Thinking (DT) workshops, as outlined in a comprehensive facilitator's manual. The manual provides a step-by-step guide to enhance various stages of the DT process, from ideation to implementation planning with Key Performance Indicators (KPIs), within a condensed 1-day workshop format. This paper highlights the benefits of leveraging Large Language Models (LLMs) and other AI tools for accelerating idea generation, streamlining data analysis, enabling rapid prototyping, and incorporating triangulation methods for robust outcomes. It also addresses the inherent challenges and limitations of AI, emphasizing the crucial role of human facilitators in ensuring ethical, accurate, and culturally sensitive innovation. Through a hybrid human-AI facilitation framework, the manual introduces pragmatic heuristics and modular templates to guide participants from problem scoping to validated solution narratives. Ultimately, this work offers a practical contribution to educators, consultants, and innovation leaders seeking to embed responsible generative AI practices into human-centered design workflows.