基于样本形态变换的高分遥感影像建筑物提取

针对高分辨率遥感影像中建筑物屋顶光谱信息多变引起建筑物提取精度降低的问题，提出基于样本形态变换的建筑物提取方法. 利用偏移阴影分析法自动提取初始建筑物样本，根据建筑物屋顶形态特征，合理利用样本旋转、偏移、缩放变换方法，构建自适应样本精细提取变换组合，以更完整、全面地提取建筑物样本；结合支持向量机（SVM）分类器进行影像分类，得到建筑物初始提取结果；提出基于形态特征的格网占比法对初始提取结果进行确认，剔除不规则非建筑物，实现对建筑物的准确提取. 对高分辨率遥感影像进行对比实验分析，以验证方法的有效性. 结果表明，与面向对象分类、反向传播（BP）神经网络、基于偏移阴影分析3种参照方法对比，所提方法的建筑物提取精度均优于参照算法.

Building extraction from high resolution remote sensing image based on samples morphological transformation

A building extraction method based on samples morphological transformation was proposed, aiming at the reduction of building extraction accuracy caused by the various spectral information on the building roof in the high-resolution remote sensing image. The shifted shadow analysis method was utilized to automatically extract the initial building samples. Rotation, offset, and zoom transformations were applied to the initial samples according to the roof shape characteristics of the building. And an adaptive sample fine extraction transformation combination was established so as to extract building samples more completely and comprehensively. The image was classified to obtain the initial extraction results of buildings, combined with the support vector machine (SVM) classifier. A grid proportion method based on morphological features was proposed to confirm the initial extraction results. The final buildings were extracted more accurately by eliminating irregular non-buildings. The comparative experiment analysis of high-resolution remote sensing images was conducted to assess the effectiveness of the proposed method. Comparison with three reference algorithms, i.e. object-oriented, back propagation (BP) neural network, and shifted shadow analysis, shows that the proposed method achieves a better accuracy of building extraction than the reference algorithms.