摘要
为了有效解决遥感影像中普遍存在的混合像元导致遥感影像定量解译精度低的问题,对两种不同混合像元端元提取算法进行了比较分析。纯净像元指数算法随着迭代次数的增加时间效率大大降低,而经典的N-FINDR算法初始端元数目选择的任意性会导致像元解混的精度不一,因此本文提出了一种基于纯净像元指数改进的N-FINDR算法。改进的N-FINDR算法相较于传统的N-FINDR算法能够准确构建候选端元集合并求得最优解。该算法结合高光谱影像数据的特点,首先利用纯净像元指数求取备选端元数目;然后以此为基础运用经典的N-FINDR算法求解最大的单形体顶点,将求解后顶点作为纯净像元,并完成丰度反演;最后使用ENVI产品中自带的经过大气校正的航空高光谱数据cup95eff.int对算法进行验证。试验结果表明,以纯净像元指数改进的N-FINDR算法在整体精度方面优于传统的N-FINDR算法。
In order to solve the low accuracy problem of the quantitative interpretation of remote sensing images caused by mixed pixels,two different mixed pixel extraction algorithms are analyzed and compared in this paper.In consideration of the pure pixel index exponential algorithm efficiency greatly reduces as the number of iterations increases and the random choice of classical N-FINDR algorithm＇s element will lead to different accuracy of the solution,an improved N-FINDR algorithm based on pure pixel index is proposed.Relative to the traditional N-FINDR algorithm,the improved N-FINDR algorithm can construct the candidate element and obtain the optimal solution more efficiently. Synthesize the characteristics of hyper-spectral image data,first,the improved N-FINDR algorithm uses the pure pixel index to calculate the number of alternate port elements,and then use the classical N-FINDR algorithm to solve the maximal single-body vertex,then use the pure element completes the abundance inversion.Finally,the algorithm is verified by the air-calibrated airborne hyper-spectral data Cup95 eff.int in the ENVI products. The experimental results show that the improved NFINDR algorithm is more efficiently than the traditional N-FINDR algorithm in the aspect of endmember extraction.
出处
《测绘通报》
CSCD
北大核心
2018年第2期89-93,共5页
Bulletin of Surveying and Mapping
基金
湖南省自然科学基金湘潭市联合基金(2016JJ5023)
关键词
高光谱遥感
混合像元分解
端元提取算法
hyper-spectral remote sensing
unmixing pixel
endmember extraction algorithm
