【作者】 王莉；

【导师】 陈龙乾； 李学友；

【作者基本信息】 中国矿业大学 ， 摄影测量与遥感， 2009， 硕士

【摘要】 高分辨率遥感影像的土地覆盖分类是研究的热点,而高分辨率遥感影像的居民地信息提取的研究尚处于初步阶段,本文以高分辨率IKONOS遥感影像为主要数据源,分别进行像素级融合、特征级融合和决策级融合的土地覆盖分类和居民地信息提取研究,旨在探索出高分辨率遥感影像有效的土地覆盖分类和居民地信息提取方法。首先,分析了遥感原始图像预处理、影像增强处理以及遥感影像特征提取的基本理论和方法,为后续影像融合、土地覆盖分类和居民地提取研究做好准备。影像像素级融合主要采用PCA、小波+PCA、IHS和Brovey四种融合方法,并对融合影像进行目视和定量评价;在像素级IKONOS遥感影像融合的基础上,分析融合影像的光谱特征和纹理特征,并分别基于光谱特征、纹理特征、光谱特征与纹理信息的特征级融合进行SVM分类,提取居民地;然后在充分分析影像不同窗口纹理特征的基础上,提出应用组合纹理特征进行土地覆盖分类和居民地信息提取方法。最后运用多数投票法和D-S证据理论对三种单一分类器(SVM,MLC和OO)的分类结果进行决策级融合提取居民地,在分析决策级融合结果的基础上,对多数投票法的决策级融合进行了改进,继而提取居民地信息。IKONOS影像像素级融合结果表明,PCA融合能够较好地提高多光谱影像的空间分辨率,同时很好地保持了多光谱影像的光谱信息,光谱畸变较小,在本实验区域具有较好融合效果;特征级融合结果表明,基于光谱信息和组合纹理特征的特征级融合土地覆盖分类和居民地信息提取,能够提高分类精度,更好地提取居民地信息;决策级融合结果表明,改进的多数投票决策级融合方法,相对于三种单一分类器的提取结果,达到了决策级融合信息提取的目的,得到了较好的居民地提取效果。更多还原

【Abstract】 Land use classification based on high-resolution remote sensing images is a hot topic, but the research on residential areas extraction is still in preliminary stage, In this thesis, the high resolution IKONOS remote sensing image was took as the main data source, and the pixel-level fusion, feature-level fusion and decision-level fusion were respectively conducted for land cover classification and residential areas extraction. All of this was aimed to explore an efficient method for land cover classification and residential areas extraction with high resolution remote sensing image.Firstly, to prepare for the image fusion, land cover classification and residential areas extraction, the theory and method of remote sensing preprocessing, image enhancement processing and the remote sensing image feature extraction were analyzed.Secondly,the image pixel-level fusion methods were adopted the PCA, wavelet plus PCA, IHS and Brovey ,and the fusion result image was studied by the visual and quantitative estimate. Thirdly, based on the pixel-level IKONOS remote sensing fusion, we analyzed the spectral feature and texture feature of the fusion result image was analyzed, the feature-level fusion was conducted by the spectral feature, texture feature, spectral plus texture feature, and using SVM classifier,the residential areas was extracted. Then, land cover classification and residential areas extraction with combined texture feature was proposed by the sufficient analysis of the texture feature with different image window. Finally, the decision-level fusion was conducted to extract residential areas with multi-vote method and D-S evidence theory for the single classifier result of SVM, MLC and OO,and based on the analysis of decision-fusion result, the decision-level fusion of multi-vote method was improved to extract residential areas information.The pixel-level fusion result with IKONOS image indicated that the PCA fusion method can preferably improve the spatial resolution of multi-spectral image, at the same time maintaining the spectral information, and lower spectral distortion. The feature-level fusion result showed that it is better for land cover classification and residential areas extraction based on spectral information and feature-level fusion with combined texture feature than other image feature.,and improved the classification accuracy.And lastly the decision-level fusion result indicated that, compared with the three single classifiers’extraction result, the improved multi-vote decision-level fusion method achieved a better classification result, and the method achieved the purpose of residential areas extraction with decision-level fusion.更多还原