高光谱图像具有丰富的光谱信息和空间信息，被广泛地用于地物分类和目标检测等。但是它具有大规模、高维和包含复杂流形结构等特点，现有的聚类算法和维度约简算法大多需要在整体数据集上计算，因而具有较高的时间复杂度，且无法有效地处理复杂流形数据。局部核心点是局部邻域中具有最大密度的对象，其发现过程充分利用了邻域信息，从而能在降低数据规模的同时有效地代表原始数据。本项目拟结合局部核心点开展以下研究：（1）通过引入局部核心点的概念，从高光谱图像数据中获得局部核心像素点代表原始像素空间，从空间维度降低数据规模；（2）在局部核心点的基础上研究高光谱图像的降维方法对光谱维度进行降维，从而避免在全部像素空间进行计算，同时达到从光谱维度降低数据规模的目的；（3）在约简后的高光谱图像数据上结合基于局部核心点的聚类算法对高光谱图像进行聚类分析。项目成果将为高光谱图像的聚类这一研究课题提供新的解决方案。

Hyperspectral image has rich spectral and spatial information, which has been widely used for feature classification and target detection. However, it is large-scale, high-dimensional, and includes complex manifold structures. Most existing clustering algorithms and dimensionality reduction algorithms need to compute on the whole data set, which has a high time complexity and cannot deal with complex manifold data. Local cores are objects with the maximum density among their local neighbors. The discovery process makes full use of neighborhood information, which makes them effectively represent the whole data and reduce the scale of data. Combining with local cores, this project intends to carry out the following research: (1) obtaining local core pixels from the hyperspectral image data to represent the original pixel space by introducing the concept of local cores, which reduces the scale of hyperspectral image data from spatial dimension; (2) doing research on dimensionality reduction method based on local cores to reduce the spectral dimensions of hyperspectral image, which avoids calculations in the whole data and reduces the size of hyperspectral image data from spectral dimension; (3) combining local cores-based clustering algorithm to cluster the reduced hyperspectral image data. This project provides a new solution for the research of clustering hyperspectral image.