A Target Recognition Method for Small Sample SAR Images Based on Heterogeneous Feature Fusion

Small sample SAR image recognition is a difficult problem in SAR image target recognition. This paper presents a SAR image recognition method based on heterogeneous feature fusion. The proposed method first uses convolutional neural network to extract similar features of SAR images from the same category, Then, a feature difference measurement method is designed to calculate the feature differences between different categories of SAR images. We call this feature difference as heterogeneous features. We fuse the similar features and heterogeneous features to achieve the feature extraction and recognition of the target. Therefore, the network can learn the same characteristics of one class of targets and the different characteristics of other classes of targets at the same time. Thus, the proposed method can reduce the demand of the network for the amount of training data and improve the accuracy of multi class target recognition under the condition of small sample. Moreover, the difference feature extraction branch can also play the effect of data amplification. The more categories, the better the effect of data amplification. Experiments on MSTAR datasets show that the overall detection accuracy and convergence speed of the proposed method are significantly improved.