中文摘要：

本文研究了针对高分辨率合成孔径声纳图像中常见的管道、线缆等重要水下设施的线目标提取方法。基于图割理论的Grab-Cut算法相比其他迭代算法具有较快的收敛速度,但需要人工辅助选定前/背景区域的初始化条件;为此,本文设计了基于尺度放缩后进行Radon变换的感兴趣区域提取环节,作为Grab-Cut的初始化步骤解决方案,使之可以快速自动解译;此外,该优化方案还缩小了模型训练的样本容量,提升了直线目标的提取精度和效率。经实验验证,该方法可以快速准确地提取直线目标,且具有相对较强的鲁棒性。

英文摘要：

In this paper, a line target extraction method of underwater cables and pipes in synthetic aperture sonar （SAS） image has been discussed. The relatively novel method, Grab Cut, which is usually known for high convergence rate, requests an extra artificial aid to label the original area of foreground and background to initialize the algorithm. In case of that, a line region of interests （ROI） marking procedure based on Radon transform of resized image is proposed in this paper as a solution to the initialization of Grab Cut, which makes it possible for sonar images to be fast and automatically interpreted, Furthermore, the optimization also results in reduction of samples required by model training and increase of detection accuracy and efficiency. Finally, it is demonstrated in series of experiments on sonar images that the method is able to extract line targets fast and accurately, and is also relatively robust.