数字海洋与水下攻防2025，Vol.8Issue(4)：472-480,9.DOI:10.19838/j.issn.2096-5753.2025.04.010
基于深度学习的侧扫声呐图像目标检测
Object Detection in Side-Scan Sonar Images Based on Deep Learning
摘要
Abstract
Object detection in side-scan sonar images holds significant application value in marine resource exploration,underwater archaeological research,and national defense security.However,existing automatic detection methods still face the challenge of low detection accuracy due to factors such as background noise,resolution,and target scale variations in side-scan sonar images.To address these issues,a new target detection method for side-scan sonar images based on deep learning is explored in this paper.By replacing the conventional convolutions in backbone network of baseline model with space-frequency convolutions,this method effectively extracts frequency texture features,significantly suppressing the impact of speckle noise in side-scan sonar images on target detection accuracy.Furthermore,a multi-scale feature fusion module is used to improve the spatial pyramid pooling-fast in the baseline model,enhancing the ability to detect small targets.Experimental results on the shipwreck side-scan sonar dataset demonstrate that the proposed method achieves improvements of 4.05%and 5.61%in mAP@50 and mAP@50 95:,respectively,compared to the baseline model.关键词
侧扫声呐图像/目标检测/深度学习/频率特征Key words
side-scan sonar image/object detection/deep learning/frequency feature分类
信息技术与安全科学引用本文复制引用
汤瑞,陈依民,高剑,郝邵文,王亚周..基于深度学习的侧扫声呐图像目标检测[J].数字海洋与水下攻防,2025,8(4):472-480,9.基金项目
国家自然科学基金"面向多源多方位探测的AUV集群分布式事件触发鲁棒协同控制研究"(52471347) （52471347）
"双一流"建设专项基金"师资队伍建设项目-国家级青年人才"(0206022GH0202) （0206022GH0202）
西北工业大学博士论文创新基金"面向少样本侧扫声呐图像的域自适应目标检测方法研究"(CX2025046). （CX2025046）
