基于前视声呐图像的AUV目标识别与跟踪

AUV target recognition and tracking based on forward looking sonar image

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2 500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,ir-regular 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 car-ried out,and the Jet mapping is used to pseudo-color map the sonar gray image to improve the data labeling speed and accur-acy,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@0.5094.17%,FPS is about 22 frames,which shows that the lightweight network has application value in un-derwater moving target tracking.