论文部分内容阅读
基于区域合并分割方法的性能在很大程度上取决于区域模型、合并准则和合并顺序,依据遥感影像的目视解译原理,分析高分辨率遥感影像的特点,设计一种新的融合光谱、形状和空间位置的合并代价函数并进行区域相似性度量。同时加入面积控制参数,使得区域在光谱值相同的情况下优先合并小区域。在合并顺序的改进中,以最优邻接链的形式来表达和获取局部范围的最小合并代价区域对,确保每次相互合并的区域都为局部最优。对QuickBird多光谱影像进行分割试验,并与eCognition的分割结果进行比较,结果证明本文方法在分割精度上有优势,更符合人的视觉感知。
The performance of the segmentation-based segmentation method depends largely on the regional models, the merging rules and the merge order. According to the principle of visual interpretation of remote sensing images, the characteristics of high-resolution remote sensing images are analyzed and a new fusion spectrum is designed. Shape and spatial location of the merger cost function and regional similarity measure. At the same time, the area control parameters are added, so that the area preferentially incorporates the small area with the same spectral value. In the improvement of the merging sequence, the minimum merging cost region pairs in the local range are expressed and obtained in the form of the optimal adjacency chain to ensure that each merging region is locally optimal. The segmentation test of QuickBird multispectral image is carried out and compared with the eCognition segmentation results. The results show that the proposed method has the advantages of segmentation accuracy and is more in line with the human visual perception.