A Remote Sensing Target Recognition Method based on Generative Adversarial Networks

Abstract

The traditional target detection algorithm uses the method of multi feature combination for recognition, which has the problems of poor anti-noise ability, weak robustness and long time consuming. In order to solve these problems, we first design a conditional generation adversarial network (GAN), and then use PCA algorithm to reduce the dimension of the feature vector, so as to retain the most distinguishing features in the sample and complete the feature extraction of remote sensing target. This paper also proposes a recognition algorithm based on template matching, which matches the processed image segmentation results with the standard template to get the type of object. Finally, we test the proposed method in a remote sensing aircraft image data set, and the accuracy reaches 95.86%, which is higher than other comparison algorithms.