Hyperspectral Image Classification Based on Different Affinity Metrics

With the development of hyperspectral sensor technologies, hyperspectral image classification has been a popular area in recent years. In this paper, we adopt different metric models: Euclidean distance and Spectral-spatial distance to learn the similarity of hyperspectral image (HSI) pixels. Then, we combine them with the smooth ordering model, which has been proposed in image processing to extract features of HSI. Finally, we utilize interpolation technology to create a decision function, which is to construct ultima classifier for the whole HSI pixels. The experiments demonstrate that these two metric combining multi-lDMEs can improve accuracy of HSI classification.