DISTORTED SAR TARGET RECOGNITION WITH VIRTUAL SVM AND AP-HOG FEATURE

Synthetic aperture radar (SAR) plays an indispensable role in national defence and civil fields due to its unique advantages. Although many methods have been proposed, the recognition to SAR images still faces many difficulties, especially to distorted images. This paper first improves the morphological methods attribute profiles (APs) which is widely employed with optical, and hyperspectral images in particular. We combine the APs method with histogram of oriented gradients (HOG) feature (called AP-HOG) to make it suitable for SAR images. Then, the SVM model is used to identify the labelled samples (i.e., Support Vectors) that are closest to the separating hyperplane with maximum margin. We generate different degrees of distorted images based on support vectors, and use the distorted image and the support vector as a new training set for retraining to improve the recognition effect of the distorted image. Finally, we perform experimental verification based on the MSTAR dataset, which proves the effectiveness of the proposed method.