Abstract: With the development of artificial intelligence technology, deep learning methods have achieved good classification results in the field of low signal-to-noise ratio Low Probability of Intercept Radar(LPI) signal recognition. However, there is relatively little research on the recognition of multipath fading signals. This article studies the identification method of multipath fading LPI signals and proposes a low signal-to-noise ratio multipath fading LPI signal recognition method based on Next-MetaFormer(Next-MF). Using Choi-Williams distribution(CWD) to generate a two-dimensional time-frequency map of radar signals, and using the Neighbor2Neighbor self-supervised method to remove background noise in the time-frequency map, it has a good denoising effect. Using the Next Generation Vision Transformer(Next-ViT) universal architecture Next-MF network model to extract signal time-frequency features and achieve classification and recognition. The simulation results show that this method has high recognition accuracy under low signal-to-noise ratio and Rayleigh multipath fading interference conditions, with an overall classification accuracy of 82.5 % for 13 types of low intercept probability radar signals.The research results can provide reference for theoretical research and engineering implementation of radar signal recognition system.