基于声图像的海底地形边界提取算法

摘要

An automatic method for seabed terrain boundary extraction based on the acoustic images was proposed in order to overcome the problem of non‐automatic extraction .The connected region satisfying region criterion was found by the connected component labeling method ,and then a single point from the region was set as the starting point for dynamic programming optimization .In the optimizing process ,the concept of energy function incorporating with grayscale ,gradient ,location and Webber resolution ,was introduced and the boundary was then extracted automatically in the field of adjacent column dynamic optimization under the criterion of energy function minimum .Evaluation experiment with 36 cases of real data demonstrates that ,the accuracy of boundary extraction achieves 92 .6% ,the similarity with golden standard is 93 .3% and the mean Euclidean distance with golden standard is 0 .23 pixels . The results indicate that ,on the basis of keeping water longitudinal information ,the proposed method could combine the spatial information better ,extract the seabed terrain boundary accurately and have good robustness .%针对海底地形边界难以自动提取的问题，提出一种基于声图像的自动提取算法．利用连通域标记找到满足面积准则的连通区域，以此区域上的单个点作为动态规划寻优的起点，在寻优过程中引入包括灰度、梯度、位置、韦伯分辨率等信息的能量函数的概念，在相邻纵列动态寻优的范围内以能量函数最小为准则实现边界的自动提取．对36例实际数据进行测试，边界提取准确率达到了92．6％，与金标准的吻合度为93．3％，与金标准的平均欧式距离为0．23个像素．结果表明：该方法能够在保留水体深度方向信息的基础上，更好地融合空间域信息，准确地提取海底地形边界，且鲁棒性好．