Considering that in Inverse Synthetic Aperture Radar (ISAR) imaging of high-speed moving targets, the phase compensation method based on traditional Linear Frequency Modulated (LFM) parameter estimation has such problems as complex calculation and large errors, a new approach based on Convolutional Neural Network (CNN) is proposed to estimate the frequency modulation slope of chirp signals.First, Wigner-Ville distribution time-frequency analysis method is used to generate time-frequency image training set for LFM signal with a certain range of frequency modulation slope.Secondly, the time-frequency image of the high-speed moving target echo signal processed by Wigner-Ville distribution is input into the CNN to identify the frequency modulation slope of the echo signal.Then, the target velocity is inversed by the identified FM slope, the compensation signal is constructed and phase compensation is made to the echo signal.Finally, clear ISAR image is obtained through the processing of Range-Doppler imaging algorithm.Simulation results show the effectiveness of the proposed method.