Phase-Based Target Classification Using Neural Network in Automotive Radar Systems

In this paper, phase-based target classification using a neural network in automotive radar systems is proposed. The signals received at the array antenna contain amplitude and phase information whose values vary depending on the type of the target. In this study, the phase information of the signals reflected from pedestrians and vehicles are mainly considered as a classification criterion. Based on the phases of received signals, the phase difference is calculated by subtracting the phases of the signals received at each antenna element to adjacent antenna element. After extracting the phase and phase difference, the two different data sets are used as inputs to distinguish between the pedestrian and vehicle. Moreover, a deep neural network is used for the classification. After determining the proper number of nodes in a hidden layer and the number of hidden layers for the network, the performance of the proposed classification is evaluated. From our measured data, the proposed method with deep neural network by setting the number of nodes and hidden layers as 25 and 1 can classify the pedestrian and vehicle with more than 90% accuracy, regardless of which of the two input types is used.