Multi-Channel Fusion Human Activity Recognition Algorithm Based on Millimeter-Wave Radar

Traditional datasets for human activity recognition (HAR) are typically obtained from single-channel radar, which does not effectively utilize the target feature information. This paper introduces a multi-channel fusion HAR algorithm based on millimeter-wave radar. The algorithm leverages the multichannel data acquisition capability of multi-input multi-output (MIMO) radar and employs three independent MobileNetV3 lightweight network models for classification and recognition. Subsequently, a hard classification algorithm is utilized to fuse the multi-channel information, enabling the identification of human behaviors. Experimental results indicate that the proposed algorithm achieves an excellent classification performance of 97.05% on the experimental dataset. Moreover, it demonstrates a marked improvement in accuracy and stability compared to traditional single-channel classification methods. Furthermore, when compared with commonly used classification networks for human behavior, MobileNetV3 used in the algorithm highlights its ability to achieve commendable classification performance with minimal computational cost. This validates the proposed method as a viable lightweight hardware-transplantation approach.