通过分析高分辨率遥感影像中平原建成区的纹理特征和局部关键点特征，提出了基于多核学习、多尺度分割以及多假设投票的平原建成区提取方法。该方法利用MR8纹理特征和尺度不变特征变换(SIFT)算法提取建成区，融合多个特征进行学习和分类，从而加强了分类器的鲁棒性和稳定性，提高了检测准确率。该方法还通过超像素分割和多假设投票将基于图像块的判别结果转化为基于像素的检测结果，完全消除块状效应，使得目标区域具有准确的边缘和形状。在多幅GF-1卫星遥感图像上进行测试，结果显示: 提出方法的平均检测精度为80%，平均召回率高于85%，平均F值可达80%以上，综合指标高于其他方法，验证了提取平原地形建成区的可行性和准确性。由于建成区提取结果已精确到了像素级别，同时避免了漏检和误检，提取出的建成区影像很准确。

By analyzing the textural features and local key points of the built-up area in a plain from high resolution remote sensing images, a method to extract the built-up area in the plain was proposed based on multi-core learning, multi-scale segmentation and multi-hypothesis voting. With the proposed method, MR8 texture characteristics and Scale Invariant Feature Transform (SIFT) algorithmwere used to extract the built-up area, and multi-characteristics was fused to implement the learning and classification to improve the robustness and stability of classifiers and to enhance the detection accuracy. Then, based on the pixel segmentation and multi-hypothesis voting, the discriminant result based on image blocks was translated into test result based on pixels to completely eliminate the block effect and to make the target area showing precise edges and shapes. The proposed method has been validated in GF-1 satellite images. The results show that the average detection precision, average recall rate and the average F-measure of the method have been achieved above 80%, 85% , and 80%, respectively. Moreover, its comprehensive performance is better than that of other methods. These results demonstrate the feasibility and accuracy of this method. As extraction precision of the built-up area has been to be the pixel level and the leak detection and error detection have been avoided, the built up area images extracted are very accurate.