Ship crack detection based on lightweight fast convolution and bidirectional weighted feature fusion network
-
摘要:
目的 针对人工目视与超声波方法的船舶裂纹检测存在效率低下、成本高昂和危险性高的特点，提出一种基于深度学习的船舶裂纹检测方法。 方法 首先，在YOLOv5s的主干网络中使用轻量化卷积结构（GSConv）替代标准卷积并融入注意力机制，在降低主干网络参数量与计算量的同时，提升主干网络对裂纹特征的提取能力；其次，在网络的颈部（Neck）使用基于PConv构建的C3_Faster替代原C3模块，提升模型的图像处理速度，增强模型快速性；最后，设计一种简化的双向加权特征融合网络（BiFFN）以改进原模型（YOLOv5s）中的特征聚合网络，提升裂纹的语义信息与位置信息的融合效果，以及模型对裂纹的识别准确度与定位精度。 结果 通过对船舶裂纹原始数据与增强数据的学习，所提方法实现了94.11%的检测精确度和93.50%的召回率，模型的计算量降低了17.93%，参数量降低了15.81%。 结论 研究表明，基于轻量化快速卷积与双向加权特征融合网络（MLF-YOLO）的船舶裂纹检测方法，实现了模型轻量化与较高的检测精确度和召回率，结果可为开发自主无人机船舶检测提供参考。 Abstract:Objectives Traditional ship crack detection methods based on artificial visual inspection and ultrasonic methods in ship repair and inspection processes have the characteristics of low efficiency, high cost, and high danger, a ship crack detection method based on deep learning is proposed. Methods First, a lightweight convolutional structure (GSConv) is used to replace the standard convolution and introduce attention mechanism in the backbone of YOLOv5s to achieve the reduction of network parameters and computation while enhancing the ability to extract crack features. Secondly, C3_Faster constructed by fast convolutional structure is used instead of the original C3 module in the neck of the network to improve the processing speed of the model and enhance its rapidity. Finally, designed a simplified bidirectional weighted feature fusion network (BiFFN) to enhance eeature aggregation in the original model (YOLOv5s) for Improved fusion of semantic and spatial information of cracks, and enhanced accuracy and localization precision in crack recognition. Results By training on both original and augmented ship crack datasets, the proposed method achieves a detection accuracy of over 94.11% and a recall rate of over 93.50%, while reducing the computational complexity by 17.93% and the parameter count by 15.81%. Conclusion The study demonstrates that the ship crack detection based on lightweight fast concolution and bidirectional weighted feature fusion network(MLF-YOLO), achieves lightweight model architecture and high detection accuracy and recall rates. This provides a reference for the development of UAV− ship autonomous inspection systems. -
图 12 YOLOv5s and MLF-YOLO模型检测结果对比
Figure 12. The detection results comparison between YOLOv5s and MLF-YOLO
表 1 消融试验结果
Table 1. Results of ablation experiment
试验 SConv GSConv C3_Faster BiFPN-S CBAM P/% R/% GFLOPs Best.pt
(MB)1 √ − − − − 87.1 89.7 16.6 14.3 2 − √ − − − 88.1 89.9 15.4 13.5 3 √ − √ − − 88.2 88.8 13.4 12.2 4 √ − − √ − 90.5 86.8 16.6 14.6 5 √ − − − √ 90.9 90.4 16.6 14.4 6 − √ − − √ 87.4 90.0 15.5 13.6 7 √ − √ − √ 88.2 87.1 12.9 12.0 8 √ − − √ √ 91.7 92.2 16.6 14.0 9 − √ √ − − 87.5 89.5 13.0 11.7 10 √ − √ √ − 88.6 88.1 13.4 12.2 11 − √ − √ − 88.5 90.1 14.0 12.1 12 − √ √ √ − 88.3 89.9 15.4 13.5 13 − √ − √ √ 88.8 88.06 14.9 13.6 14 − √ √ − √ 87.8 88.7 13.1 12.9 15 √ − √ √ √ 86.6 87.2 13.5 12.3 16 − √ √ √ √ 94.1 93.5 13.6 12.2 
下载: 导出CSV
-
[1] 王浩亮, 尹晨阳, 卢丽宇, 等. 面向海上搜救的UAV与USV集群协同路径跟踪控制[J]. 中国舰船研究, 2022, 17(5): 157–165. doi: 10.19693/j.issn.1673-3185.02916WANG H L, YIN C Y, LU L Y, et al. Cooperative path following control of UAV and USV cluster formaritime search and rescue[J]. Chinese Journal of Ship Research, 2022, 17(5): 157–165 (in Chinese). doi: 10.19693/j.issn.1673-3185.02916 [2] YANN L, BENGIO Y, HINTON G. Deep learning [J] Nature, 2015, 521(7553): 436-444. [3] LI Y T, TENG F B, XIAN J H, et al. Underwater crack pixel-wise identification and quantification for dams via lightweight semantic segmentation and transfer learning[J]. Automation in Construction, 2022, 9(144): 104600. [4] DUNG C V, ANH L D. Autonomous concrete crack detection using deep fully convolutional neural network[J]. Automation in Construction, 2019, 99(01): 52–58. [5] 任秋兵, 李明超, 沈扬, 等. 水工混凝土裂缝像素级形态分割与特征量化方法[J]. 水力发电学报, 2021, 40(2): 234–246. doi: 10.11660/slfdxb.20210224REN Q B, LI M C, SHEN Y, et al. Pixel-level shape segmentation and feature quantification of hydraulic concrete cracks based on digital images[J]. Journal of Hydroelectric Engineering, 2021, 40(2): 234–246 (in Chinese) . doi: 10.11660/slfdxb.20210224 [6] ZHAO X F, LI S Y. A method of crack detection based on convolutional neural networks [C]// Proceedings of the 11th International Workshop on Structural Health Monitoring. Stanford, CA, USA: DEStech Publications, Inc, 2017. [7] LI L F, Ma W F, LI L, et al. Research on detection algorithm for bridge cracks based on deep learning[J]. Acta Automatica Sinica, 2019, 45(9): 1727–1742. [8] CHA Y J, CHOI W, SUH G, et al. Autonomous structural visual inspection using region‐based deep learning for detecting multiple damage types[J]. Computer‐Aided Civil and Infrastructure Engineering, 2018, 33(9): 731–747. doi: 10.1111/mice.12334 [9] 余加勇, 李锋, 薛现凯, 等. 基于无人机及Mask R-CNN的桥梁结构裂缝智能识别[J]. 中国公路学报, 2021, 34(12): 80–90. doi: 10.3969/j.issn.1001-7372.2021.12.007YU J Y, LI F, XUE X K, et al. Intellgent identification of bridge structural cracks based on unmanned aerial vehicle and Mask R-CNN[J]. China Journal of Highway and Transport, 2021, 34(12): 80–90 (in Chinese) . doi: 10.3969/j.issn.1001-7372.2021.12.007 [10] LIN T Y, DOLLAR P, GIRSHICK R , et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE 2017 Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA: IEEE, 2017. [11] LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA : IEEE, 2018: 8759−8768. [12] TAN M, PANG R, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: IEEE, 2020: 10781-10790. [13] HOWARD, A G, ZHU M L, CHEN B, et al. Mobilenets: efficient convolutional neural networks for mobile vision applications [J]. arXiv preprint arXiv. (2017-04-17)[2023-06-07]. https://arxiv.org/pdf/1704.04861.pdf. [14] ZHANG X Y, ZHOU X Y, LIN M X, et al. Shufflenet: an extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA: IEEE, 2018: 6848−6856. [15] HAN K, WANG Y H, TIAN Q, et al. Ghostnet: more features from cheap operations[C]// Proceedings of the 2020 IEEE /CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA : IEEE, 2020: 1580−1589. [16] LI H L, LI J, WEN H B, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[J/OL]. arXiv preprint arXiv. (2022-06-06)[ 2023-06-07]. https://arxiv.org/ftp/arxiv/papers/2206/2206.02424.pdf. [17] CHEN J R, KAO S H, HE H, et al. Run, don't walk: chasing higher FLOPS for faster neural networks [C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, BC, Canada: IEEE, 2023. [18] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[J/OL]. arXiv preprint arXiv. (2015-06-08)[ 2023-06-07]. https://arxiv.org/pdf/1506.02640.pdf. [19] WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision (ECCV). [S. 1. ]: Springer, 2018: 3−19. [20] ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[C]//Proceedings of the 34th AAAI Conference on Artificial Intelligence. New York Hilton Midtown, New York, USA: AAAI Press, 2020, 34(7): 12993−13000. -
点击查看大图
图(18) / 表(1)
计量
- 文章访问数: 123
- HTML全文浏览量: 5
- PDF下载量: 0
- 被引次数: 0
