

浏览全部资源
扫码关注微信
1.湖南大学电气与信息工程学院, 湖南 长沙 410082
2.机器人视觉感知与控制技术国家工程研究中心, 湖南 长沙 410082
3.湖南星邦智能装备股份有限公司, 湖南 长沙 410600
Received:07 December 2024,
Revised:2024-12-25,
Accepted:25 December 2024,
Online First:07 April 2025,
Published:25 January 2026
移动端阅览
袁小芳,李潘,孙荣武等.基于轻量化与注意力机制的船舶除漆机器人实时目标检测[J].中山大学学报(自然科学版)(中英文),2026,65(01):13-22.
YUAN Xiaofang,LI Pan,SUN Rongwu,et al.Ship paint-removal robots real-time object detection based on lightweight and attention mechanism[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2026,65(01):13-22.
袁小芳,李潘,孙荣武等.基于轻量化与注意力机制的船舶除漆机器人实时目标检测[J].中山大学学报(自然科学版)(中英文),2026,65(01):13-22. DOI: 10.13471/j.cnki.acta.snus.ZR20240344.
YUAN Xiaofang,LI Pan,SUN Rongwu,et al.Ship paint-removal robots real-time object detection based on lightweight and attention mechanism[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2026,65(01):13-22. DOI: 10.13471/j.cnki.acta.snus.ZR20240344.
自动巡航船舶除漆机器人目标检测受外部干扰时,存在算法检测精度下降、难以达到实时性要求等问题。为了解决这些问题,首先将重参深度可分离移动网络模块(Repvit-MobileNet block)引入到YOLOV5的主干网络中,提高检测速度。其次,在骨干网络每个阶段后增加位置注意力机制,扩大模型的全局感受野,提升模型的目标定位及抗干扰能力。然后,将卷积块注意力模块(CBMA)引入到颈部网络中,通过融合CBMA模块增强特征提取能力,提高网络模型的检测性能。最后,提出了一种Refine-Loss损失函数,通过优化预测框和真实框的几何关系、兼顾IOU的权重和置信度信息,提高对机器人目标位置的检测精度。在船舶机器人实验数据集中进行测试与验证,结果表明:融合Repvit-MobileNet block与注意力机制的YOLOV5轻量化网络平均检测精度达到了84.1%,在边缘设备上的推理运算速度达到了26.6 f/s,满足船舶除漆机器人目标检测工业应用的需求。
When the automatic ship paint-removal robot encounters external interference, existing algorithms suffering from performance degradation and insufficient real-time processing capability. To address these challenges,the Repvit-MobileNet block is integrated into the backbone network of YOLOV5 to enhance detection speed. Additionally,the positional attention mechanism has been incorporated after each stage of the backbone network, broadening the model's global receptive field and improving both target localization and interference resistance. Then, a convolutional block attention module(CBAM) is implemented in the neck network, and the feature extraction ability is enhanced by integrating the CBMA module to improve the detection performance of the network model. Lastly,a Refine-Loss loss function is proposed to optimize the geometric relationship between the predicted bounding box and the true bounding box which also balances weight and confidence information related to IOU,leading to improved accuracy in detecting the robot's target position.Subsequent experiments from ship robotic datasets show that the lightweight YOLOV5 network combining Repvit-MobileNet block and attention mechanism can reach 84.1% in the experiment with average precision, and the inference speed on the edge device reaches 26.6 f/s, which meets the need of industrial applications for object detection of ship paint-removal robots.
方璇 , 刘俊锋 , 陈勇 , 等 , 2023 . 基于SLAM的爬壁机器人自主移动研究 [J]. 制造业自动化 , 45 ( 6 ): 85 - 88 .
姜泽 , 王珉 , 赵哲 , 等 , 2023 . 爬壁机器人发展现状与关键技术研究综述 [J]. 包装工程 , 44 ( 12 ): 29 - 38 .
李磊 , 杨幸 , 秦绪杰 , 等 , 2023 . 爬壁机器人研究现状及发展趋势 [J]. 信息对抗技术 , 1 : 1 - 6 .
李希平 , 舒勇 , 魏莹吉 , 等 , 2024 . 基于视觉辅助导航的船用除漆爬壁机器人设计 [J]. 工程机械 , 36 ( 11 ): 35 - 40 .
张文 , 丁雨林 , 陈咏华 , 等 , 2022 . 基于外部视觉与机载IMU组合的爬壁机器人自主定位方法 [J]. 清华大学学报(自然科学版) , 62 ( 9 ): 1524 - 1531 .
DING X , ZHANG X , MA X Y , et al , 2021 . RepVGG:Making VGG-style ConvNets great again [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Nashville, TN, USA : 13728 - 13737 .
HARIS M , HOU J , 2020 . Obstacle detection and safely navigate the autonomous vehicle from unexpected obstacles on the driving lane [J]. Sensors , 20 ( 17 ): 4719 .
HOWARD A , SANDLER M , CHEN B , et al , 2019 . MobileNetV 3 : Searching for MobileNetV3 [C]// IEEE/CVF International Conference on Computer Vision . Seoul, Korea : 1314 - 1324 .
HOWARD A G , ZHU M L , BO C , et al , 2018 . MobileNets: Efficient convolutional neural networks for mobile vision applications [EB/OL]. arXiv: 1704.04861 .
HOU Q B , ZHOU D Q , FENG J S , et al , 2021 . Coordinate attention for efficient mobile network design [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Virtual : 13708 - 13717 .
HU J , LI S , SUN G , et al , 2018 . Squeeze-and-excitation networks [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Salt Lake City, UT, USA : 7132 - 7141 .
LEI R , SHI C , RAN X , 2012 . Target detection of maritime search and rescue: Saliency accumulation method [C]// International Conference on Fuzzy Systems and Knowledge Discovery . Chongqing, China : 1972 - 1976 .
LI C Y , LI L L , JIANG H L , et al , 2024 . YOLOV6: A single-stage object detection framework for industrial applications [J]. arXiv: 2209.02976 .
LIN T Y , DOLLÁR P , GIRSHICK R , et al , 2017 . Feature pyramid networks for object detection [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Honolulu, HI, USA : 936 - 944 .
LV S D , ZHU L , WANG W W , et al , 2020 . Improving SSD for detecting small target in remote sensing image [C]// Chinese Automation Congress . Shanghai, China : 567 - 571 .
SANDLER M , HOWARD A , ZHU M , et al , 2018 . MobileNetV2: Inverted residuals and linear bottlenecks [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Salt Lake City, UT, USA : 4510 - 4520 .
SZEGEDY C , LIU W , JIA Y , et al , 2015 . Going deeper with convolutions [C]// IEEE Conference on Computer Vision and Pattern Recognition . Boston, MA, USA : 1 - 9 .
WANG A , CHEN H , L Z J , et al , 2024 . Rep ViT: Revisiting MobileNet CNN from ViT perspective [C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition . Seattle, WA, USA : 15909 - 15920 .
WOO S , PARK J , LEE J Y , et al , 2018 . CBAM: Convolutional block attention module [C]// Computer Vision—ECCV 2018 . Munich, Germany : 3 - 19 .
0
Views
628
下载量
0
CSCD
Publicity Resources
Related Articles
Related Author
Related Institution
京公网安备11010802024621