Hyperspectral image classification based on extrema morphological profiles and total variation

Abstract Hyperspectral image (HSI) contains spectral information and spatial information of a scene. Because of the unique spectral characteristics of materials, HSI has been proved to be particularly useful for distinguishing the sort of materials in the scene. In this paper, a HSI classification method is proposed. Firstly, we propose a new extrema morphological profiles (EMPs) to extract the spatial feature in HSI. Secondly, total variation (TV) is used to integrate the spatial feature into a fused feature map. Finally, support vector machine (SVM) is utilized to accurately classify the fused feature maps. In order to evaluate the proposed method, we take quantitative classification analyses on a HSI dataset, class accuracy (CA), overall accuracy (OA), average accuracy (AA), and Kappa coefficient are adopted as the metrics. The experimental results indicate that the proposed method can efficiently achieve high classification accuracy.