一种实时、自适应的侧扫声呐小目标分割算法  

A Real-time and Adaptive Side Scan Sonar Small Target Segmentation Algorithm

作　　者：李秋实 王旭旸 李国林[3] 张福海[1] 刘佳[4] 许枫[4] 谢翔[2] Li Qiushi;Wang Xuyang;Li Guolin;Zhang Fuhai;Liu Jia;Xu Feng;Xie Xiang(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China;Department of Microelectronics and Nano-electronics,Tsinghua University,Beijing 100083,China;Department of Electronic Engineering,Tsinghua University,Beijing 100083,China;Institute of Acoustics of Chinese,Academy of Sciences,Beijing 100190,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350 [2]清华大学微电子与纳电子学系,北京100083 [3]清华大学电子工程系,北京100083 [4]中国科学院声学所,北京100190

出　　处：《南开大学学报（自然科学版）》2022年第3期15-20,共6页Acta Scientiarum Naturalium Universitatis Nankaiensis

摘　　要：针对侧扫成像声呐获取的水下小目标,提出了一种基于声呐原始回波数据的实时、自适应分割算法.该算法利用水底回波分布的统计特点,采用了一种基于威布尔分布模型的自适应阈值确定方法,实现对目标的初始分割;在此基础上提出了一种基于亮区回波特点的区域生长方法,得到精确的分割结果;之后利用霍夫检测方法合并过分割区域,利用小目标物理尺寸信息进行尺寸检测消除山脊、岩石等尺寸较大的虚假目标.实验结果表明,仅利用目标物理尺寸的检测方法,其速度能够达到3 s/帧,目标召回率达到100%,虚警率仅为34.62%.A real-time and adaptive segmentation algorithm based on sonar original echo data is proposed for underwater small targets acquired by lateral scan imaging sonar.The algorithm uses the statistical characteristics of underwater echo distribution,and adopts an adaptive threshold determination method based on Weibull distribution model to realize the initial segmentation of the target.Based on this,a region based on the characteristics of bright region echo is proposed.The growth method is used to obtain accurate segmentation results.Then,the Hough detection method is used to merge the segmented regions,and the small target physical size information is used for size detection to eliminate large-scale false targets such as ridges and rocks.The experimental results show that the detection speed of the target physical size can reach 3 s/frame,the target recall rate reaches 100%,and the false alarm rate is only 34.62%.

关 键 词：侧扫声呐原始回波数据 目标分割 威布尔函数 区域生长 霍夫变换 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]