Aerial image classification based on sparse representation and deep belief network

Due to the rapid development of satellite sensor technology, a rich aerial image data set can be easily acquired. How to efficiently classify and recognize the aerial image has become a critical task. In this paper, we propose an aerial image classification approach based on sparse representation and deep belief network (DBN). Low-level features are extracted by using scale-invariant feature transform (SIFT). These extracted features are encoded in terms of an improved sparse encoding mode by combining structural similarity and sparse coding to generate new sparse representation. DBN is used to express the relationship between low-level feature and high-level semantic representation and complete image classification. We apply our approach to UC Merced data set. The proposed approach obtained results that were equal to or even better than the previous results with the UC Merced data set.