现代电子技术2025，Vol.48Issue(9)：28-35,8.DOI:10.16652/j.issn.1004-373x.2025.09.005
基于Star_YOLOv8的水下珍品检测方法研究
Research on underwater treasure detection method based on Star_YOLOv8
摘要
Abstract
To effectively achieve the rapid and accurate detection of all kinds of treasures in underwater images,and to solve the problems of underwater treasure detection,such as occlusion,low accuracy and slow inference speed,this paper proposes an underwater treasure detection algorithm based on Star_YOLOv8.Firstly,the C2f_StarNB module is fused in the backbone network to capture the underwater image treasure features and achieve lightweight and fast detection of the model.Secondly,EMA attention is added to achieve the fusion of channel and spatial features,improve the efficiency of cross-attention computation of underwater image treasure features of different scales,reduce the impact of noise,and improve the expression of salient features and detection accuracy of complex scenes and multi-scale targets in underwater images.Finally,the dynamic focusing loss function Repulsion Loss is introduced to improve the convergence ability of the model and the overall detection effect of the multi-scale distribution of occluded treasures.In order to verify the effectiveness of the improved method,experimental validation is carried out on the underwater treasure dataset,and the mAP@0.5 are used to verify that the incorporation outperforms the other methods in terms of lightness and attention,improved by 8%and 9.4%,respectively.In addition,the proposed method achieves a detection performance of 0.863 on the mAP@0.5 compared to previous state-of-the-art underwater treasure detection methods.The experimental results show that the proposed method has a certain improvement in terms of model performance such as detection accuracy and speed for all types of treasures in underwater images.关键词
YOLOv8/水下珍品/目标检测/StarNet/注意力机制/Repulsion Loss损失函数Key words
YOLOv8/underwater treasure/object detection/StarNet/attention mechanism/loss function Repulsion Loss分类
信息技术与安全科学引用本文复制引用
夏建军,高定国,许松涛..基于Star_YOLOv8的水下珍品检测方法研究[J].现代电子技术,2025,48(9):28-35,8.基金项目
国家自然科学基金项目(62166038) （62166038）
拉萨市科技计划项目(LSKJ202306) （LSKJ202306）
