Radar target classification method based on transfer learning

The invention relates to a radar target classification method based on transfer learning. According to the invention, a TL-Swinin-Transform network is provided, features are extracted from a channel dimension and a space dimension by using a convolution attention mechanism module, the features are fused into the Swinin-Transform module to improve the extraction capability of multi-head attention information in a small scale, regions useful for classification are more concerned, it is ensured that more local detail features of a time-frequency graph are learned before down-sampling, and the classification efficiency is improved. The time-frequency image features are better extracted; a transfer learning strategy is introduced, the weight of a model network is initialized by utilizing a model parameter file pre-trained on an ImageNet data set, the original random initialization operation is replaced, global fine tuning is carried out, other training processes are carried out as usual, and then inter-domain tran