An Optimization Method for Hyperspectral Endmember Extraction Based on K-SVD

• DOI: 10.14358/pers.85.12.879
• Publication Date: 2019-12-18T04:57:27Z
• Authors: Xiaoxiao Feng; Luxiao He; Ya Zhang; Yun Tang
• Keywords: Endmember

Mixed pixels are common in hyperspectral imagery (<small>HSI</small>). Due to the complexity of ground object distribution, some end-member extraction methods cannot obtain good results and processes complex. Therefore, this paper proposes an optimization method for <small>HSI</small> endmember extraction, which improves accuracy based on K-singular value decomposition (<small>K-SVD</small>). The proposed comprises three core steps. (1) Based contribution initial endmembers, partially observed data selected according appropriate confidence participate calculation. (2) Construction error model eliminate background noise. (3) Using <small>K-SVD</small> perform column-by-column iteration endmembers achieve overall optimality. Experiments with real images applied, demonstrating can improve by 15.1%–55.7% compared original methods.