论文部分内容阅读
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。
Aiming at the problem of extracting information from solar greenhouse on high resolution image, this paper presents a method of extracting contiguous solar greenhouse and single greenhouse by using support vector machine and nearest neighbor algorithm combined with texture features. Experiments show that the texture features can improve the classification accuracy. On the large-scale layer, the classification accuracy is greatly improved, but in the small-scale layer, the accuracy of classification accuracy is relatively small. The more the number of the involved operations, the more the classification The accuracy of spectral + texture combination is the highest in most cases. The best scheme of extracting contiguous greenhouse is SVM and spectrum + shape + texture (7 pixels × 7 pixels) with the total accuracy of 92.86%. Kappa The coefficient was 0.90, while the optimal solution for single-house solar greenhouse was SVM and spectral + texture (11 pixels × 11 pixels) with the total accuracy of 88.39% and Kappa coefficient of 0.86.