Wireless modulation signal pattern recognition is widely used in military and civilian applications. This paper proposes a method based on deep learning for wireless modulation signal pattern recognition for many types in practical applications. In this paper, a 34-layer convolutional neural network is designed to identify 16 typical wireless modulated signals. The robustness of the recognition algorithm is improved by using various methods. The data is enhanced by means of interpolation and extraction, power normalization, Gaussian noise, etc. The feature information matrix is constructed by extracting the time domain, frequency domain and phase domain information of the signal by Hilbert transform. The Additive Margin Softmax [1] method is used to make the signal have larger inter-class spacing and smaller intra-class spacing. Experiments show that under the condition of 0dB signal-to-noise ratio, the recognition rate can reach more than 95%, which is better than the traditional machine learning algorithm.