中文摘要：

针对光学和雷达遥感协同应用于城市地物类型高精度识别难这一问题,提出了一种信息融合与自适应提升（adaptive boosting,adaboost）和引导集成（bootstrap aggregation,bagging）分类器集成模型的遥感影像地物识别方法。该方法充分利用光学和雷达遥感数据提供的不同信息,达到提高遥感图像在地物识别方面应用的潜力。首先选择GS（Gram-Schmidt）、主成分变换（principal components transform,PCT）、HSV（hue,saturation,value）和改进的多孔小波算法（a trous algorithm for wavelet,ATWT）融合算法对信息源进行融合,然后采用bagging和adaboost集成算法对支持向量机（a library for Support Vector Machines,LibSVM）、功能树（function tree,FT）、快速有效的裂具算法（ripper alogrithm for fast,effective rule induction,JRIP）、序列最小优化算法（sequential minimal optimization,SMO）分类器进行集训练学习提高地类识别精度。研究以意大利北部帕维亚地区的ERS SAR与Landsat TM影像对为信息源,通过融合定量指标评价和典型地物识别应用验证,结果表明提出的多分类器模型能够充分利用光学和雷达遥感信息,稳健有效地进行地物类别提取,地物识别精度比单独使用光学和雷达数据提高15到17个百分点。

英文摘要：

In order to improve the effectiveness of optical and SAR data collaboratively processing for land use/ land cover classification,a （bootstrap aggregation, bagging） and （adaptive boosting, adaboost） based multi-classifier ensemble method is proposed in this research. The modified wavelet transform and color domain transform methods include （a trous algorithm for wavelet and intensity,hue and Saturation ATWT_IHS）, （Gram-Schmidt, GS） , （principal components transform, PC T）, and （Hue,Saturation, Value, HSV） were applied on ERS SAR and Landsat TM optical dataset over PAVIA city, Italy. The processing outcomes were further processed with different classifiers for typical object identification. And the bagging and adaboost algorithms were explored for LibSVM, FI, JRIP, and SMO classifiers training. The results were evaluated by quantitative indexes for fused images and overall accuracy for classification maps, which demonstrated that the proposed multi-classifier ensemble strategy could sufficiently utilize information provided by both optical and SAR remotely sensed data for typical land use/cover types robust and efficient identification, and land use/cover identification accuracy of proposed method for optical and SAR could improve 15,17%.