Aircraft discrimination in high resolution SAR images based on texture analysis

Target discrimination is the key step of automatic target detection in synthetic aperture radar (SAR) images. Aiming at the issue of aircraft discrimination in high resolution SAR images, a novel discrimination method is proposed with using texture features. First of all the method of gray level co-occurrence matrix is used to generate eight discrimination texture features: mean, variance, deficit moment, inertia moment, entropy, angular second moment, relevance and non-similarity and then forming a feature vector. Differing with the common method of extracting the holistic texture features of image to represent the target, the texture features of each pixel are extracted and the feature vectors of all pixels are used to represent the target. Then J-M distance is used to measure the different targets, and supervised training method is applied to achieve the parameters of discrimination rule. Finally, suspected targets are discriminated to different classes by the trained discrimination rule and large numbers of false alarms are eliminated efficiently. The experiments show that the aircraft has small distance to other aircrafts while large difference to false alarms, so this discrimination method has high accuracy with excellent applicability.