声呐图像由于水体不均匀、边界不规则以及声呐设备本身性能的限制，导致图像噪声明显、亮度不均、分辨率低，使得水下AUV装备在使用前视声呐进行水下目标检测时难度较大。针对该问题，基于m750d声呐探测获得的AUV声呐数据，进行了数据提取、高斯滤波处理、扇形映射处理，并采用Jet映射对声呐灰度图像进行了伪彩色映射提高数据标注速度和精度，制作获得了4组2500张声呐图像的AUV目标检测数据集；采用YOLOv4-tiny目标检测算法开展AUV目标检测研究，研究结果表明该方法在该数据集上表现优秀，mAP@0.50达到94.17%，FPS在22帧左右，说明该轻量级网络在水下AUV目标识别与跟踪应用上具有较好的应用价值。
The sonar image has obvious noise, uneven brightness and low resolution due to the uneven water body, irregular boundary and the limitation of the performance of sonar equipment itself, which makes it difficult for underwater AUV equipment to detect underwater targets using forward-looking sonar. To solve this problem, based on the AUV sonar data obtained from the m750d sonar detection, the data extraction, Gaussian filtering and sector mapping processing are carried out, and the Jet mapping is used to pseudo-color map the sonar gray image to improve the data labeling speed and accuracy, and four sets of AUV target detection data sets of 2500 sonar images are produced; The YOLOv4-tiny target detection algorithm is used to carry out AUV target detection research. The research results show that this method performs well in this dataset, mAP@0.50 94.17%, FPS is about 22 frames, which shows that the lightweight network has application value in underwater moving target tracking.
2024,46(5): 115-119 收稿日期：2023-03-02
DOI：10.3404/j.issn.1672-7649.2024.05.021
分类号：U666.7
作者简介：郑鹏(1994-),男,硕士,工程师,研究方向为AUV控制与水下探测
参考文献：
[1] 檀盼龙, 吴小兵, 张晓宇. 基于声呐图像的水下目标识别研究综述[J]. 数字海洋与水下攻防, 2022, 5(4): 342-353.
TAN Pan-long, WU Xiao-bing, ZHANG Xiao-yu. Review on underwater target recognition based on sonar image[J]. Digital Ocean & Underwater Warfare, 2022, 5(4): 342-353.
[2] 王凯, 秦丽萍, 卢丙举. 大噪声环境下前视声呐图像目标识别方法研究[J]. 舰船科学技术, 2022, 44(1): 125-130.
WANG Kai, QIN Li-ping, LU Bing-ju. Research on target recognition method of forward-looking sonar image in large noise environment[J]. Ship Science and Technology, 2022, 44(1): 125-130.
[3] 乔鹏飞, 邵成, 覃月明. 基于多波束前视声呐的水下静态目标的探测识别技术[J]. 数字海洋与水下攻防, 2021, 4(1): 46-52.
QIAO Peng-fei, SHAO Cheng, QIN Yue-ming. Detection and recognition technology of underwater static target based on multi-beam forward-looking sonar[J]. Digital Ocean & Underwater Warfare, 2021, 4(1): 46-52.
[4] 李宝奇, 任露露, 陈发, 等. 基于前视三维声呐的轨条砦识别方法[J]. 水下无人系统学报, 2022, 30(6): 747-753.
LI Bao-qi, REN Lu-lu, CHEN Fa, et al. A method of erect rail barricade recognition based on forward-looking 3D sonar[J]. Journal of Unmanned Undersea Systems, 2022, 30(6): 747-753.
[5] 张家铭, 丁迎迎. 基于深度学习的声呐图像目标识别[J]. 舰船科学技术, 2020, 42(23): 133-136.
ZHANG Jia-ming, DING Ying-ying. Sonar image target recognition based on deep learning[J]. Ship Science and Technology, 2020, 42(23): 133-136.
[6] 刘昊搏, 刘铁军, 汪海林, 等. 一种基于前视声呐的目标检测与跟踪方法[J]. 舰船科学技术, 2021, 43(21): 143-148.
LIU Hao-bo, LIU Tie-jun, WANG Hai-lin, et al. A target detection and tracking method based on forward looking sonar[J], Ship Science and Technology , 2021, 43(21): 143-148.
[7] ZHANG Tie-dong, LIU Shu-wei, et al. Underwater target tracking using forward-looking sonar for autonomous underwater vehicles[J]. Sensors (Basel, Switzerland), 2019, 20(1).
[8] QUIDU I, JAULIN L, BERTHOLOM A, et al. Robust multitarget tracking in forward-looking sonar image sequences using navigational data[J]. IEEE Journal of Oceanic Engineering, 2012, 37(3): 417-430.
[9] GALCERAN E, DJAPIC V, CARRERAS M, et al, A real-time underwater object detection algorithm for multi-beam forward looking sonar[J]. IFAC Proceedings Volumes, 2012, 45(5): 306-311.
[10] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[J]. IEEE, 2016. DOI:10.1109/CVPR.2016.91
[11] REDMON J, FARHADI A. YOLO9000: Better, Faster, Stronger[J]. IEEE, 2017: 6517-6525. DOI:10.1109/CVPR.2017.690.
[12] REDMON J, FARHADI A. YOLOv3: An Incremental Improvement[J]. arxive-prints, 2018.
[13] BOCHKOVSKIY A, WANG C Y, LIAO H . YOLOv4: Optimal Speed and Accuracy of Object Detection[J]. arxiv 2020(4).
