To correctly identify specific signals in ultra short wave communication, an approach is proposed by using the timefrequency spectrogram and the convolution neural network for the ultra short wave specific signal recognition, which transforms the classification of specific signals into image recognition. The time frequency spectrogram of specific signals are obtained by using the shorttime Fourier transform. Then the time frequency spectrogram is used to train the modified convolution neural network model. Finally the network model is tested to realize the ultrashort wave specific signal recognition. Simulation results show that the recognition rate of the proposed approach can reach 98% for a specific signal, the recognition rate can reach 97% when the signal-to-noise ratio (SNR) is 0 dB, and the recognition rate can reach 90% when the aliasing interference is 50%. Compared with traditional algorithms, the proposed approach has a better ability in low SNR and aliasing interference, which verifies the effectiveness of the convolution neural network in the field of specific signal recognition, and lays a foundation for subsequent research in this field.