SAR image segmentation using unsupervised spectral regression and Gabor filter bank

Segmentation of synthetic aperture radar (SAR) is a challenge topic in recent years. Many statistical and structural methods have been proposed for this goal. Some of them are based on clustering, such as the sparse spectral clustering and Nyström method. These methods suffer from the low speed and high computational complexity because of the use of the eigen-decomposition in their algorithm. In this paper, we proposed an unsupervised feature learning method in which the features of different areas of SAR images are extracted, and then they will be learned using an unsupervised manner and finally the learned features will be clustered. The proposed algorithm improved the accuracy compared with other methods and it also has a shorter run time.