两种改进的遥感影像地物分类方法对比研究

摘要

In the segmentation of remote sensing images,the texture feature is an important feature in the classification of the feature class.In this paper,we propose a method of using remote sensing image classification based on Bayesian formula.According to judgment rule,we choose the one with the greatest probability of improvement as the result of the decision classification.In contrast,an improved K-mean al-gorithm based on Daubechies1 wavelet is proposed.The two-dimensional single-scale wavelet transform is used to decompose the image pixels into different dimension spaces.The average energy value of each sub-block band of each image is used as a feature component for fusion.Finally,wavelet transform is used to restore the classified image.We implement the algorithm by MATLAB 2010a software,and carry out the experiment based on the remote sensing image data of Haikou Town,Dianchi Lake Basin,Yunnan Province.At last,the results are qualitative analysis and quantitative evaluation.The results show that two kinds of remote sensing image segmentation algorithm have a better effects in the object classification of high-resolution remote sensing image.%遥感影像的分割中纹理特征是分辨、提取影像中地物类别的一个重要特征,文中提出一种利用贝叶斯公式进行遥感影像分类的思路,根据判断规则,选择改进概率最大的一个作为决策分类的结果.对比提出了另一种基于Daubechies1小波的改进K-mean算法影像分割技术,利用二维单尺度小波变换将图像像素分解到不同的维度空间,利用每个图像子块频带的平均能量值作为特征分量进行融合,最后利用小波逆变换还原分类图像.文中使用MATLAB 2010a软件进行算法实现,利用云南省滇池流域海口镇的2015年遥感影像数据进行了实验,最后对结果进行了定性分析和定量评价.结果表明,上述两种遥感影像分割算法在高分辨率遥感图像地物目标分类提取中取得较好的应用结果.