日光温室的纹理特征分层提取

针对高分辨率影像上日光温室的信息提取问题,该文提出了利用支持向量机、最近邻算法结合纹理特征在不同层上分别提取连片日光温室和独栋日光温室的方法。实验表明:纹理特征能提高分类精度,在大尺度的层上,分类精度提升幅度较大,但在小尺度的层上,分类精度提升幅度会比较小;并不是参与运算特征数越多,分类精度越高,多数情况下光谱+纹理组合的分类精度最高;提取连片日光温室的最优方案是支持向量机和光谱+形状+纹理(7像素×7像素),总精度为92.86%,Kappa系数为0.90,而提取独栋日光温室最优方案为SVM和光谱+纹理(11像素×11像素),总精度为88.39%,Kappa系数为0.86。

Extraction of greenhouse information using multiscale texture features

Aiming at the problem of extracting greenhouse information from high resolution image,in this article,a method using support vector machine (SVM) and nearest neighbor classifier with textural features extracted continuous and single greenhouses on multi-layers were proposed.The results showed that:texture feature could improve classification accuracy,which was improved more on layers with bigger scale and improved less on layers with smaller scale;classification accuracy was not always higher with more features,in most cases it would get better classification by using the combination of spectral and texture features;the best combination for extracting continuous greenhouse was SVM with regard to spectral,shape,texture features in the window size of 7pixel× 7pixel,overall accuracy and Kappa reached 92.85％ and 0.90,and the best combination for extracting single greenhouse was SVM with regard to spectral,shape,texture features in the window size of 11 pixel× 11 pixel,overall accuracy and Kappa reached 88.03 ％ and 0.85.