论文部分内容阅读
高空间分辨率遥感影像在许多领域均有应用。由于遥感影像数据量大且内容复杂,目前少有针对这种影像的有效分割方法。引入一种快速、稳健的多尺度分割方法——均值漂移,该方法是一种通过简单迭代快速自适应上升的模式搜索法。基于均值漂移算法的分割方法,并充分利用光谱特征与空间特征,通过具有一定物理意义的参数控制分割精度,与目前商用软件eCognition提出的分割算法相比,同样达到与视觉分割一致的效果,并且速度更快。
High spatial resolution remote sensing images have applications in many fields. Due to the large amount of remote sensing image data and the complicated content, there are few effective methods for this kind of image segmentation. A fast and robust multiscale segmentation method, mean shift, is introduced, which is a pattern search method that is rapidly and adaptively increased by simple iterations. Based on the mean shift algorithm segmentation method and making full use of the spectral features and spatial features, the segmentation accuracy is controlled by the parameters with certain physical meaning. Compared with the segmentation algorithm proposed by the current commercial software eCognition, the segmentation algorithm also achieves the same effect as the visual segmentation, and faster.