Spectral-Spatial Hyperspectral Image Classification via Adaptive Total Variation Filtering

Abstract

It is unavoidable that existing noise interference in hyperspectral image (HSI). In order to reduce the noise in HSI and obtain a higher classification result, a spectral-spatial HSI classification via adaptive total variation filtering (ATVF) is proposed in this paper, which consists of the following steps: first, the principal component analysis (PCA) method is used for dimension reduction of HSI. Then, the adaptive total variation filtering is performed on the principal components so as to reduce the sensitiveness of noise and obtain a coarse contour feature. Next, the ensemble empirical mode decomposition is used to decompose each spectrum band into serial components, the characteristics of HSI can be further integrated in a transform domain. Finally, a pixel-level classifier (such as SVM) is used for classification of the processed image. The paper analyzes the effect of different parameters of ATVF method on the classification performance in detail, tests the proposed algorithm on the real hyperspectral data sets, and finally verifies the superiority of the proposed algorithm based on a contrastive analysis of different algorithms.

This work was supported the National Natural Science Foundation of China under Grant 51704115, by the Key Laboratory Open Fund Project of Hunan Province University under Grant 17K040 and 15K051, and by the Science and Technology Program of Hunan Province under Grant 2016TP1021.