端元快速提取的光谱梯度特征搜索法

由于数据量大,目前大多数端元提取算法均需较长的计算时间,限制了这些算法的有效应用。本文提出了以光谱梯度特征为搜索条件的快速端元提取方法,其核心包括基于光谱梯度特征的候选端元快速筛选和基于光谱解混误差的端元识别两部分。由于能够从影像中快速筛选出少量的像元光谱作为候选端元,故具有较好的计算性能;同时由于避免了非端元光谱参与端元识别,使得识别的结果具有更高的精度。试验表明,相比经典的IEA算法和ECHO算法,该算法不仅能大幅度提高端元提取速度,而且具有更准确的端元识别能力。同时,基于该算法原理,也可对现有各种算法进行改进,提升现有的各种端元提取算法的运算速度。

A Fast Endmember Extraction Algorithm Using Spectrum Gradient Features

Due to the large amount of image data ,most algorithms for endmember extraction cost huge time ,which limits the wide application of them .A fast endmember extraction algorithm is proposed by using SpectrumGradient Features as the searching rule .The core idea is composed of two parts ,namely , rapid screening of candidate endmembers based on Spectral Gradient Features and endmember identifica‐tion based on spectrum unmixing residual .Being able to quickly screen out a small amount of pixels from the image as candidate endmembers , the algorithm has excellent computational performance .This algorithm can also avoid non‐endmember spectrum participating in endmember identification and can obtain a result of higher accuracy .The experimental result shows that this new algorithm can greatly improve the endmember extraction speed and recognize endmembers more accurately compared with IEA and ECHO .What’s more ,existing algorithms for endmember extraction can be applied better based on the principle of this algorithm ,and the extraction speed can be improved remarkably .