针对风电叶片内腔结构复杂、缺陷种类多样、难以准确检测等问题，提出了改进的单次多边界框检测器（SSD）缺陷检测算法，并提出 3方面改进：通过将 SSD 基础网络由可变形卷积神经网络（VGG-16）变成残差网络（ResNet101）以优化预测边界框的回归和分类任务的输入特征；通过加入全卷积空间注意力模块（FCSE）使模型更加关注重要特征，从而提高检测的准确性；通过在损失函数中添加超参数来控制平滑区域，使模型更加稳定。在自建风电叶片内腔数据集上的对比实验表明，改进的 SSD 模型平均精度（mAP）值为 83.6%，比原始 SSD模型提升了 9.4个百分点，检测精度优于其他基于 SSD 框架的主流模型，同时该模型大幅减少了模型参数量，降低了模型的复杂度和存储需求，检测速度为31.6 f/s，可满足实际生产工作中的检测速度需要。
Abstract
This paper aims to solve the problem of accurate detection of various types of defects in the complex internal cavity structure of wind turbine blades. An improved SSD (single shot multibox detector) algorithm is thus proposed and three aspects of improvement are made: 1) in terms of network framework, the base network of SSD is changed from VGG-16 to ResNet101 to optimize the input features for the regression and classification tasks of predicting bounding boxes; 2) an FCSE attention module is added to make the model pay more attention to important features and improve its detection accuracy; 3) the loss function is improved by adding a hyperparameter to control the smooth region, making the model more robust. Through comparative experiments on a self-built wind turbine blade internal cavity dataset, the improved SSD model achieves an mAP value of 83.6%, which is 9.4 percentage points higher than that of the original SSD model, and has advantages over other mainstream models based on SSD framework in detection accuracy, while greatly reducing the model parameter quantity, lowering the model complexity and storage requirements, and achieving a detection speed of 31.6 f/s, meeting the detection speed needs in practical production.
关键词
风电叶片 /
单次多边界框检测器 /
缺陷检测
{{custom_keyword}} /
Key words
turbine blades /
single shot multibox detector /
defect detection
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Chua L O. CNN: A vision of complexity[J]. International Journal of Bifurcation and Chaos, 1997, 7(10): 2219-2425.
[2] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition. New York: ACM, 2014: 580-587.
[3] Girshick R. Fast R-CNN[C]//Proceedings of IEEE International Conference on Computer Vision (ICCV). Piscataway, NJ: IEEE, 2015: 1440-1448.
[4] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[C]//Advances in Neural Information Processing Systems, Montreal, Canada: NIPS, 2015: 91-99.
[5] He K M, Gkioxari G, Dollar P, et al. Mask R-CNN[C]//Proceedings of IEEE International Conference on Computer Vision (ICCV). Piscataway, NJ: IEEE, 2017: 2961-2969.
[6] Jiang P Y, Ergu D J, Liu F Y, et al. A review of yolo algorithm developments[J]. Procedia Computer Science, 2022, 199: 1066-1073.
[7] Liu W, Anguelov D, Erhan D, et al. SSD: single shot MultiBox detector[M]//Leibe B, Matas J, Sebe N, et al. Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[8] Fu C Y, Liu W, Ranga A, et al. DSSD: Deconvolutional single shot detector[J]. arXiv preprint arXiv: 1701.06659, 2017.
[9] Jeong J, Park H, Kwak N. Enhancement of SSD by concatenating feature maps for object detection[J]. arXiv preprint arXiv:1705.09587, 2017.
[10] Li Z, Zhou F. FSSD: Feature fusion single shot multibox detector[J]. arXiv preprint arXiv:1712.00960, 2017.
[11] Yi J R, Wu P X, Metaxas D N. ASSD: Attentive single shot multibox detector[J]. Computer Vision and Image Understanding, 2019, 189: 102827.
[12] 郑晓娟, 王晓东, 郭建军, 等. 基于声发射技术的复合材料风电叶片缺陷监测技术研究 [D]. 杭州: 浙江大学, 2019.
[13] 蒋善超. 风电叶片检测技术的研究进展[J]. 光电子, 2019, 9(1): 28-33.
[14] 陈晓峰, 王晓东, 郭建军, 等. 风电叶片典型缺陷的性能评估与模式识别[J]. 复合材料学报, 2019, 36(11): 2995-3004.
[15] Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of IEEE International Conference on Computer Vision (ICCV). Piscataway, NJ: IEEE, 2017: 2980-2988.
[16] Tammina S. Transfer learning using VGG-16 with deep convolutional neural network for classifying images[J]. International Journal of Scientific and Research Publications (IJSRP), 2019, 9(10): 143-150.
[17] Wu Z F, Shen C H, van den Hengel A. Wider or deeper: Revisiting the ResNet model for visual recognition[J]. Pattern Recognition, 2019, 90(C): 119-133.
[18] Torralba A, Russell B C, Yuen J. LabelMe: Online image annotation and applications[J]. Proceedings of the IEEE, 2010, 98(8): 1467-1484.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
PDF(1564 KB)
({{custom_author.role_cn}}), {{javascript:window.custom_author_cn_index++;}}
