The Aerial Target Type Recognition Based on Box-Cox Transform and Convolutional Neural Network

Abstract

Aiming at the weak micro characteristics of traditional narrowband radar targets, an aerial target type recognition method based on Box-Cox transform and convolutional neural network is proposed. The method does not need to compensate the fuselage component, and carries out Box-Cox nonlinear transformation directly to the radar echo data, which enhances the characteristics of micro component. Then, two-dimensional time-frequency images are generated by short-time Fourier transform, which are input to convolutional neural network for feature learning, and helicopter, propeller aircraft, jet aircraft type recognition is completed. Finally, a comparative experiment was carried out on the recognition effect under three influencing factors of SNR, observation time and pulse repetition frequency. The experimental results show that the Box-Cox transform can effectively improve the recognition rate under the condition of high SNR.