关键词: 信息处理技术, 波段选择, 端元选择, 算法移植, 高光谱图像

Abstract: High dimensionality of hyperspectral imagery leads to the existing of much redundant information. This case brings difficulty to the analysis and processing of hyperspectral data. Band selection can remove the redundant information and so reduce the computational cost. Endmember selection is another important technique of hyperspectral imagery processing. Theoretically speaking, it is possible to transplant endmember selection algorithm to band selection. After the analysis of feasibility, three endmember selection algorithms are used in band selection, and the subsequent technique problems are also resolved by introduced automatic spatial partition, kernel based PCA (principal component analysis) and distance equivalent principle. Numerical experiments show the validation of this kind of algorithm transplantation.

Key words: information processing, band selection, endmember selection, algorithm transplantation, hyperspectral imagery