在遥感图像中，利用模糊扩展的概率测度描述混合像元各端元的丰度含量，是模糊集合理论进行混合像元分解的主要手段。然而由于非线性因素的影响以及自然地物的复杂性，往往导致了端元光谱的变异性，这是造成混合像元分解误差的主要原因之一。从模糊集合的另一种测度——可能性测度出发，讨论了对端元光谱的变异性的描述：在遥感图像多维正态分布假设的前提下，利用1维连续的卡方分布，提出了面向遥感图像可能性分布的模糊描述，并进一步提出了可能性分布的模糊划分算法，对整个图像进行端元的模糊划分，从而描述端元光谱的变异性。通过对真实的高光谱遥感图像进行检验，结果表明，该方法能充分地表达端元光谱的变异性，可为今后进一步提高遥感图像混合像元分解精度提供重要的参考。

In remote sensing image，the event of mixed pixel can be presented by fuzzy extended probability measures.These measures are considered by fuzzy classification.However，due to the effect of nonlinear factor and the complexity of nature materials，spectral variability occurs to the endmembers，which leads to the primary error in the process of spectral unmixing.This paper focused on the fuzzy presentation on the endmember spectral variability by possibility distribution for remote sensing image: on the assumption of the multidimensional normal distribution，chi-square distribution is proposed for the fuzzy presentation.Based on the chi-square distribution，a possibility distribution fuzzy partition arithmetic is developed for fuzzy partition of endmembers in the whole image，which presents the endmember spectral variability.In order to evaluate the performance of the algorithm，an experiment using real hyperspectral image is demonstrated in this paper.The proposed algorithm clearly reveals the spectral variability of endmembers in the image，which provides valuable insight into the accuracy of mixed pixel analysis.