摘要
湿地资源含水量高,光谱特征混淆,一般方法难以有效提取湿地信息。采用支持向量机(SVM)方法对ALOS AVNIR-2数据进行湿地信息提取实验。首先,以分割和分类迭代的方法获得多尺度对象层;其次,提取对象的光谱特征、纹理特征、形状特征和自定义特征,采用SVM分类器进行影像分类;最后,通过地类的实际分布情况与初步分类结果的比较,分析SVM结果的错分情况,从而构建优化规则修正结果。两个地区湿地信息提取结果的总体精度分别为94.45%和94.18%,对应kappa系数为0.91和0.92,该方法明显提高了湿地类别的识别精度。
It is usually difficult to effectively extract wetland information with spectral features only,since wetland classes contain high percentage of water,which possibly confuses the spectral classification process.This paper integrates iterative segmentation and spatial optimization rules to improve wetland classification accuracy.The proposed method is a three-step classification routine that involves the integration of:(1)segmentation of the images into objects at different scales through the iterative process of segmentation and classification;(2)classification of the wetland areas supported by SVM with the spectral features,textures,shape features and customized features;(3)optimizing the classification results using spatial dependent rules and predicted according classification.Experimental results show that this method can significantly improve the wetland extracting accuracy.The overall accuracies of the two regions are 94.45%and 94.18%,and the kappa coefficients are 0.91and 0.92,respectively.
出处
《遥感信息》
CSCD
2014年第2期89-93,99,共6页
Remote Sensing Information
基金
福建省重点项目(2011Y0036)