Human Activity Recognition From Radar Signals Based on FBSE-EWT and Quantum Convolution Neural Network

Frequency-modulated continuous wave (FMCW) radar signals reflected from the human activities are analyzed to enable accurate recognition and classification, forming the basis for activity recognition applications. The analysis of micro- and macromotions of human activities plays a vital role in automatic detection of suspicious activity in smart home automation and clinical applications. The use of convolutional neural networks (CNNs) with time–frequency representations (TFRs) for human activity recognition (HAR) is limited due to their reliance on large datasets, poor generalization to complex radar signals, and large model size. To mitigate these limitations, a novel Fourier–Bessel series expansion (FBSE)-based empirical wavelet transform (FBSE-EWT) and quantum CNN (QCNN)-based approach is proposed. The FBSE-EWT method is utilized for the TFRs of a reflected multicomponent nonstationary signal with enhanced frequency resolution. The QCNN effectively classifies the human activities, enabling accurate and efficient HAR. The HAR from radar signals based on FBSE-EWT and QCNNs has not been studied earlier based on our knowledge. This is the first instance to utilize the proposed method for HAR using FMCW radar signals. We have also conducted extensive study of FBSE-EWT-based method with CNN approaches (such as GoogLeNet and VGG19) in the same framework. The proposed approach achieves 95.71% test accuracy compared to existing CNNs, demonstrating its superior performance for HAR.