Human Activity Recognition Method Based on Scattering Separation Using Multifrequency Radar Data

The human body displays typical properties of multiple sca-ttering when targeted by radar, and the strong echo signal scattered from the torso often masks the weak echo signal scattered from the other body parts such as limbs and the head, limiting the performance of activity recognition. To address this issue, a human activity recognition method based on scattering separation using multifrequency radar data is proposed. First, the multifrequency echo data of human activity collected from stepped-frequency continuous wave radar is stacked, followed by principal component analysis to separate the trunk signal and the branch signal into the first two components, which helps in avoiding the interference caused by masking effects. Subsequently, the two time-frequency spectrograms that express the characteristics of human activities jointly are put into two parallel convolutional neural networks to complete feature extraction and classification. Datasets encompassing six activities were gathered using the stepped-frequency radar. Test results demonstrate that this method can enhance the average recognition effectually compared to the approach without scattering separation.