高光谱影像是一个三维的海量数据立方体，如果对高光谱图像直接进行分割，那么算法运算量会很大；如果对高光谱影像先进行数据降维再进行分割，则会损失图像的部分细节信息，影响分割效果。本文提出一种基于光谱角空间变换的高光谱图像分割方法，首先计算每个像元与其周围领域像元之间的光谱夹角，并把这些光谱角的值作为坐标值，将像元映射到一个低维空间中，计算低维空间中样本点到原点的距离并将其转换为灰度值，从而生成一幅突出了地物区块边缘信息的灰度图像。然后利用分水岭变换对生成的灰度图像进行分割，提取分割后各区块局部极小值点的光谱矢量，进行对比分析，将具有相似光谱矢量的区块合并，以解决分水岭变换的过分割问题。最后采用美国印第安纳州的 AVIRIS高光谱数据对本文算法进行了验证和分析。实验结果表明，相比 SVM、ECHO、SVMMSF等高光谱图像分割方法，本文方法能够获得更高的图像分割精度。

Hyperspectral images are massively large three-dimensional data sets, and as such, the segmentation algorithms for hyperspectral images are usually much more computationally complex than those for gray image segmentation. Reducing the dimensions of the hyperspectral image before segmentation leads to the loss of some details of the image, and therefore the segmentation quality cannot be guaranteed. In this paper, a novel hyperspectral image segmentation method based on spectral angular space transformation is proposed. Firstly, the spectral angles between each pixel and its neighboring pixels are calculated, and these are used as coordinate values to map the pixels into a low-dimensional space. The distances from the sample points to the origin in the low-dimensional space are calculated and converted into gray values. A grayscale image highlighting the edge information of the ground object distribution blocks is subsequently generated. The grayscale image is segmented by watershed transform and the spectral vectors at the local minimum points of all the segmented blocks are extracted and compared. Finally, the blocks with similar spectral vectors at the local minimum points are merged together to reduce the over-segmentation phenomena. In experiment, the AVIRIS hyperspectral data from Indiana Pines is used to validate and analyze the proposed method. The experimental results show that the proposed method can achieve higher image segmentation accuracy compared with other hyperspectral image segmentation methods such as SVM, ECHO, SVMMSF.