针对风电叶片表面缺陷形状不一,传统卷积神经网络涉及阈值筛选和非极大值抑制过程增加计算的复杂性等问题,提出一种基于改进Transformer（RT-DETR）的检测模型RH-DETR。通过对比ResNet18、SwinTransformer和CswinTransformer等主干网络,选取检测精度尚可且计算量较低的ResNet18作为主干特征提取网络;引入改进后的重参数模块（RP-Block）来更新主干网络的中basic-block,在不牺牲检测精度的前提下提高模型的推理速度;考虑到风电叶片表面缺陷形状不规则,使用基于高低频特征分离位置编码（HiLo）结构重构原有的尺度内特征交互模块（AIFI）来解决低层特征的丢失问题,提高模型的特征提取能力。实验结果表明RH-DETR模型取得了96.1%的准确率以及93.1%的平均精度均值,优于目前主流的YOLO检测模型,相较于原始模型RT-DETR,分别提高2.4和1.3个百分点,且模型计算复杂度降低50%以上,检测速度由32.6帧/s提升至64.7帧/s,满足工业实时检测的要求,可为风电叶片的检测维护提供技术支撑。
Abstract
In view of the different shapes of surface defects of wind turbine blades,the traditional convolutional neural network involves threshold screening and non-maximum suppression processes that increase the complexity of calculation,a detection model RH-DETR based on improved Transformer （real-time-detection transformer,RT-DETR） is proposed by comparing backbone networks such as ResNet18,SwinTransformer and CSwinTransformer,ResNet18 with relatively balanced detection accuracy and low computational complexity is selected as the backbone feature extraction network; the improved re-parameter module （RP-Block） is introduced to update the basic-block of the backbone network,while not losing detection accuracy,improving the inference speed of the model; considering the irregular shape of surface defects on wind turbine blades, the HiLo structure is used to reconstruct the original AIFI structure to solve the problem of low-level feature loss and improve the feature extraction ability of the model. The experimental results show that the accuracy and average accuracy of RH-DETR are 96.1% and 93.1% respectively, which are better than the current mainstream YOLO detection model. Compared with the original model RT-DETR,they are improved by 2.4 and 1.3 percentage points, respectively. In addition, the computational complexity of the model is reduced by over 50%. The detection speed is increased from 32.6 f/s to 64.7 f/s,which meets the requirements of industrial real-time detection and provides technical support for the detection and maintenance of wind blade.
关键词
风电叶片 /
表面缺陷 /
深度学习 /
RT-DETR /
注意力机制 /
轻量化网络
Key words
wind turbine blades /
surface defects /
deep learning /
RT-DETR /
attention mechanism /
lightweight network
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] ROGA S, BARDHAN S, KUMAR Y, et al.Recent technology and challenges of wind energy generation: a review[J]. Sustainable energy technologies and assessments, 2022, 52: 102239.
[2] Global Wind Energy Council. Global wind energy report 2025[R]. Brussels Wind Energy Council, 2025.
[3] MOURAD A I, ALMOMANI A, AHMAD SHEIKH I, et al.Failure analysis of gas and wind turbine blades: a review[J]. Engineering failure analysis, 2023, 146: 107107.
[4] 石腾, 许波峰, 陈鹏, 等. 基于机器视觉的风电机组叶片多类型损伤检测方法研究[J]. 太阳能学报, 2024, 45(6): 487-494.
SHI T, XU B F, CHEN P, et al.Study on multi-type damage detection method for wind turbine blades based on machine vision technology[J]. Acta energiae solaris sinica, 2024, 45(6): 487-494.
[5] 殷孝雎, 潘雪, 左雁斌, 等. 基于迁移学习的风电机组叶片损伤检测与分析[J]. 太阳能学报, 2024, 45(10): 506-511.
YIN X J, PAN X, ZUO Y B, et al.Wind turbine blade damage detection and analysis based on transfer learning[J]. Acta energiae solaris sinica, 2024, 45(10): 506-511.
[6] 邵延华, 张铎, 楚红雨, 等. 基于深度学习的YOLO目标检测综述[J]. 电子与信息学报, 2022, 44(10): 3697-3708.
SHAO Y H, ZHANG D, CHU H Y, et al.A review of YOLO object detection based on deep learning[J]. Journal of electronics & information technology, 2022, 44(10): 3697-3708.
[7] GUO J H, LIU C, CAO J F, et al.Damage identification of wind turbine blades with deep convolutional neural networks[J]. Renewable energy, 2021, 174: 122-133.
[8] ZHU X X, HANG X Y, GAO X X, et al.Research on crack detection method of wind turbine blade based on a deep learning method[J]. Applied energy, 2022, 328: 120241.
[9] MA L M, JIANG X Y, TANG Z Z, et al.Wind turbine blade defect detection algorithm based on lightweight MES-YOLOv8n[J]. IEEE sensors journal, 2024, 24(17): 28409-28418.
[10] ZHU Y F, LIU X Y.A lightweight CNN for wind turbine blade defect detection based on spectrograms[J]. Machines, 2023, 11(1): 99.
[11] HUANG C, CHEN M H, WANG L.Semi-supervised surface defect detection of wind turbine blades with YOLOv4[J]. Global energy interconnection, 2024, 7(3): 284-292.
[12] ZHANG J J, COSMA G, WATKINS J.Image enhanced mask R-CNN: a deep learning pipeline with new evaluation measures for wind turbine blade defect detection and classification[J]. Journal of imaging, 2021, 7(3): 46.
[13] ZHANG R, WEN C B.SOD-YOLO: a small target defect detection algorithm for wind turbine blades based on improved YOLOv5[J]. Advanced theory and simulations, 2022, 5(7): 2100631.
[14] ZHAO Y A, LV W Y, XU S L, et al.DETRs beat YOLOs on real-time object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, WA, USA, 2024: 16965-16974.
[15] 尹泽宇, 杨波, 陈金令, 等. 基于STD-DETR的轻量化小目标检测算法[J]. 激光与光电子学进展, 2025, 62(8): 136-146.
YIN Z Y, YANG B, CHEN J L, et al.Lightweight small object detection algorithm based on STD-DETR[J]. Laser & optoelectronics progress, 2025, 62(8): 136-146.
[16] 姜香菊, 王瑞彤, 马彦鸿. 基于轻量级改进RT-DETR边缘部署算法的绝缘子缺陷检测[J]. 电工技术学报, 2025, 40(3): 842-854.
JIANG X J, WANG R T, MA Y H.Insulator defect detection based on lightweight improved RT-DETR edge deployment algorithm[J]. Transactions of China Electrotechnical Society, 2025, 40(3): 842-854.
[17] 刘亚蒙, 赵友全, 孙振涛, 等. 构建改进RT-DETR算法检测隐形眼镜环状波纹缺陷[J]. 电子测量与仪器学报, 2024, 38(5): 1-9.
LIU Y M, ZHAO Y Q, SUN Z T, et al.Constructing an enhanced RT-DETR algorithm for detecting annular ripple defects in contact lenses[J]. Journal of electronic measurement and instrumentation, 2024, 38(5): 1-9.
[18] HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA, 2016: 770-778.
[19] CAI J F, PAN Z Z, ZHUANG B H.Fast vision transformers with HiLo attention[C]//Advances in Neural Information Processing Systems 35. New Orleans, Louisiana, USA, 2022: 14541-14554.
[20] ZHANG H, LI F, LIU S L, et al. DINO: DETR with improved DeNoising anchor boxes for end-to-end object detection[EB/OL].2022: arXiv: 2203.03605. https://arxiv.org/abs/2203.03605
[21] LIU Z, LIN Y T, CAO Y, et al.Swin transformer: hierarchical vision transformer using shifted windows[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, QC, Canada, 2022: 9992-10002.
[22] DONG X Y, BAO J M, CHEN D D, et al.CSWin transformer: a general vision transformer backbone with cross-shaped windows[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans, LA, USA, 2022: 12114-12124.
[23] DING X H, ZHANG X Y, MA N N, et al.RepVGG: making VGG-style ConvNets great again[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, TN, USA, 2021: 13728-13737.
[24] CHEN J R, KAO S H, HE H, et al.Run, don’t walk: chasing higher FLOPS for faster neural networks[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, BC, Canada, 2023: 12021-12031.
[25] DEVRIES T, TAYLOR G W. Improved regularization of convolutional neural networks with cutout[EB/OL].2017: arXiv: 1708.04552. https://arxiv.org/abs/1708.04552
[26] 廖力达, 罗晓, 黄斌, 等. 复杂背景下的多晶硅太阳电池缺陷检测[J]. 太阳能学报, 2024, 45(9): 295-303.
LIAO L D, LUO X, HUANG B, et al.Detection of defects in polysilicon solar cells in complex backgrounds[J]. Acta energiae solaris sinica, 2024, 45(9): 295-303.
[27] LI C Y, LI L L, JIANG H L, et al. YOLOv6: a single-stage object detection framework for industrial applications[EB/OL].2022: arXiv: 2209.02976. https://arxiv.org/abs/2209.02976
[28] LIU L Z, LI P R, WANG D H, et al.A wind turbine damage detection algorithm designed based on YOLOv8[J]. Applied soft computing, 2024, 154: 111364.
[29] 白家宇, 沙玲, 魏丹, 等. 基于改进YOLOv5的风机叶片表面缺陷检测方法[J]. 热力发电, 2025, 54(9): 95-103.
BAI J Y, SHA L, WEI D, et al.Surface defect detection method for wind turbine blade based on improved YOLOv5[J]. Thermal power generation, 2025, 54(9): 95-103.
PDF(3903 KB)
