基于无人机影像的快速分割方法

Study of method for fast segmentation based on UAV image

为解决地震灾区震后空间数据难以及时获取的问题,考虑到无人机(unmanned aerial vehicle,UAV)遥感影像自身特点,提出了一种改进的均值漂移(Mean Shift)分割算法。按照影像特征将UAV影像划分为纹理区和匀色区,对匀色区直接进行Mean Shift算法分割得到初分结果;对纹理区综合提取颜色、纹理、形状信息构造高维特征空间,并根据归一化分布密度值求得其合适的带宽,再使用Mean Shift算法对特征空间进行模式分类得到分割结果;通过构造代价函数进行区域合并,消除过分割区域,得到最终分割结果。针对芦山地震后获取的高空间分辨率UAV影像进行分割实验,并提出一种考虑到面积和光谱的分割匹配指数对分割结果进行评价。实验结果表明:所提出的改进的Mean Shift算法的分割精度优于传统的Mean Shift算法,为后续的震害信息提取提供了数据保障。

In order to solve the problem that it is difficult to obtain spatial data in time in the earthquake stricken area,the authors propose in this paper a more efficient new segmentation algorithm on the basis of the own features of unmanned aerial vehicle（ UAV） remote sensing images. Firstly,the image is divided into several homogeneous color areas and texture areas through variance detection on the color space. Secondly,preliminary partition of the homogeneous color area is directly achieved by Mean Shift method. Meanwhile,for the texture area,a high dimensional feature space is set up based on the color,texture,and shape information,and the proper bandwidth is calculated according to the normalized distribution density before applying Mean Shift algorithm on the feature space to model classification so as to reach the partition. Finally,an object function is set up to realize area merging and then reach the final partition results by smoothing over partitioned areas. Tests were conducted for high spatial resolution remote sensing image segmentation on UAV images of Lushan earthquake stricken area. A segmentation matching index which considers area and spectrum is proposed to evaluate the segmentation result. The experimental results show that the improved method performs better than the traditional method,and can provide data protection for subsequent damage information extraction.