With the characteristics of typical LPI radar signals and the requirements for modulation pattern recognition,a method of modulation recognition for typical LPI radar signal based on Gaussian smoothing ambiguity function and sDAE_LIBSVM is proposed.Firstly,the ambiguity function transformation combined with Gaussian smoothing is adopted to complete the construction of feature images.Secondly,a recognition network is built with the fusing of stack Denoising AutoEncoder (sDAE) and LIBSVM,which is used for the classification and recognition of feature images.The simulation results show that when SNR is -7 dB,the Probability of Successful Recognition(PSR) of the proposed method can achieve 97% for eight typical modulation patterns of LPI radars,including BPSK,Costas,Frank,LFM and T1~T4,and it has strong stability and robustness.Compared with other methods,it has better recognition performance.