基于空间信息的高光谱图像聚类算法

Hyperspectral image clustering algorithm based on spatial information

遥感图像具有丰富的光谱信息和空间信息,因此遥感图像分类被广泛的应用于气象预报、国防预警等多个领域。文中对遥感图像进行非监督分类,提出了基于区域合并的遥感图像分类方法。首先采用主成分首先采用主成分分析法(PCA)对高光谱遥感数据集进行降维处理,在尽可能保留图像信息的情况下滤除大部分的光谱维,接着利用灰度直方图将空间信息引入降维后的遥感图像,然后对于由此增加的空间信息维度进行第二次降维,再后采用高斯混合模型通过迭代收敛最终完成聚类。文章最后在PaviaUniversity数据集上进行遥感图像分类实验,通过比较总体分类精度与kappa系数发现,此算法在遥感图像分类的精确性与正确性提高了约40%。

Remote sensing images have rich spectral information and spatial information. Therefore, remote sensing image classification is widely used in many fields such as weather forecasting and national defense early warning. In this paper,the unsupervised classification of remote sensing images is proposed,and a remote sensing image classification method based on regional merging is proposed. Firstly,the principal component is firstly used to reduce the dimensionality of the hyperspectral remote sensing dataset by principal component analysis(PCA). Most of the spectral dimensions are filtered out while retaining the image information as much as possible,and then the spatial information is filtered by the gray histogram. The reduced-dimensionality remote sensing image is introduced,and then the second dimension reduction is performed on the spatial information dimension thus increased,and then the clustering is completed by iterative convergence using the Gaussian mixture model. At the end of the paper,the remote sensing image classification experiment was carried out on the PaviaUniversity dataset. By comparing the overall classification accuracy with the kappa coefficient,the accuracy and correctness of the algorithm in the classification of remote sensing images was improved by about 40%.