建筑物的纹理和光谱信息的多样性一直是自动化识别的瓶颈。 针对此问题, 提出了一种彩色遥感图像建筑物提取方法, 该算法结合中性集和均值漂移, 对转换到中性集空间的影像进行均值漂移分割, 生成以影像中主要地物类型为核心的光谱类别图像, 提取建筑物。 通过中性集空间的增强及分割, 克服了传统均值漂移分割稳定性低、 光谱不连续及信息混杂的缺陷, 避免了地物识别前提取连通区等操作。 实验证明, 提出的算法可以简捷、 完整、 准确、 稳定地提取建筑物, 满足高分辨率遥感影像建筑物的提取要求。

The diversity of spectral and textural information of constructions has become an obstacle in automatically detecting buildings. To overcome that, a method to detect constructions based on remote sensed images is proposed. It synthesizes neutrosophic set and mean shift algorithms to segment images transformed to neutrosophic set domain. After segmentation, an image is generated whose pixels mainly describe spectral categories based on main land features. Constructions can be extracted by generated spectral information. The algorithm overcomes the shortcomings of low stability, spectral discontinuity and complicated spectral information by enhancing and segmenting image in neutrosophic set domain. It avoids operations of extracting connected area before recognizing land features as well. Experiments show that the proposed algorithm can not only extract entire constructions steadily, precisely, completely and simply but also satisfies the demand of extracting constructions from high resolution remote sensing images.