Human Similar Activity Recognition Using Millimeter-Wave Radar Based on CNN-BiLSTM and Class Activation Mapping

Human activity recognition (HAR) is an important field in the application of millimeter-wave radar. Radar-based HARs typically use Doppler signatures as primary data. However, some common similar human activities exhibit similar features that are difficult to distinguish. Therefore, the identification of similar activities is a great challenge. Given this problem, a recognition method based on convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM), and class activation mapping (CAM) is proposed in this paper. The spectrogram is formed by processing the radar echo signal. The high-dimensional features are extracted by CNN, and then the corresponding feature vectors are fed into the BiLSTM to obtain the recognition results. Finally, the class activation mapping is used to visualize the decision recognition process of the model. Based on the data of four similar activities of different people collected by mm-wave radar, the experimental results show that the recognition accuracy of the proposed model reached 94.63%. Additionally, the output results of this model have strong robustness and generalization ability. It provides a new way to improve the accuracy of human similar posture recognition.