基于相似性编组的航标自动提取算法

Automatic navigation mark detecting based on correlative groups

针对高分辨率光学遥感影像,利用航标间具有很强相似性的特点,提出了航标相似性编组的自动提取算法。首先采用单类支持向量机对遥感影像进行水陆分割,确定出水陆的边界线,再对水域内空洞目标进行轮廓检测,将检测出的小目标作为候选目标。利用航标的几何及灰度特征对候选目标进行初步筛选,获得疑似航标目标。最后计算疑似航标目标间的相关系数并以此为依据进行相关性编组,得到不含虚警的航标组。采用QuickBird 0.6m分辨率的融合影像进行实验,结果表明该方法可以提取出区域内80%以上的航标,具有很强的可行性。

As navigation marks have strong similarity among each other, a new method for detecting navigation marks in high resolution remote sensing imagery is proposed. On the basis of segmenting the land and the water using one-class support vector machine, the small targets within the water regions are detected and regarded as the candidate ones. The targets which satisfy the pixel intensity and the geometric feature of navigation marks will be saved and used for organizing correlative groups. If the relation coefficient between two navigation marks meets the given conditions, they are considered as one group, then the largest group is regarded as the group of navigation marks. Finally, the method is tested and verified with the QuickBird fusion 0.6 m imagery. With 82 % navigation marks detected, the method has been proven to be effective in the experimental region.