高光谱影像端元数目估计的HAHT模型

端元提取是高光谱遥感研究的重点内容之一，在高光谱影像信息识别、环境监测、资源探测和丰度反演等领域有着重要运用。为了实现有效的端元提取，如何准确估计（尤其是未知区域）高光谱影像中端元数目就显得更为关键，特别是在无人或境外地区的遥感探测方面极有实际价值。端元数目估计过多或者过少都会影响端元提取和混合像元分解的精度。本文基于谐波分析（Harmonic Analysis，HA）理论实现了高光谱影像有效去噪，并结合二元假设检验方法构建了一种高光谱影像端元数目估计的谐波分析假设检验（HA-Hypothesis Testing，HAHT）模型。通过AVIRIS和Hyperion高光谱影像的可行性分析与普适性验证，并与HFC（Harsanyi Farrand Chang）、特征值极大似然函数（Eigenvalue Likelihood Maximization, ELM）和最小误差高光谱信号辨识法（Hyperspectral Signal Identification by Minimum Error, Hysime）等常规的端元数估计算法应用成果相对比，表明HAHT模型所估计的端元数目与实际地物数具有更高的吻合度。同时，采用较成熟的连续最大角凸锥（Sequential Maximum Angle Convex Cone，SMACC）方法提取了端元波谱曲线，通过比较设置2（HFC估计数）、8（HAHT估计数）和14（Hysime估计数）不同端元数的提取结果，也证明HAHT模型在估计端元数目时具有较高准确性，以及较好的适用性和应用前景。

Harmonic Analysis-Hypothesis Testing Model on Estimating the Endmember Numbers of Hyperspectral Image

As one of the key contents of hyperspectral remote sensing research, endmember extracting on hyperspectral image has the important applications in information recognition, environmental monitoring, resource exploration and abundance inversion. It is crucial how to estimate the endmember numbers of hyperspectral image acquired especially from the unknown zone in order to extract endmembers effectively, and the estimation would have highly practical value in particular with the remote sensing detection on the unmanned or overseas regions. The numbers estimated too much or too little would affect the accuracy on endmember extracting and pixel unmixing. In view of the above, an effective de-noising method on hyperspectral image was realized based on the harmonic analysis(HA) theory, and a new model of the HA-hypothesis testing(HAHT) was constructed to estimate the endmember numbers of hyperspectral image combined with the two hypothesis testing. By feasible analyzing and universal verifying based on the HAHT with AVIRIS and Hyperion hyperspectral image, and comparing with the results obtained by the conventional algorithms such as Harsanyi-Farrand-Chang(HFC), Eigenvalue Likelihood Maximization(ELM) and Hyperspectral signal identification by minimum error(Hysime), it indicated that the endmember numbers estimated by the HAHT are more consistent with actual numbers of the features. At the same time, the common Sequential Maximum Angle Convex Cone (SMACC) algorithm was used to extract respectively 2, 8 and 14 endmember spectra of the AVIRIS image according to setting up the SMACC parameters of extracting endmember numbers that are estimated by the HFC, the HAHT and the Hysime correspondingly, the extracting results of 8 endmember spectra also proved the estimation on endmember numbers of HAHT model has higher accuracy, better applicability and application prospect.