Spatial Co-occurrence Kernel Based Aerial Image Classification

We propose to use fuzzy relation based co-occurrence kernel for classification of high-resolution aerial images. By analyzing the characteristics of aerial images, it points out that the imagery does not have an absolute reference frame. For this reason, it uses a local descriptor called MROGH which is inherently rotation invariant to extract low-level features of aerial images. It then uses fuzzy relation based spatial co-occurrence kernel to build the holistic representation of aerial images. Experiments results on publicly available aerial scene imagery dataset show that our method gets a better classification result. In addition, we make a consistent comparative analysis of different classification frameworks based on aerial image dataset.