Research on SAR ATR technology based on feature fusion and spiking neural network

Synthetic aperture radar (SAR) images often suffer from sample missing problems, which are a consequence of the high cost of imaging. However, deep learning methods heavily rely on large-scale, high-quality labelled data, while traditional feature-based classification requires manual classifier design. To resolve above limitations, we proposed a SAR target recognition method based on feature fusion and spiking neural network (FF-SNN). Initially, we extract features from SAR images and fuse these features with different strategies. Subsequently, transformed the fused SAR image features into spiking signals. Then inputting the spiking signals into a Leaky Integrate-and-Fire neuron model, and transmitting the spiking signal information in a time driven manner. To verify the effectiveness of the model, we compared the FF-SNN with several traditional classifiers and popular deep learning algorithms. The experimental results showed that FF-SNN can effectively improve the accuracy of aircraft target recognition on civilian aircraft dataset. Among them, the fusion effect of Gabor filter and HOG feature demonstrated the best performance, with an accuracy of 84.84% on SAR-AIRcraft.1.0. Then, using this fused feature to conduct the sample missing experiment. The experiment achieved positive results, confirming the robustness of FF-SNN.