A radar micro-doppler target recognition method based on deep learning

Abstract

The application discloses a radar micro-Doppler target recognition method based on deep learning, and belongs to the field of radar target recognition. The method uses the collected target Doppler data, constructs an N*M two-dimensional array after pulse compression and cyclic shift, and generates a data set with an amplitude of 0~255; the data set is trained by using a ResNet18 model on an embedded server platform, parameters are adjusted until the recognition probability of the verification set reaches the standard; the model is converted into a format suitable for embedding NPU and deployed, real-time data inference is called through API, and the target type corresponding to the maximum probability that reaches the standard is taken. The specific implementation process of the method is described by using a Ku wave band FMCW radar data set. The application solves the problems of long traditional STFT residence time and information loss of artificial features, improves the recognition efficiency and accuracy, and is suitable for realizing real-time target classification in a radar system containing NPU.

Images