Automated detection/classification of objects in side-scan sonar imagery
Automated detection/classification of objects in side-scan sonar imagery
复制标题
侧扫声纳图像中物体的自动检测/分类
DOI：
--
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
Chunhua Zhang
中科院分区：
文献类型：
--
作者：
Jie Tian;Chunhua Zhang
针对海底水声图像中目标的检测与识别问题，提出了一种改进的算法。该先进算法由基于非线性匹配滤波器的改进检测算法、分类特征提取器和支持向量机（SVM）分类器组成。检测阶段使用滑动匹配掩模来识别图像中的哪个区域与地雷的特征非常相似。根据侧扫声纳图像中海底目标的成像特点设计了掩模，并设计了SID，以保证较低的漏检率。对于每个检测到的目标样区域，特征提取器计算准备用于识别阶段的一组特征。水下目标识别是一个典型的小样本问题，因此选择支持向量机分类器，以保证良好的泛化性能。将该方法应用于实际的水声图像，实验结果证明了该方法的有效性。
An advanced algorithm is propbsed in this paper to solve the problem of detecting and recognizing the targets settled on sea-bottom underwater acoustic images. This advanced algorithm consists of an improved detection algorithm on the basis of a nonlinear matched filter, a classification feature extractor and a Support Vector Machine (SVM) classifier. The detection stage uses a sliding match mask to identify which region in the images is closely similar to the mine's signature. The mask is designed according to the imaging characteristics of bottom targets in side-scan sonar images, SID it is supposed to ensure a low missed detection rate. For each detected target-like region, the feature extractor calculates a set of features prepared to the recognition stage. Recognition of underwater objects is a typical small sample problem, so a Support Vector Machine classifkr is selected so that favorable generalization performance can be guaranteed. This scheme was experimentally used to some practical underwater acoustic images, and the efficiency was proved by the experimental results.