The integration of Micro Electronic Mechanical Systems (MEMS) sensor technology in smartphones has greatly improved the capability for Human Activity Recognition (HAR). By utilizing Machine Learning (ML) techniques and data from these sensors, various human motion activities can be classified. This study performed experiments and compiled a large dataset of nine daily activities, including Laying Down, Stationary, Walking, Brisk Walking, Running, Stairs-Up, Stairs-Down, Squatting, and Cycling. Several ML models, such as Decision Tree Classifier, Random Forest Classifier, K Neighbors Classifier, Multinomial Logistic Regression, Gaussian Naive Bayes, and Support Vector Machine, were trained on sensor data collected from accelerometer, gyroscope, and magnetometer embedded in smartphones and wearable devices. The highest test accuracy of 95% was achieved using the random forest algorithm. Additionally, a custom-built Bidirectional Long-Short-Term Memory (Bi-LSTM) model, a type of Recurrent Neural Network (RNN), was proposed and yielded an improved test accuracy of 98.1%. This approach differs from traditional algorithmic-based human activity detection used in current wearable technologies, resulting in improved accuracy.

Recently, human activity recognition (HAR) has gained a lot of importance due to its wide range of applications in virtual reality, healthcare, surveillance, security, automated control systems, etc. Latest mobile phones have advanced computational capabilities along with several embedded MEMS sensors, which enable us to detect various physical activities unobtrusively. Incorporating personalized motion samples can improve the accuracy of motion detection by mobile devices or wearable devices that are tailored to the individual. Recent works have demonstrated that the use of machine learning and statistical techniques can detect human activities more accurately. In this paper, we have classified two different physical activities, viz., walking and brisk-walking using a deep neural network (DNN). The personalized data has been collected using multiple sensors of the mobile phone with the two kinds of physical activity. Using mobile phone sensors like accelerometer, gyroscope, magnetometer, etc, data has been collected, examined, and used for training and testing the DNN model. Out of multiple sensors on the phone, we have identified the sensors which are more appropriate for the given motion/activities. Finally, we have achieved a classification accuracy of 96.5%.

As problems like climate change, greenhouse emissions, etc. become common, Electric Vehicles (EVs) are now being looked at as an alternative. In order for EVs to succeed, numerous charging networks must be established in a user-friendly environment and the optimum charging solution must be selected. Wireless Power Transfer (WPT) System is an option that simplifies the battery charging process for EVs as it eliminates the need for annoying cables and ensures the user's safety. Inductive power transfer is currently the most popular and mature WPT technology, which uses magnetic fields for power transfer. In order to comply with transmission and safety criteria, Foreign Object Detection (FOD) is an essential module that has to be taken care of. Deep Learning (DL) has been successfully applied to various applications by using algorithms such as CNNs, You Only Look Once (YOLO), etc. In this work, a unique approach has been taken toward FOD by using DL, particularly the YOLO v7 algorithm. The pre-trained YOLO v7 algorithm has been further customized to detect (a) coins (b) screws and (c) paper clips as part of FOD and towards the end, Mean Average Precision of 97%, 93.1% precision, and 95.3% recall values have been achieved.

The burning of bricks affects air quality and human health, and the toxic pollutants released into the air seriously impact the surrounding areas. In order to curb the pollution caused by these brick kilns, it is very important to identify them. As part of this study, an analysis in the Delhi- NCR region is performed. An approach for mapping brick kiln locations by using various Machine Learning (ML) algorithms is presented in this work. The Sentinel-2 imagery with 10m resolution was used to create four classes, namely, built up, field, water, and brick kiln. It is possible to distinguish brick kilns from other classes by using the said dataset due to its unique spectral and geometric features. Several ML classifiers have been trained on the sentinel-2 dataset, including the Random Forest Classifier and four built-in classifiers of Google Earth Engine, such as Smile Random Forest, Smile CART (Classification and Regression Trees), Smile Gradient Tree Boost, and Smile Naive Bayes Classifier. In order to train the model based on different classifiers, 70% of data points are used for training and 30% for validation purposes. Among the machine learning classifiers, Smile Random Forest achieved the maximum accuracy of 96.4%. As a result of this study, an ML model has been proposed for the classification of brick kilns from Sentinel-2 imagery, and a comparative analysis has been done between the various ML models. The proposed ML model can classify brick kilns from Sentinel-2 images with an accuracy of 96.4%, precision of 0.88, and recall of 0.73. The functionality of the classification model lies in its utilization by local authorities and various pollution control bodies, so that proper action can be initiated. Hence contributing to a greener environment and battling serious issues like climate change and global warming.

Sleep apnea is a common sleep disorder characterized by frequent breathing interruptions during sleep, which can lead to a range of negative health consequences. The diagnosis and treatment of sleep apnea require accurate detection of these breathing events, which can be challenging to detect using traditional polysomnography methods. To address this challenge, various automated sleep apnea detection methods have been developed using machine learning techniques. In this research work, we propose the development of a power-efficient and effective Support Vector Machine (SVM) classifier for the detection of sleep apnea using Field Programmable Gate Array (FPGA) technology, which is known for its high processing speed and low power consumption. The SVM classifier is implemented on the Xilinx Nexys Artix-7 FPGA board. Our experimental results show that our proposed SVM classifier achieves high classification accuracy of 98% for testing data, with a low power consumption of 144 mW. This power consumption is significantly less than that of other SVM classifiers executed on a general-purpose processor.