Superpixel-based composite kernel for hyperspectral image classification

摘要: We propose a superpixel-based composite kernel framework for hyperspectral image (HSI) classification. Composite methods can utilize both the spectral and spatial information HSI However, setting optimal neighborhood different structures is non-trivial issue. In order to adaptively exploit contextual information, we superpixel obtain information. A be regarded as local neighborhood, whose size shape adjusted according in HSI. Then, features are extracted by computing mean of pixels within each superpixel. Finally, with support vector machine implemented on real Experiments two HSIs demonstrate outstanding performance proposed method.