Road Targets Recognition Based on Deep Learning and Micro-Doppler Features

The ability of automotive millimeter-wave (MMW) radar to identify the road targets has a direct impact on the performance of Advanced Driver Assist System(ADAS). There are obvious differences in the Micro-Doppler features of the radar echo of pedestrians, automobiles and motorcycles. In this paper, a recognition method based on deep learning is proposed to realize the targets identification and classification by training and learning the Micro-Doppler features of the road targets. In this method, the time-frequency spectrogram of moving target echoes of the millimeter-wave radar is used as the training sample, convolutional neural network (CNN) is adopted to extract features and reduce dimension, and the output layer of the CNN is extended and fed into a recurrent neural networks (RNN) for sequence learning since the Micro-Doppler features of moving targets have obvious temporal correlation. The output layer of the network uses softmax regression to identify specific road targets. The Recognition results of measured data show that the proposed method has higher recognition accuracy and better recognition ability of road moving targets, compared with other machine learning methods that only use CNN.