UNSUPERVISED FEATURE LEARNING FOR SCENE CLASSIFICATION OF HIGH RESOLUTION REMOTE SENSING IMAGE

Due to the rapid development of various satellite sensors, a large amount of high resolution remote sensing images can be obtained. In order to efficiently represent the scenes from these high resolution images, an unsupervised feature learning method is proposed for high resolution image scene classification. In the proposed method, a set of filter banks are learned in an unsupervised manner from the unlabeled image patches, which are robust, efficient and do not need elaborately designed descriptors such as SIFT. And then, each image is encoded by these filter banks using a soft distance assignment scheme, generating a final feature vector to excellently represent the image scene. Finally, by virtue of the traditional SVM classifier, the sematic concepts of different scenes can be categorized. Experimental evaluation on the the high resolution remote sensing images demonstrates the effectiveness and good performance of the proposed method.