Effective feature extraction algorithm for spotlight synthetic aperture radar images

An effective algorithm of feature extraction for spotlight synthetic aperture radar images is presented. The signal noise ratio of the image is improved by denoising using wavelet transform, and edge detection is performed by canny operator. According to the characteristics of radar image, a method of image segmentation is suggested by performing threshold processing directly after edge detection instead of close curves. The Hu moments, which are rotation, scale and translation invariant, are extracted as feature vector and normalized after image preprocessing as mentioned above, and clustering analysis is applied in the training phase. The recognition capability of this feature extraction algorithm is tested with the MSTAR experimental data using both the nearest neighbor classifier and the back propagation neural network classifier, and the effectivity of this algorithm is validated.