论文部分内容阅读
混合像元是遥感影像中普遍存在的现象,对此,本文提出基于加权后验概率的支持向量机进行影像混合像元分解。该分类算法可判定端元种类的同时得到每种地物的后验概率,从而进行非线性模型的混合像元分解。由于加权后验概率的支持向量机分类算法能够减少分类器受土地覆盖类型模糊样本点的干扰,因此,改善了非线性混合像元分解模型的精度。首先,由样本点计算得到核函数参数值,然后,计算影像中每一种土地覆盖类型的后验概率,将其作为各个两类支持向量机分类器的权系数并求得多类后验概率值,确定影像每一种土地覆盖类型并得到丰度值。本文采用TM多波段遥感影像验证该方法的可行性,实验区位于我国东北部的大兴安岭中北段地区,土地覆盖类型包含农田、居民地、水体、荒地等。将本文提出的混合像元分解方法结果与标准支持向量机模型分解的结果对比表明,以加权后验概率的支持向量机遥感影像混合像元分解方法精度优于标准支持向量机模型。
Mixed pixel is a common phenomenon in remote sensing images. In this paper, we propose a support vector machine based on weighted posterior probability for image mixed pixel decomposition. This classification algorithm can determine the posterior probabilities of each feature by determining the endmember types, and then perform the mixed pixel decomposition of the nonlinear model. Since the weighted posteriori probability SVM classification algorithm can reduce the interference of the classifier by the fuzzy sample points of the land cover type, the accuracy of the nonlinear mixed pixel decomposition model is improved. First, the kernel function parameter values are calculated from the sample points. Then, the posterior probability of each land cover type in the image is calculated and used as the weight coefficient of each two types of support vector machine classifiers and multiple types of posterior probability Value, determine each type of land cover image and get the abundance value. This paper verifies the feasibility of this method using TM multi-band remote sensing images. The experimental area is located in the middle and northern part of Daxinganling in the northeast of China. The land cover types include farmland, residential area, water body and wasteland. Comparing the results of the hybrid pixel decomposition method proposed in this paper with the results of the standard support vector machine model decomposition, it shows that the precision of the hybrid pixel decomposition method using weighted posteriori probabilistic support vector machine remote sensing images is better than the standard support vector machine model.