Deep Kronecker LeNet for human motion classification with feature extraction

Abstract

Human motion classification is gaining more interest among researchers, and it is significant in various applications. Human motion classification and assessment play a significant role in health science and security. Technology-based human motion evaluation deploys motion sensors and infrared cameras for capturing essential portions of human motion and key facial elements. Nevertheless, the prime concern is providing effectual monitoring sensors amidst several stages with less privacy. To overcome this issue, we have developed a human motion categorization system called Deep Kronecker LeNet (DKLeNet), which uses a hybrid network.The system design of impulse radio Ultra-Wide Band (IR-UWB) through-wall radar (TWR) is devised, and the UWB radar acquires the signal. The acquired signal is passed through the gridding phase, and then the feature extraction unit is executed. A new module DKLeNet, which is tuned by Spotted Grey Wolf Optimizer (SGWO), wherein the layers of these networks are modified by applying the Fuzzy concept. In this model, the enhanced technique DKLeNet is unified by Deep Kronecker Network (DKN) and LeNet as well as the optimization modules SGWO is devised by Spotted Hyena Optimizer (SHO) and Grey Wolf Optimizer (GWO). The classified output of human motion is based on human walking, standing still, and empty. The analytic measures of DKLeNet_SGWO are Accuracy, True positive rate (TPR), True Negative rate (TNR), and Mean squared error (MSE) observed as 95.8%, 95.0%, 95.2%, and 38.5%, as well as the computational time observed less value in both training and testing data when compared to other modules with 4.099 min and 3.012 s.