To solve the problem of radar jamming signal recognition in complex electromagnetic environment, from the perspective of optimizing the convolution neural network structure, this paper proposes a method to add a batch normalization layer and change the activation function to the convolution neural network structure LeNet-5. This method can accelerate the network convergence and improve the network learning efficiency. In this paper, the ship target model is first established, and the differences in time-frequency domain between noise amplitude modulation jamming, noise frequency modulation jamming, comb spectrum jamming and the real target echo signal without jamming are analyzed. Then the data sets are generated for the ship target model by using four kinds of signals. Finally, the automatic recognition of radar jamming is realized by the method proposed in this paper. The simulation results show that under the condition of full signal-to-noise ratio (SNR), the recognition accuracy of the proposed method for four signals reaches 98.1%, indicating that the proposed method has good stability and robustness.