Synthetic Aperture Radar Images Target Detection and Recognition with Multiscale Feature Extraction and Fusion Based on Convolutional Neural Networks

In order to improve the precision of target detection and recognition for synthetic aperture radar (SAR) images, in this paper, we proposed the multiscale feature extraction and fusion method for SAR images based on the convolutional neural networks. We constructed training and testing data based on the MSTAR dataset. Since there are not enough SAR image data, we used image processing methods to do the data augmentation. In order to improve the accuracy of target detection, we also used the method of transfer learning. Eventually we trained and tested the model on a small data set, the final mAP reached 96.58%, a relatively high score which proved the effectiveness of multiscale feature extraction and fusion. In order to better understand the principle of this technology, we also did some visualization analysis for the feature maps. This proved the reliability of the method.