基于OLI影像多参数设置的SVM分类研究

在遥感影像自动分类中仅使用光谱特征很难产生正确的分类,OLI影像是波段数较多的多光谱影像,如果增加纹理、几何等多种特征以提高分类精度,就会使得特征的维度很高.支持向量机善于解决小样本、非线性和高维的影像分类问题,但是核函数和参数的设置只能依靠实验来获得.文中在OLI影像中提取了23个特征,逐个测试核函数和参数值对分类结果的影响.研究的主要结论如下：RBF核的支持向量机分类精度最高,Sigmoid核支持向量机分类精度最低;核函数的选择对分类精度的影响最大;核函数和参数值的变化不会影响重要特征的使用,3种核的支持向量机分类所使用的重要特征基本一致.

Study on SVM classifications with multi-features of OLI images

OLI images are multispeetral images with many bands. Image classification can＇t get high accuracy only with spectral features. If texture, geometry and other features are used to the classification, the feature dimension will increase rapidly. Support vector machines are good at solving small sample, nonlinear and high-dimensional feature classification problems. But the kernel function and parameter settings can only be obtained by experiments. The 23 features of OLI images are extracted, kernel function and parameter values are tested for the classification results one by one. Main conclusions are as followed： when SVM with RBF kernel get the highest accuracy, SVM with Sigmoid kernel will get the worst result, and the kernel function impact on the classification accuracy will increase greatly; kernel function and parameter value changes do not affect the important features, and the most important features with three classifiers are basically the same.