论文部分内容阅读
本论文尝试讨论两个主题:主题一为利用主成分分析PCA方法应用于像元阶层资料融合技术的研究。主题二为应用Dempster-Shafer evidence theory方法于特征阶层数据融合技术的研究。在第一个主题中,由于合成孔径雷达的数据具有全偏极特性,在此选取了对植被较为敏感的HV极化合成孔径雷达数据,与具有光谱特性的光学SPOT数据做数据融合处理以利接下来的地物分类。首先,本研究利用小波转换技术来滤除合成孔径雷达斑驳噪声,在接下来融合步骤中,主成分分析出来的第一部分(PC1)是用做完滤除噪声后的合成孔径雷达取代,在数据融合后,进行地物分类是采用最大似然法来分类融合影像。在第二个主题中,利用全偏极雷达数据的极化特性结合SPOT数据的光谱特性,其主要目的是为了增加分类的精确度。首先使用李式滤波器滤除全偏极雷达数据噪声,接下来同样是使用采用最大似然法来分类融合影像,(不同的在于全偏极雷达影像使用Wishart几率分布,在光学影像采用multivariate Gaussian几率分布)将每个类别中每个像元属于某个类别的几率值计算出来,再利用Dempster-Shafer evidence theory来结合这些类别的机率值。最后产生出一张新的分类影像。实验的结果显示分类的精确度比较于未融合的资料都有明显提升的效果,也证明了此两个数据融合方法对于不同数据特性的融合都是很成功的。
This thesis attempts to discuss two main topics: Topic one is the research on the PCA method using principal component analysis (PCA) applied to pixel-level data fusion. The second topic is the application of Dempster-Shafer evidence theory in feature-level data fusion. In the first topic, the data of synthetic aperture radar (SAR) have the characteristics of fully biased. Here we selected the HV polarization SAR data sensitive to vegetation and data fusion with the optical SPOT data with spectral characteristics The next feature classification. First of all, in this study, wavelet transform is used to filter mottle noise of synthetic aperture radar. In the next fusion step, the first part (PC1) of principal component analysis is replaced by Synthetic Aperture Radar Fusion, the classification of features is the use of maximum likelihood method to classify the fusion image. In the second topic, the main purpose of using the polarization characteristics of fully polarimetric radar data in combination with the spectral characteristics of SPOT data is to increase the classification accuracy. First, the Li-style filter is used to filter out the data of the all-polarimetric radar. Next, the fused image is classified using the maximum likelihood method (the difference is that the Wishart probability distribution is used in the all-polarimetric radar image and the multivariate Gaussian Probability distribution) Calculate the probability that each pixel belongs to a category in each category, and then use Dempster-Shafer evidence theory to combine the probability values of these categories. Finally produce a new classification of images. The experimental results show that the accuracy of the classification is significantly improved compared with the unfused data, and it is also proved that the two data fusion methods are very successful for the fusion of different data features.