针对近红外场景仿真中需要将不同材质分类的问题，提出一种谱聚类的灰度纹理图像分割方法。首先利用mean-shift 算法，将原始图像预分割为多个具有准确边界的同质区域，并将这些区域描述为超像素；进而，将超像素构造的无向加权图作为谱聚类的输入，通过谱聚类的方法解决超像素的过分割问题。本文的方法在谱聚类过程中考虑了超像素的纹理特征，弥补了灰度图像在谱聚类过程中只顾及灰度和空间信息的不足。实验结果表明，利用本文分割方法不但减少了运算量，并且为精细的近红外场景仿真奠定了基础。

In order to solve the problem of classification of different materials in near-infrared scene simulation, an image segmentation method for spectral clustering is proposed. First, the original image is pre-segmented into multiple homogeneous regions with exact boundaries using the mean-shift algorithm, and these regions are described as super-pixels. Furthermore, the non-weighted graph of the super-pixel structure is used as the input of spectral clustering, and the over-segmentation problem of super-pixels is solved by spectral clustering. The method of this study takes into account the texture characteristics of super-pixels in the process of spectral clustering, which makes up for the lack of grayscale and spatial information in the process of spectral clustering. Experimental results show that the segmentation method reduces computational complexity and establishes the foundation of fine near-infrared scene simulation.