Research On Intelligent Recognition Method Of Human Target Motion Exploiting Ultra-Wideband Radar

Ultra-wideband(UWB)radar human motion recognition is a research hotspot in the field of pattern recognition in recent years,and has important application value in disaster rescue,human-computer interaction,stability maintenance,patient monitoring and so on.Human motion recognition model based on deep learning can effectively solve the problem of insufficient or mismatched feature extraction in traditional and machine learning based motion recognition algorithms.Therefore,recognition models based on deep learning have become the mainstream in the field of UWB radar human motion recognition.At present,UWB radar human target motion recognition based on deep learning mainly faces problems such as low model recognition accuracy,too large parameters,too few samples,poor real-time performance,and limited viewing angle.Aiming at the above problems,this thesis conducts research on methods such as parameter optimization,lightweight network design,transfer learning,and multi-radar collaborative real-time recognition of UWB radar non-direct-looking human target recognition model based on deep learning.The main work and innovations are as follows:1.Aiming at the problem of low recognition accuracy,a human motion recognition method based on Reg Net network parameter optimization is proposed.The model combines the advantages of semi-automatic design network and neural network architecture search,which significantly improves the recognition accuracy of human motion,and provides certain guiding significance for the understanding and design analysis of human motion recognition model architecture.2.Aiming at the problem of too large model parameters,a human motion recognition method based on the attention mechanism of Mobile Net lightweight network space-time fusion is proposed.The method utilizes the multi-scale and multi-channel information enhancement ability of the attention mechanism to the behavior samples and the lightweight characteristics of the Mobile Net network to achieve efficient and accurate recognition of single-view human target motion,and provides a theoretical basis for the model to be embedded in portable devices.3.Aiming at the problem of small samples and the limitation of radar viewing angle,a multi-radar collaborative human motion recognition method based on heterogeneous transfer learning is proposed.The model uses the public image data set to pre-train the human motion recognition model,and retrains the model with the radar data after knowledge transfer,which solves the problem of model overfitting due to too few samples and the problem of degradation of recognition model performance due to viewing angle limitations.4.Aiming at the problems of poor real-time performance and insufficient utilization of information,a multi-view real-time recognition method based on the combination of bidirectional recurrent neural network and Stacking ensemble learning model is proposed.The model method the ability of bidirectional recurrent neural network to correlate historical and future information and the multi-radar data fusion strategy of Stacking ensemble learning,which solves the problems of poor real-time human motion recognition and insufficient information utilization.The proposed method performs well in terms of accuracy,maximum delay and angle sensitivity.The above-mentioned UWB radar human motion recognition methods are tested and verified by the measured data.The results show that the methods proposed in this thesis can effectively improve the recognition performance of the network,and can be applied to ultra-wideband intelligent radar systems.