A Two-Staged Feature Extraction Method Based on Total Variation for Hyperspectral Images

Abstract

Effective feature extraction (FE) has always been the focus of hyperspectral images (HSIs). For aerial remote-sensing HSIs processing and its land cover classification, in this article, an efficient two-staged hyperspectral FE method based on total variation (TV) is proposed. In the first stage, the average fusion method was used to reduce the spectral dimension. Then, the anisotropic TV model with different regularization parameters was utilized to obtain featured blocks of different smoothness, each containing multi-scale structure information, and we stacked them as the next stage’s input. In the second stage, equipped with singular value transformation to reduce the dimension again, we followed an isotropic TV model based on split Bregman algorithm for further detail smoothing. Finally, the feature-extracted block was fed to the support vector machine for classification experiments. The results, with three hyperspectral datasets, demonstrate that our proposed method can competitively outperform state-of-the-art methods in terms of its classification accuracy and computing time. Also, our proposed method delivers robustness and stability by comprehensive parameter analysis.

The Tensor Discriminant Ridge Regression Model With Extreme Learning Machine for Hyperspectral Image Classification

“…Let (C×J) . W (1) ∈ R C×h can give each band an appropriate weight. W (2) ∈ R 1×l , W (3) ∈ R 1×ω can give each pixel vector an appropriate weight, that is similar as AWF [34].…”

“…The second case is to optimize the spectral information. that is how do we optimize W (1) while keeping E (1) constant.…”

“…Secondly, the second case is discussed, and the 1 th pattern weight W (1) ∈ R C×h need to be solved. Denote all the t th class samples of E (1) ∈ R h×J as F…”

“…is the mean matrix of E (1) . Let us denote a ij as an entry of row i and column j of W (1) F (1) ∈ R C×I ; then a ij is the projection of the j th sample of F (1) onto the i th row vector of W (1) .…”

“…H YPERSPECTRAL images (HSIs) provide rich spatial and spectral information, allowing for a detailed classification and recognition of land cover objects. However, the following characteristics of HSIs bring difficulties to their application [1]. Firstly, because of the characteristics of the This work is supported in part by the National Nature Science Foundation of China under Grant U1701266, Grant 61372173, and Grant 61671163; in part by the Team Project of the Education Ministry of the Guangdong Province under Grant 2017KCXTD011; the Guangdong Higher Education Engineering Technology Research Center for Big Data on Manufacturing Knowledge Patent under Grant 501130144; the Guangdong Province Intellectual Property Key Laboratory Project under Grant 2018B030322016; the Hong Kong Innovation and Technology Commission, Enterprise Support Scheme, under Grant S/E/070/17; the Guangdong University Ideological and Political Education Project under Grant 2019GXSZ013; Natural Science Foundation of Guangdong Province (2019A1515012048); Special project in key fields of Guangdong Education Department(2020ZDZX1062); Guangdong Science and Technology Department project(KTP20210286); National Social Science Fund project(AJA190013); Ministry of Education planning Fund project(20YJA880058); Guangdong Education Department project(KTP20210286); Special project in key fields of Guangdong Education Department(2020ZDZX1062); National Natural Science Foundation of China (62072122); and Education Department of Guangdong Province (2019KSYS009).…”