The invention provides a radar signal intra-pulse modulation identification method, which comprises the following steps of performing Cohen type time-frequency distribution processing on an intercepted radar signal to obtain a time-frequency image, preprocessing the time-frequency image, namely adjusting the size and the amplitude, and then grouping and tagging to make a training set and a test set, designing a deep convolutional neural network, adjusting the size and the amplitude of a radar signal of an unknown modulation type, putting the radar signal into the trained deep convolutional neural network, and automatically judging the radar signal type by the network to complete identification. Compared with a kernel function in the Choi-Williams distribution, the time-frequency analysis kernel function provided by the invention has a better effect on inhibiting the cross terms of radar signals, and the signal robustness characteristic is more obvious. The classification network is pre-trained by using the convolutional de-noising auto-encoder, so that the information loss of signal energy caused by time-frequency image preprocessing can be avoided, the classification accuracy of the whole system is improved, and the method is simple to operate and is easier to implement.