基于图像分割和LSSVM的高光谱图像分类

摘要

A hyperspectral image classification method based on image segmentation and LSSVM is proposed,which com⁃bines spatial information to realize the hyperspectral imagery classification. Firstly hyperspectral image is segmented with mean⁃shift algorithm,and then dimension reduction of the data in each segmentation region is conducted and LSSVM classification of the data after dimension reduction is carried out. Finally the maximum voting method is used to fuse segmented map and obtain the final classification result. With the proposed method,the similarity matrices of segmented regions are derived and the new training sample set is trained to derive the low rank coefficient matrix,and the eigenvalue equation is built by means of similari⁃ty matrices and low rank coefficient matrix to solve the dimension reduction matrix,and then LSSVM is used to classify the data after dimension reduction. The exp⁃erimental results show that the hyperspectral image classification method based on image seg⁃mentation and LSSVM can effectively improve the classification accuracy of hyperspectral images.%提出一种基于图像分割和LSSVM的高光谱图像分类方法，将空谱信息结合起来进行高光谱图像的分类。首先利用均值漂移算法对高光谱图像进行分割，然后对每一块分割区域数据进行降维并且对降维后的数据LSSVM分类，最后用最大投票方法融合分割图和分类得到最终的分类结果。该文分类方法先对分割后的区域求出相似性矩阵并训练新样本集求出低秩系数矩阵，由相似性矩阵和低秩系数矩阵构造特征值方程求解出降维矩阵，然后利用混合核LSSVM对降维后的数据进行分类。实验结果表明，提出的基于图像分割和LSSVM的高光谱图像分类方法有效提高了高光谱图像的分类精度。