结合局部空谱信息的高光谱图像多端元提取

Multiple Endmember Extraction with Local Spatial-Spectral Information for Hyperspectral Imagery

针对传统高光谱单端元提取算法不能描述光谱变异、混合像元分解精度不高的缺点,提出一种结合局部空谱信息的高光谱图像多端元提取(multiple endmember extraction algorithm with local spatial-spectral information,MEELSI)方法。首先将原始高光谱图像进行图像子空间划分获取不重叠的图像块,并利用自动目标生成算法分别在图像块上提取候选端元;然后对候选端元的邻域像元进行光谱相似性分析,优化精选候选端元;最后利用K-means聚类算法对所有端元集进行聚类分析,得到最终的多端元光谱集。仿真数据和真实高光谱数据的实验结果表明,与传统单端元提取方法相比较,MEELSI算法具有表征遥感图像中光谱变异的能力,能够有效提高混合像元分解精度。

Traditional hyperspectral endmember extraction technology ignores the fact that the endmember spectral variability exists in practical situations,which results in the low accuracy of the decomposition of mixed pixels.In this study,we proposes a multiple endmember extraction algorithm with local spatial-spectral information(MEELSI)to account for spectral variability.Firstly,the original hyperspectral image is divided into some non-overlapping image blocks,and automatic target generation process algorithm is run on these blocks to extract candidate endmembers.Then,the local spatial-spectral information is used to refine the candidate endmember set.Finally,K-means clustering algorithm is used to select multiple endmembers per scene component.The performances of different unmixing methods are compared using both synthetic hyperspectral image and real hyperspectral image.The experimental results demonstrate that the proposed multiple endmember extraction-based spectral unmixing algorithm is effective,and the endmembers extracted by the proposed algorithm are more accurate than the state-of-the-art single endmember extraction algorithms.