SAR Target Recognition Based on Convolutional Feature Aggregation and Decision Combination

Synthetic aperture radar (SAR) target recognition based on deep convolutional neural networks (CNN) has achieved great success. Whereas, CNN usually needs massive labeled data to learn. In some cases, it may be difficult to collect a large number of labeled data, especially in SAR target recognition field. Thus, we develop a new method called SAR target recognition based on convolutional feature aggregation and decision combination (FADC) to improve the classification accuracy when labeled SAR data is limited. In FADC, we propose to concatenate the feature maps of different convolutional layers to extract discriminative feature. Then, the first-order statistical features of different layers are used to train extra classifiers. We can obtain two pieces of soft classification results yielded by softmax layer and extra classifiers for a query SAR target image. These soft classification results are combined by weighted arithmetic average rule whose weights are learnt by minimizing the mean squared error between fusion results and ground truth on labeled SAR target images. FADC was tested on MSTAR dataset, and the experiment results demonstrate that it can effectively improve the classification accuracy compared with a variety of advanced methods.