Frequency-hopping signal shows good performance in anti-interference. Accurately identifying the modulation methods of frequency-hopping signals can provide strong support for military information warfare such as judging the attributes of enemy and enemy targets and interfering with enemy signals. Nevertheless, there is still a big gap in the modulation recognition of frequency hopping signals at home and abroad. A frequency-hopping signal modulation recognition method based on time-frequency features is proposed. Through Smoothed Pseudo Wigner-Ville Distribution(SPWVD) time-frequency transformation, time-frequency images of frequency-hopping signals of different modulation types are obtained, and the time-frequency images are sent to a Convolutional Neural Network(CNN) for feature extraction and classification recognition. Simulation experiments prove that the proposed CNN model has achieved better recognition results under low Signal-to-Noise Ratios(SNRs).