Aerial Scene Classification by Surf Feature Extarction Using Unsupervised Learning

The high-resolution satellite imagery which consisted of rich data allow us to directly model aerial scenes by understanding their spatial and structural patterns. Efficient representation and recognition of scenes from image data are challenging. For satellite image analysis, pixel and object based classification approaches are widely used but these approaches often exploit the high-fidelity image data only in a limited way. In this paper, we explore a supervised feature learning approach for aerial scene classification. This system follows some peculiar steps like Noise Removal, Feature Extraction (SURF), Feature Learning and Classification. SURF features are extracted from the input image and classification is done by Latent Dirichlet Allocation Algorithm. This technique can be applied to several challenging aerial scene data sets: ORNL-I data set consisting of 1-m spatial resolution satellite imagery, UCMERCED data set with sub-meter resolution, and ORNL-II data set for large-facility scene detection. The proposed method is highly effective in developing a detection system that can be used to automatically scan large-scale high-resolution satellite imagery for detecting large facilities such as a shopping mall.