Human Activity Classification with Multi-frequency Spectrogram Fusion and Deep Learning

In this paper, we propose an approach that uses deep convolutional neural network (DCNN) based on multi-frequency spectrogram fusion to classify the human activity with respect to stepped frequency continuous wave (SFCW) radar. The main principle of the fused spectrogram is to adopt the guide image filter (GIF) to decompose all spectrograms into a noise-free base layer and a detail layer containing high-frequency detail information and noise. The weighted averaging (WA) method is used to fuse the body information contained in the base layer. The detail layer is denoised by the adaptive enhancement and denoising technology and then adopt direct superposition to keep the details in the fused detail layer. Finally, both of them are recombined to obtain the fused spectrogram. The fused spectrogram as an input to DCNN network achieves the classification. The algorithm based on spectrogram fusion reduces complexity of DCNN construction, and simultaneously improves classification accuracy.