水下目标物侧扫声呐图像自动识别  

Automatic Recognition of Side-Scan Sonar Image for Underwater Target

作　　者：吴彬 方振 Wu Bin;Fang Zhen(Shanghai Dahua surveying&mapping technology Co.,Ltd.,Shanghai 201208,China)

机构地区：[1]上海达华测绘科技有限公司,上海201208

出　　处：《港口航道与近海工程》2024年第2期85-88,99,共5页Port,Waterway and Offshore Engineering

摘　　要：为解决现有港航工程、海洋工程建设中采用侧扫声呐进行水下目标检测和识别方法的局限性问题,通过引入YOLOV3深度学习方法,利用人工标记的侧扫声呐图像对深度神经网络进行训练和测试,检测水下沉船目标;采用转移学习方法,利用预先训练好的卷积神经网络对侧扫声呐数据进行特征提取、感兴趣区域(ROI)汇聚、目标定位和分类,实现目标自动检测和识别,提高了效率,且目标检测的平均识别精度达到88%。In order to solve the limitations of side-scan sonar in detecting and recognizing of underwater target in the construction of current port,channel and Marine works,YOLOV3 deep learning method was applied,namely,deep neural network was trained and tested by using manually labeled side-scan sonar images in order to detect underwater wrecks.The transfer learning method was also adopted based on side-scan sonar data,that is,a pre-trained convolutional neural network was used to extract features,converge regions of interest(ROI),locate and classify objects,achieve automatic detection and recognition of objects,improve the working efficiency.The average recognition accuracy of detected objects was up to 88%.

关 键 词：水下目标 侧扫声呐图像 深度学习 迁移学习 自动识别 

分 类 号：U652.4[交通运输工程—港口、海岸及近海工程]