In complex electromagnetic environments,the image transmission signal of non-cooperative UAV is prone to be covered by interference,and traditional detection methods cannot effectively identify the signal,so a UAV image transmission signal recognition method based on the improved AlexNet model is proposed.According to the time-frequency spectrum characteristics of the image transmission signal in interference coverage scenarios,this method improves and optimizes the AlexNet model and deepens the network structure without increasing computational complexity by splitting the convolution kernel,reducing the number of nodes in the fully-connected layer,and adding a global average pooling layer,which effectively improve the image transmission signal recognition capability.In indoor anechoic chamber and real-world field environments,the time-frequency image datasets under different interference intensities are prepared to train the model.The results show that,when the Signal to Interference plus Noise Ratio (SINR) is -15 dB, the improved AlexNet model can still maintain a verification accuracy above 90%,and the unit training time can be shortened by more than one second compared with other CNN models.