In synthetic aperture sonar (SAS) image, the underwater shipwreck targets are often buried by sediment or badly damaged. Only a part of the characteristics of artificial objects is retained. In this paper, firstly, based on the analysis of the ocean buried background, the Meanshift filtering is used to smooth the original image and convert the color image into binary one. Secondly, the residual contour of artificial target is extracted through the modified Canny edge detection algorithm. Thirdly, the Region Growing method is taken to remove the discrete interference and keep the intact edge of the line. Consideration with the principle of line alignment, the contours of shipwreck targets are gradually connected and aggregated. Finally, a large amount of measured practical SAS images are tested. The experimental results verified that the proposed algorithm can accurately detect the shipwreck target based on residual contour information, meanwhile with an acceptable timeliness for large size sonar image data.

中文翻译：

---

基于残差轮廓信息的沉船目标检测算法

在合成孔径声纳（SAS）图像中，水下沉船目标经常被沉积物掩埋或严重损坏。只保留了人造物体的一部分特征。本文首先在海洋掩埋背景分析的基础上，利用Meanshift滤波对原始图像进行平滑处理，将彩色图像转化为二值图像。其次，通过改进的Canny边缘检测算法提取人工目标的剩余轮廓。第三，采用区域生长法去除离散干扰，保持线条边缘完整。考虑到线对齐的原则，沉船目标的轮廓逐渐连接和聚合。最后，对大量实测的 SAS 图像进行了测试。