基于SVM的资源三号影像林地分类及精度评价研究.pdfVIP

西北林学院学报 2016 ， 31 (1): 180~185 Journal of Northwest Forestry University doi:10. 3969/j. issn.1001-746 1. 2016. 01. 32 基于 SVM 的资源、三号影像林地分类及精度评价研究 侯逸晨，赵鹏祥祷，杨伟志，张晓莉 (西北农林科技大学林学院，陕西杨陵 712100) 摘 要:运用 ZY-3 影像全色和多光谱影像，采用支持向量机(SVM) 法对黄龙山林区蔡家川林场林 地进行分类研究，探讨 SVM 法的分类能力及不同核函数、纹理窗口大小对森林植被分类精度的影 响。结果表明 :SVM 法在研究区 ZY-3 影像林地分类中精度比传统的极大似然法高;将光谱信息 与灰皮共生矩阵(GLCM) 构造派生的纹理信息结合能有效提高分类精度;采用 SVM 法分类时不 同核函数对分类结果的精度影响不显著;在选用 3X3 、 5X5 纹理窗口时分类精度更高。 关键词:遥感;资源三号影像;森林分类;支持向量机 中图分类号:S77 1. 8 文章编号:lOOl-7461(2016)Ol-0180-06 文献标志码:A Forest Classification and Accuracy Assessment in ZY3 Image with SVM Method HOU Yi-chen ,ZHAO Peng-xiang* ,YANG Wei-zhi ,ZHANG Xiao-li (College of Forestry , Northwest AF University. Yangling .Shaanxi 712100 .China) Abstract: Based on ZY-3 panchromatic and multispectral image ,the forest stands in H uanglong Mountain- ous areas were classified by support vector machine (SVM) method. The classification ability of SVM was tested ,and the influence of different kernel functions and texture window size on the accuracy of classifica- tion were examined. The data showed that the accuracy of the SVM on forest vegetation classification of ZY-3 image was higher than that of the traditional maximum likelihood method. The accuracy of classifica- tion could be evidently improved by combining spectral information and the image texture information based on gray level co-occurrence matrix (GLCM). Different kernel functions had no significant influence on the accuracy of classification by using SVM method. In addition ,the accuracy of classification was higher with 3 X 3 or 5 X 5 windows. Key words: remote sensing; ZY-3 image; forest classification; support vector