Diabetes Mellitus is a significant global health issue, affecting over half a billion people worldwide. Current glucose monitoring methods are invasive, painful, and require skilled application, highlighting the need for development of effective, non-invasive, and easy to use methods. This paper presents our work on the design, development, and evaluation of a non-invasive blood glucose monitoring system, utilizing Near-Infrared Spectroscopy technique for glucose monitoring. The proposed system comprises of MAX30102 biosensor connected to an ESP32 microcontroller. The biosensor captures the photoplethysmogram signals, which are then processed by a microcontroller to evaluate blood glucose level. In order to increase the accuracy of the results, we have incorporated linear regression with Clarke error grid analysis to calibrate our system. The linear regression model is trained by comparing the results obtained through the developed system with that of commercial-off-the-self invasive device. The glucose levels obtained through the developed system are displayed in real-time on an Organic LED (OLED) screen and uploaded to a cloud server via Internet of Things (IoT) for remote monitoring. To validate the performance of the proposed system, we have compared the performance metrics of our system against existing solutions published in the literature. Performance comparison show that our system achieves a reasonably good accuracy with a root mean square error of 13.8 mg/dl and a mean absolute relative difference of 12%. The proposed system offers a painless, reliable, and convenient solution, potentially improving glucose monitoring for patients worldwide.

Emotion recognition is a valuable technique to monitor the emotional well-being of human being. It is found that around 60% of people suffer from different psychological conditions like depression, anxiety and other mental issues. Mental health studies explore how different emotional expressions are linked to specific psychological condition. Recognizing these patterns and identifying their emotions is complex in human being since it varies from each individual. Emotion represents the state of mind in response to the particular situation. These emotions that are collected using EEG electrode needs a fine grain emotional analysis to contribute for clinical analysis and personalized health monitoring. Most of the research works are based on valence and arousal (VA) resulting in two, three and four emotional classes based on their combinations. The main objective of this paper is to include dominance along with valence and arousal (VAD) resulting in the classification of 16 classes of emotional states and thereby improve the number of emotions to be identified. This paper also considers 2-class emotion, 4-class emotion and 16-class emotion classification problem and applies different models and discusses the evaluation methodology in order to select the best one. Among the six machine learning models, KNN proved to be the best model with the classification accuracy of 95.8% for 2- class, 91.78% for 4-class and 89.26% for 16-class . Performance metrics like Precision, ROC, Recall, F1-Score and Accuracy are evaluated. Additionally, statistical analysis has been performed using Friedmanchisquare test to validate the results. With the help of the experimentation, a suitable machine learning model that could perform for various classes has been identified.

Mass spectrometry is the most reliable and accurate approach for analyzing a complex biological sample and identifying its protein content, which is time-consuming and reasonably expensive. One possible option to overcome such limitations is to use potentiometric sensors based on transistors. However, for such technology to work, a protein database that contains information for billions of small peptides and amino acids (AA) is required. The only practical way to build such database is to use machine learning and this paper shows the initial steps towards achieving this aim. This study sheds light on the possibility of a new approach for peptide sequencing combining analytical simulations with Large Language Models (LLM) based on Sequence-2-Sequence (Seq-2-Seq) architecture built by Long-Short-Term Memory (LSTM) networks. 11573 tokenized data points (voltage and capacitance cross-over points) with a vocabulary size of 504 are fed to the model, 80% of data is used for training and validation, and 20% is used for testing. The model is tested on unseen data and the accuracy during the test is 71.74%, which is significant if compared to expensive and time-consuming conventional methods, i.e., spectrometry. In conclusion, the output results of this study show that the proposed Seq-2-Seq LLM architecture could be used to build a material database for a potentiometric sensor to replace the mass spectrometry method.

Precision Agriculture is one of the emerging technologies that is promising to solve the problem of food security worldwide. These focus on collecting, analyzing, and taking actions based on data available from the crop and its environment. Building low-cost and reliable plant health-related sensors is very important and helpful in the agriculture industry. This study builds a leaf reflectance sensor comprising a white LED source and an S1133 photodiode detector. The angle between the source and detector varied from 30, 45, 60, and 90 to determine the angle at which it would have an optimal reflectance value. The white LED source was connected to a 3-volt and 0.3-ampere power supply, while the S1133 photodiode detector was connected to an oscilloscope to measure the response voltage. Different green intensities were used using an RGB color scheme that imitates the color of the leaf that characterizes its health status. Reflectance intensities were calibrated using white standard reflectance. The result shows a 45-degree angle between the source and detector, with the highest correlation value (R2= 0.958) between varying green and normalized intensity compared to the reflectance angles. This study provides a portable setup for a reflectance sensor that will be used to assess plant health status and help improve crop yield in the agricultural sector.

Vibration energy harvesting has emerged as a promising approach to power small electronic devices and sensors, particularly in remote or inaccessible locations where traditional power sources are impractical. Among the various transducer mechanisms, electromagnetic vibration energy harvesters (EVEHs) have garnered significant attention due to their relatively simple design, high energy conversion efficiency, and scalability. However, the performance of EVEHs is heavily dependent on the structural configuration and relative positioning of the magnets and coils within the transducer. This study investigates the optimization of the EVEH structure to maximize the harvested voltages and power output by a comparatively study using three (3) design scenarios namely single coil-four magnets circuit, single coil-split magnet circuit and two coils-split magnet circuit within an equivalent coil magnet volume. A comprehensive parametric analysis is conducted to examine the effects of the magnet-coil gap, coil position, and magnetic flux density on the EVEH's performance. Analytical models are developed to predict the open-circuit voltage and power output on a prototype EVEH system. This work will provide valuable insights for the design and implementation of high-performance EVEHs, contributing to the advancement of self-powered and sustainable electronics. The results will demonstrate that by strategically positioning the magnets and coils, the bandwidth, harvested voltages and power can be significantly enhanced or compromised.

State of Charge (SoC) estimation is important for improving performance and longevity of lithium-ion batteries in electric vehicles (EVs). Traditional methods such as voltage measurements and Coulomb counting lie in the inability to account for factors like battery aging and operational conditions variations, leading to potential errors in SoC estimation. Accordingly, this work overcomes these limitations by utilizing Ensemble Projected Gated Recurrent Units (E-PGRUs) for enhancing SoC estimation. Traditional methods often struggle with the non-linear dynamics and transient behaviors of battery systems, leading to suboptimal predictions. The proposed E-PGRU model leverages the adaptability of GRU, which efficiently handles time-series data, while employing an ensemble strategy to mitigate the risks of overfitting and improve generalization. In our methodology, we employed a publically available dataset specifically dedicated to the particular topic of real-world EV operations involving driving cycles and capturing varying operating conditions. E-PGRU architecture consists of multiple GRU networks, with projected layers features, each trained on different subsets of the data, and their outputs are aggregated to produce a more reliable SoC estimate. This ensemble technique targets specifically variability in prediction (i.e., standard deviation minimization), increasing prediction confidence and allowing the model to learn complex patterns in the battery's operational behavior. While, the experiments of this work are ongoing, it is expected to reach higher coefficient of determination, providing an explanation of the variance in dependent variable by independent variables in SoC estimation model. The expected result will demonstrate improvements in prediction performance compared to baseline models of recurrent neural networks in both coefficient of determination (i.e., due to ensemble learning) and computational time (i.e., due to projection layers) indicating a strong alignment with SoC values. Furthermore, E-PGRU expected to show superior adaptability to different usage scenarios and conditions, suggesting its potential for application in battery management systems.

Precision agriculture (PA), leveraging wireless sensor networks (WSN) for efficient data collection, is set to revolutionize intelligent farming. However, challenges such as energy efficiency, data quality, redundant data transmission, latency, and limited WSN lifespan persist. We propose an edge computing-enabled WSN framework for PA designed to enhance network longevity by optimizing energy use through controlled data redundancy and minimized data transmission to the sink. This framework involves a two-step data aggregation process: within clusters, where the cluster head (CH) aggregates data, and at a central network point, where an edge computing-enabled gateway node (GN) performs further aggregation. Our MATLAB simulation evaluates the proposed approach against the Low-energy adaptive clustering hierarchy (LEACH) protocol and two classic sensing strategies, Periodically Sensing with All Nodes (PSAN) and Effective Node Sensing (ESN). Results reveal significant energy efficiency and network lifespan improvements. Due to reduced long-range transmissions, nodes in our scheme dissipate energy over 1500 rounds, compared to 500 rounds in LEACH. Our method sends Data packets to the CH and GN for 3000 rounds, while LEACH stops at 1500 rounds. Our approach improves network stability and lifetime, with the first node dying at 790 rounds versus 500 rounds in LEACH and the last node remaining functional until 3000 rounds compared to 1500 rounds in LEACH. Our edge computing-driven aggregation technique (ECDAT) outperforms PSAN and ESN in latency, energy usage, and QoD. At 50 Mbps, ECDAT improves communication latency by 10% over ESN and 20% over PSAN. ECDAT maintains a QoD of 100% across various valid sensor and node counts, surpassing ESN and PSAN. Our contribution integrates edge computing with WSN for PA, enhancing energy utilization and data aggregation. This approach effectively tackles data redundancy, transmission efficiency, and network longevity, providing a robust solution for precision agriculture.

This study introduces a prototype of the Warehouse Robot Localization and Communication System designed to optimize battery management and maintain an uninterrupted workflow in warehouse environments. The system includes autonomous mobile robots equipped with advanced localization and wireless communication technologies. When a robot currently assigned a task has its battery level drop below a predefined threshold, it communicates with the main computer via Wi-Fi to request assistance. An available robot then adjusts its task, navigating the shortest path to the low-battery robot's location, guided by a webcam system. Then, the low-battery robot proceeds to a charging station after transferring its task to the arrived assisting robot. The study aims to enhance productivity by reducing downtime through efficient battery management, precise localization, reliable communication, and a user-friendly control interface. To achieve localization, an overhead camera was used to capture a comprehensive image of the work floor. This image was processed to generate a detailed map tracing paths and obstacles. This mapping is only repeated if the work floor changes. The A* algorithm was integrated into the main controller program to ensure optimal pathfinding based on real-time data. The positions, orientations, and movements of the mobile robots were accurately tracked using color codes and their shapes. A Wi-Fi-based communication was integrated to facilitate data exchange with the main computer, including battery levels, orientation, path coordinates, and assistance requests. A Python-based user interface was designed for monitoring and controlling purposes. After implementation, the mobile robots successfully detected their battery levels, communicated with the main computer, and autonomously assigned tasks and localized themselves, effectively eliminating the need for manual intervention. This prototype system has the potential for further development into a market-ready product for industrial applications. Future work includes enhancing multitasking capabilities, incorporating additional sensors for improved accuracy, and optimizing power consumption management.

Fiber optic-based sensors have recently gained popularity in a variety of industrial areas. Currently, these sensors are employed to measure a variety of physical and chemical characteristics. Among the many configurations investigated, sensors based on Long Period Fiber Gratings (LPFGs) stand out for their simplicity, low-cost, and multiplexing capabilities, which allows for simultaneous monitoring of multiple parameters. These devices are periodic structures engraved in optical fibers that couple propagation modes at certain wavelengths, resulting in resonances that are sensitive to environmental fluctuations. Traditionally, measurements in these sensors were based on power fluctuations or spectrum shifts. However, the possibility for employing higher-order core-excited modes to forecast changes in observed variables has not yet been investigated. This methodology may be practical using signal processing methodologies and deep learning techniques. In this study, we developed and implemented a convolutional neural network (CNN) to predict thermal variations based on the modal distribution in LPFGs. An LPFG with a period of 560 µm and length of 33.6 mm was constructed in a few-mode optical fiber (SM2000) using a CO₂ laser etching technique. To train and verify the CNN-based model, a database of 355 empirically acquired near-field images corresponding to the LP11 propagation modes was used. The images were captured with a WIDY SWIR 640 VS infrared camera and a 980 nm laser. Similarly, the model's hyperparameters were tuned using the computational tool Optuna, which improved its overall performance. The findings show that the constructed deep learning model can predict temperature with 98.5% accuracy over a range of 26°C to 120°C, with a maximum error of 3.77°C. The root mean square error (RMSE) of the forecasts was 0.94°C, indicating that the model was accurate. Finally, the inference time for a batch of 32 images was 0.055 seconds, confirming the effectiveness of the proposed approach.

Vision and touch are fundamental sensory modalities that enable humans to perceive and interact with objects in their environment. Vision facilitates the perception of attributes such as shape, color, and texture from a distance, while touch provides detailed information at the contact level, including fine textures and material properties. Despite their distinct roles, the processing of visual and tactile information shares underlying similarities, presenting a unique opportunity to enhance artificial systems that integrate these modalities. However, existing methods for combining vision and touch often rely on data fusion at the decision level, requiring extensive labeled data and facing challenges in generalizing to novel situations.

In this paper, we leverage contrastive learning to train a convolutional neural network on textile data using both visual and tactile inputs. Our objective is to develop a network capable of extracting unified representations from both modalities without the need for extensive labeled datasets. We explore two distinct contrastive loss functions to optimize the learning process. Our results demonstrate that the shared representations effectively capture critical data structures and features from both sensory modalities, enabling successful differentiation between object classes based on both vision and touch. We validate our approach through a series of experiments, optimizing hyperparameters to maximize performance. The findings suggest that extracting shared representations for vision and touch not only enhances the integration of visual and tactile information but also provides a robust framework for multimodal perception in artificial systems.