Advancements in AI have elevated speech recognition, with convolutional neural networks (CNNs) proving effective in processing spectrogram-transformed speech signals. CNNs, with lower parameters and higher accuracy compared to traditional models, are particularly efficient for deployment on storage-limited embedded devices. Artificial neural networks excel in predicting inputs within their expected output range but struggle with anomalies. This is usually harmful to a speech recognition system. In this paper, the neural network classifier for speech recognition is trained with a “negative branch” method, incorporating directional regularization with out-of-distribution training data, allowing it to maintain a high confidence score to the input within distribution while expressing a low confidence score to the anomaly input. It can enhance the performance of anomaly detection of the classifier, addressing issues like misclassifying the speech command that is out of the distribution. The result of the experiment suggests that the accuracy of the CNN model will not be affected by the regularization of the “negative branch”, and the performance of abnormal detection will be improved as the number of kernels of the convolutional layer increases.

Hazard detection in industrial and public environments is critical for ensuring safety and regulatory compliance. This paper presents an energy-efficient, cost-effective IoT-based hazard detection system utilizing an ESP32-CAM microcontroller integrated with temperature (DHT22) and motion (PIR) sensors. A custom-built convolutional neural network (CNN) deployed on a Flask server enabled real-time classification of hazard signs, including “high voltage”, “radioactive”, “corrosive”, “flammable”, “no hazard”, “no smoking”, and “wear gloves”. The CNN model, optimized for embedded applications, achieves high classification accuracy with an F1 score of 85.9%, ensuring reliable detection in diverse environmental conditions. A key feature of the system is its adaptive monitoring mechanism, which dynamically adjusts image capture frequency based on detected activity, leading to 31–37% energy savings compared to continuous monitoring approaches. This mechanism ensures efficient power usage by minimizing redundant image captures while maintaining real-time responsiveness in high-activity scenarios. Unlike traditional surveillance systems, which rely on high-cost infrastructure, centralized monitoring, and subscription-based alerting mechanisms, the proposed system operates at a total cost of SGD 38.60 (~USD 28.50) per unit and leverages free Telegram notifications for real-time alerts. The system was validated through experimental testing, demonstrating high classification accuracy, energy efficiency, and cost-effectiveness. In this study, a hazard refers to any environmental condition or object that poses a potential safety risk, including electrical hazards, chemical spills, fire outbreaks, and industrial dangers. The proposed system provides a scalable and adaptable solution for hazard detection in resource-constrained environments, such as construction sites, industrial facilities, and remote locations. The proposed approach effectively balances accuracy, real-time responsiveness, and low-power operation, making it suitable for large-scale deployment.

This research addresses the rapid aging phenomenon prevalent in Asian societies, which has led to a significant increase in elderly individuals relocating to nursing homes due to health-related issues. This trend has resulted in social isolation and loneliness among the elderly, contributing to physical and mental ailments such as hypertension and cardiovascular diseases, as highlighted by the World Health Organization (WHO). To mitigate these issues, the research proposes leveraging technology, specifically the integration of robotics, to alleviate the caregiver shortage and enhance human interaction for the elderly. The novel approach involves developing a social robot designed to bridge the gap between humans and machines, combining knowledge from embedded systems, robotics, and essential soft skills for effective interaction. The authors found that this technological solution holds promise in addressing the caregiver shortage and improving the well-being of elderly individuals by reducing their sense of isolation, fostering better mental and physical health outcomes, and potentially transforming the landscape of elderly care through innovative technological applications. Future work includes expanding pilot studies and collaborating with healthcare institutions to further validate the effectiveness of the solution.

Micro-light-emitting diodes (μLEDs) are semiconductor devices that have been shown to have higher luminous efficacy, higher contrast ratio, and higher energy efficiency than existing mainstream technologies based on liquid crystals or organic LEDs (OLEDs). Portable display applications such as wearable devices and head-up display are some of the interesting applications of μLED displays. However, this technology has not yet been mass-produced for commercial devices due to process yields, costs, and manufacturability issues. This article presents a novel technology for the heterogeneous integration of a μLED matrix display with bipolar complementary metal-oxide- semiconductor (CMOS) DMOS (BCD) circuits that could improve manufacturability by eliminating the need for a dedicated bond stack in the bump-bonding process. To validate the concept, custom high-performance, 2-D arrays of parallel-addressed GaN blue μLEDs matrices were fabricated. The individual μLED pixel diameters are 20 and 50 μm, respectively, and the overall dimension of the array is 650 μm2. In addition, a μLED display driver-integrated circuit (IC) with a compact size of 3×4.4 mm ² has been designed, implemented, and verified experimentally for the μLED matrices. Measured output optical power-forward bias current-forward bias voltage (P-I-V) curves of the individual μLED pixel are shown. The 4 × 4 μLED matrix has also been successfully driven using active-matrix driving and display pictures to demonstrate the function of active-matrix driving are presented.

A programmable hysteresis comparator with tunable asymmetry for blood pressure peak detection is proposed. The hysteresis can be controlled by two programmable width transistors independently. The total current for the designed comparator is approximately 3.4 ¿A with a 3.0 V supply. The hysteresis change does not affect the comparator's power consumption. The measured results show that with the flexible programming of the hysteresis, the peak detection circuit has good immunity for either 50 Hz power line noise or low frequency motion artifacts, and can be suitable for most blood pressure peak extraction cases. The design can be easily extended to the low-voltage low-power pulse counting and peak detection circuitries in other applications.

Prognostic techniques attempt to predict the Remaining Useful Life (RUL) of a subsystem or a component. Such techniques often use sensor data which are periodically measured and recorded into a time series data set. Such multivariate data sets form complex and non-linear inter-dependencies through recorded time steps and between sensors. Many current existing algorithms for prognostic purposes starts to explore Deep Neural Network (DNN) and its effectiveness in the field. Although Deep Learning (DL) techniques outperform the traditional prognostic algorithms, the networks are generally complex to deploy or train. This paper proposes a Multi-variable Time Series (MTS) focused approach to prognostics that implements a lightweight Convolutional Neural Network (CNN) with attention mechanism. The convolution filters work to extract the abstract temporal patterns from the multiple time series, while the attention mechanisms review the information across the time axis and select the relevant information. The results suggest that the proposed method not only produces a superior accuracy of RUL estimation but it also trains many folds faster than the reported works. The superiority of deploying the network is also demonstrated on a lightweight hardware platform by not just being much compact, but also more efficient for the resource restricted environment.