基于纹理特征的高分辨率遥感影像分类方法  

Remote Sensing Image Classification Based on Texture Features

作　　者：韦春桃[1,2] 王宁[1] 张利恒[1] 原凯敏[1] 邹瑄[1] 

机构地区：[1]桂林理工大学测绘地理信息学院,广西桂林541004 [2]桂林理工大学广西空间信息与测绘重点实验室,广西桂林541004

出　　处：《桂林理工大学学报》2013年第1期80-85,共6页Journal of Guilin University of Technology

基　　金：国家自然科学基金项目(41161073);广西青年科学基金项目(2012GXNSFBA053131);广西空间信息与测绘重点实验室研究基金项目(桂科能1103108-17;1207115-03);广西自然科学基金创新研究团队项目(2012GXNSFGA060001)

摘　　要：灰度共生矩阵能较好反映影像灰度统计规律,小波变换能较好反映影像的多尺度特性,利用两者结合进行了纹理特征提取。将灰度共生矩阵和小波变换提取纹理特征作为分类特征向量,建立基于支持向量机分类模型对高分辨率遥感影像进行分类;在支持向量机参数优化问题上,利用遗传算法进行参数寻优,有效的避免多学习和欠学习状态的发生。分类实验结果表明了本方法的有效性。Gray level co-occurrence matrix can reflect the space information of different pixel position,and wavelet transformation expresses the multi-scale image.This paper gives full play to the characteristics of GLCM and wavelet transformation,and extracts texture feature by combining them.The feature vector from the low-dimensional space is mapped into a high dimensional space and the optimal separating hyperplane is found in a high dimensional feature space through the support vector machine by solving the optimization problem so as to solve the classification problem of complex data.SVM parameter optimization using genetic algorithms can in avoid the excessive learning and less learning state.The classification of the high-resolution remote sensing image is established based on support vector machine classification model utilizing GLCM and wavelet transformation to extract texture features as the classification of feature vectors.Classification experiment shows the effectiveness of the method in this study.

关 键 词：纹理特征 遥感影像分类 灰度共生矩阵 小波变换 支持向量机 遗传算法 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]