Detection of MMW Radar Target Based on Doppler Characteristics and Deep Learning

In recent years, unmanned technology has been continuously developed. millimeter - wave (MMW)radar has been widely used in driverless vehicles because of its performance characteristics. Target detection is also one of the hot issues studied by experts and scholars in the field of driverless driving. According to the target detection problem of millimeter - wave radar, a deep learning - based target detection method is proposed. It uses 77G HZ on - board millimeter - wave radar Spectro graph data to mark the target existence area and form a standard data set through data preprocessing. An improved model of Doppler image detection of RetinaNet radar was subsequently proposed. The model uses ResNet101 as a feature extraction network, uses group normalization (GN) as a normalization method, improves the network accuracy and convergence speed, introduces the attention mechanism in the feature extraction network, and enhances the feature expression capability of the model. The improved RetinaNet model improves the average accuracy of radar Doppler image detection by 7.2 % and 91.5%, which provides ideas for the development of radar target detection and unmanned driving technology, and has engineering application value.