基于毫米波雷达的行为检测研究OA

Targeting at the requirements of non-contact and high-precision behavior detection in medical care environments,a behavior de-tection system is proposed based on millimeter wave radar.Firstly,an experimental platform is built to collect data.Then,a feature extrac-tion method based on moving target indication and time-frequency analysis is used to suppress clutter information and extract micro-Doppler features.Due to its lightweight architecture and high efficiency in feature extraction,ResNet-18 is employed.Finally,a fusion net-work based on ResNet-18 and long short-term memory(LSTM)network is proposed to extract both time-frequency features and sequence features.Experimental results of behavior detection on the public dataset of University of Glasgow show that average detection accuracy of the proposed model are 93.4%,which are higher than the values of average detection accuracy of AlexNet model(90.0%),VGG-16 model(88.9%),ResNet-18 model(92.3%),LSTM model(80.5%)and 4-layer convolutional neural network's(86.0%).On self-built dataset,the proposed model achieves an accuracy of 94.2%,which is an improvement over the existing models.