The object recognition concept is being widely used a result of increasing CCTV surveillance and the need for automatic object or activity detection from images or video. Increases in the use of various sensor networks have also raised the need of lightweight process frameworks. Much research has been carried out in this area, but the research scope is colossal as it deals with open-ended problems such as being able to achieve high accuracy in little time using lightweight process frameworks. Convolution Neural Networks and their variants are widely used in various computer vision activities, but most of the architectures of CNN are application-specific. There is always a need for generic architectures with better performance. This paper introduces the Dimension-Based Generic Convolution Block (DBGC), which can be used with any CNN to make the architecture generic and provide a dimension-wise selection of various height, width, and depth kernels. This single unit which uses the separable convolution concept provides multiple combinations using various dimension-based kernels. This single unit can be used for height-based, width-based, or depth-based dimensions; the same unit can even be used for height and width, width and depth, and depth and height dimensions. It can also be used for combinations involving all three dimensions of height, width, and depth. The main novelty of DBGC lies in the dimension selector block included in the proposed architecture. Proposed unoptimized kernel dimensions reduce FLOPs by around one third and also reduce the accuracy by around one half; semi-optimized kernel dimensions yield almost the same or higher accuracy with half the FLOPs of the original architecture, while optimized kernel dimensions provide 5 to 6% higher accuracy with around a 10 M reduction in FLOPs.

This article presents a location-invariant air pollution prediction model with good geographic generalizability. The model uses a light GBR as part of a machine-learning framework to capture the spatial identification of air contaminants. Given the dynamic nature of air pollution, the model also uses a random forest to capture temporal dependencies in the data. Our model uses a transfer learning strategy to deal with location variability. The algorithm can learn concentration patterns because it has been trained on a vast data set of air quality measurements from various locations. The trained model is then improved using information from a particular target site, customizing it to the features of the target area. Experiments are carried out on a comprehensive data set containing air pollution measurements from various places to assess the efficacy of the proposed model. The recommended method performs better than standard models at forecasting air pollution levels, proving its dependability in various geographical settings. An interpretability analysis is also performed to learn about the variables affecting air pollution levels. We identify the geographical patterns associated with high-pollutant concentrations by visualizing the learned representations within the model, giving important information for environmental planning and mitigation methods. The observations show that the model outperforms state-of-the-art forecasting based on recurrent neural network and transformer-based models. The suggested methodology for forecasting air contaminants has the potential to improve air quality management and aid in decision-making across numerous regions. This helps safeguard the environment and public health by creating more precise and dependable air pollution forecast systems.

In this paper, we elaborate the design, implementation, and testing of a system – FallSense, which is able to detect accidental falls of human beings in a precise way. The distinguishing feature of FallSense is its endeavors beyond the scope of accelerometer, which is a component of traditional Body Sensor Network (BSN). Along with this acceleration measuring unit, FallSense uses the benefits of an Internet-of-Things (IoT)-enabled environment, which consists of a number of infrared transmitter-receiver pairs, and ultrasonic sensors. Employing a Fuzzy Inference System (FIS), FallSense fuses data from multiple sensors and becomes over-sure before inferring that a fall has occurred. Depending on the inputs from multiple sensors, FallSense generate a value between 0 and 1, which signifies the chance of fall. Results show that multi-sensor based FallSense achieves overall 16% improvement in comparison with the existing approaches, on an average. Beyond the theoretical modeling, this work also practically implemented the same with the help of real sensing units.

This article presents a bidirectional long short-term memory (LSTM) model for the detection of landslides. Previous uses of machine learning (ML) in this setting have demonstrated its general potential, which necessitates the implementation of a suitable algorithm. Landslides are natural disasters that can cause significant destruction and disruption in the affected areas. Early detection is the key to minimizing the impact of landslides, so it is important to develop accurate and efficient models. An area selected for this study is located in Mawiongrim, Meghalaya, India, which is an active landslide zone. The proposed model uses a bidirectional LSTM to capture the temporal patterns of the input data collected from a long-term real-time monitoring system set up in the area. To evaluate the effectiveness of the predictions, the model is trained using a data set composed of various landslide-related characteristics, such as topography, rainfall, hydrological, and soil properties. The results show that the suggested model is capable of detecting landslides with greater accuracy and the lowest error value relative to other models. Additionally, the model is also able to provide a real-time warning system, making it a viable tool for early landslide detection. The research also highlights the prediction models for matric suction and groundwater level, which are crucial in determining slope stability.

Human movement is a significant factor in extensive spatial-transmission models of contagious viruses. The proposed COUNTERACT system recognizes infectious sites by retrieving location data from a mobile phone device linked with a particular infected subject. The proposed approach is computing an incubation phase for the subject's infection, backpropagation through the subjects’ location data to investigate a location where the subject has been during the incubation period. Classifying to each such site as a contagious site, informing exposed suspects who have been to the contagious location, and seeking near real-time or real-time feedback from suspects to affirm, discard, or improve the recognition of the infectious site. This technique is based on the contraption to gather confirmed infected subject and possibly carrier suspect area location, correlating location for the incubation days. Security and privacy are a specific thing in the present research, and the system is used only through authentication and authorization. The proposed approach is for healthcare officials primarily. It is different from other existing systems where all the subjects have to install the application. The cell phone associated with the global positioning system (GPS) location data is collected from the COVID-19 subjects.