Defect detection of weld X-ray image based on edge AI
-
摘要: 为了提高深度学习在X射线焊缝缺陷检测中的实用性，降低缺陷检测任务的硬件需求，提出模型参数量仅为3.6 M的YOLO-M网络. 通过在网络中引入轻量级的倒残差结构，减少网络计算量；采用多尺度预测机制，网络分层预测不同缺陷特征；跨网格扩增图片正样本信息，加快网络训练过程中的收敛速度. 结果表明，YOLO-M网络不仅应用于传统计算机，而且成功试验于超低功耗边缘人工智能芯片勘智K210中. 所提方法在嵌入式端的检测准确度为93.5%，检测速度为11帧/s. 该方法具有良好的检测准确度，极大降低了缺陷检测的成本.Abstract: In order to improve the practicability of deep learning in X-ray weld defect detection and reduce the hardware requirements of defect detection task, the YOLO-M network with merely 3.6 M parameters is proposed. The amount of network calculation is reduced by introducing a lightweight inverse residual structure into the network; the multi-scale prediction mechanism is used to predict different defect characteristics by network layer; the positive sample information of the image is amplified across the grid to speed up the convergence speed in the process of network training. The network is not only applied to traditional computers, but also tested in the ultra-low power edge artificial intelligence chip Kanzhi K210. The results show that the detection accuracy of the proposed method is 93.5% and the detection speed is 11 fps. This method has good detection accuracy and greatly reduces the cost of defect quality inspection.
-
Keywords:
- defect detection /
- deep learning /
- lightweight /
- embedded equipment
-
-
![]()
图 5 CR图像检测过程可视化
Figure 5. Visualization of CR image detection process. (a) input; (b) output; (c) confidence; (d) class
![]()
图 6 PO图像检测过程可视化
Figure 6. Visualization of PO image detection process. (a) input; (b) output; (c) confidence; (d) class
![]()
图 7 LOP图像检测过程可视化
Figure 7. Visualization of LOP image detection process. (a) input; (b) output; (c) confidence; (d) class
![]()
图 8 LOF图像检测过程可视化
Figure 8. Visualization of LOF image detection process. (a) input; (b) output; (c) confidence; (d) class
![]()
图 9 嵌入式设备和检测结果
Figure 9. Embedded device and detection results. (a) K210 development board; (b) bubble defect detection results
表 1 焊缝识别效果对比
Table 1 Comparison of weld recognition effect
测试网络 准确率YAP(%) 平均准确率
YmAP(%)检测速度
YFPS/(帧·s−1)模型参数量
YMP (M)CR LOF LOP PO YOLO V3 93.12 93.11 94.30 98.18 94.68 5.25 61.54 Faster-RCNN 85.33 95.40 95.26 92.11 92.03 6.30 41.14 Cascade-RCNN 87.17 92.31 98.30 95.33 93.28 7.88 68.94 RegNet 89.97 94.77 98.51 96.87 95.03 9.00 31.48 CentripetalNet 84.05 90.13 93.19 91.04 89.60 7.87 205.68 YOLO-M 92.53 93.30 93.40 96.46 93.92 100 3.60 
下载: 导出CSV
-
[1] 迟大钊, 马子奇, 程怡, 等. 不等厚板搭接焊缝缺陷数字X射线检测[J]. 焊接学报, 2019, 40(11): 45 − 48. doi: 10.12073/j.hjxb.2019400286 Chi Dazhao, Ma Ziqi, Cheng Yi, et al. Digital X-ray detection of lap weld defects in unequal-thickness plates[J]. Transactions of the China Welding Institution, 2019, 40(11): 45 − 48. doi: 10.12073/j.hjxb.2019400286
[2] Yan Z H, Xu H, Huang P F. Multi-scale multi-intensity defect detection in ray image of weld bead[J]. NDT & E International, 2020, 116: 102342 − 102357.
[3] Hou W, Zhang D, Wei Y, et al. Review on computer aided weld defect detection from radiography images[J]. Applied Sciences, 2020, 10(5): 1878 − 1893. doi: 10.3390/app10051878
[4] 修延飞, 李海超, 胡广泽, 等. 一种用于穿孔塞焊焊缝特征提取的视觉识别算法[J]. 焊接学报, 2020, 41(2): 75 − 79, 86. Xiu Yanfei, Li Haichao, Hu Guangze, et al. A visual recognition algorithm for feature extraction of perforated plug weld[J]. Transactions of the China Welding Institution, 2020, 41(2): 75 − 79, 86.
[5] 刘涵, 郭润元. 基于X射线图像和卷积神经网络的石油钢管焊缝缺陷检测与识别[J]. 仪器仪表学报, 2018, 39(4): 247 − 256. Liu Han, Guo Runyuan. Detection and recognition of weld defects of petroleum steel pipe based on X-ray image and convolutional neural network[J]. Chinese Journal of Scientific Instrument, 2018, 39(4): 247 − 256.
[6] 樊丁, 胡桉得, 黄健康, 等. 基于改进卷积神经网络的管焊缝X射线图像缺陷识别方法[J]. 焊接学报, 2020, 41(1): 7 − 11. Fan Ding, Hu Yude, Huang Jiankang, et al. X-ray image defect recognition method of pipe weld based on improved convolutional neural network[J]. Transactions of the China Welding Institution, 2020, 41(1): 7 − 11.
[7] 谢经明, 刘默耘, 何文卓, 等. 基于轻量化YOLO的X射线焊缝图像信息检测[J]. 华中科技大学学报(自然科学版), 2021, 49(1): 1 − 5. Xie Jingming, Liu Moyun, He Wenzhuo, et al. X-ray weld image information detection based on lightweight YOLO[J]. Journal of Huazhong University of Science and Technology (Natural Science Edition), 2021, 49(1): 1 − 5.
[8] Ren S, He K, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 39(6): 1137 − 1149.
[9] 鞠默然, 罗海波, 王仲博, 等. 改进的YOLO V3算法及其在小目标检测中的应用[J]. 光学学报, 2019, 39(7): 253 − 260. Ju Moran, Luo Haibo, Wang Zhongbo, et al. Improved YOLO V3 algorithm and its application in small target detection[J]. Journal of Optics, 2019, 39(7): 253 − 260.
[10] Nacereddine N, Goumeidane A B, Ziou D. Unsupervised weld defect classification in radiographic images using multivariate generalized Gaussian mixture model with exact computation of mean and shape parameters[J]. Computers in Industry, 2019, 108: 132 − 149. doi: 10.1016/j.compind.2019.02.010
[11] Hou D, Miao Z, Xing H, et al. Exploiting low dimensional features from the MobileNets for remote sensing image retrieval[J]. Earth Science Informatics, 2020, 13(4): 1437 − 1443. doi: 10.1007/s12145-020-00484-3
[12] Wang X, Han Y, Wang C, et al. In-Edge AI: Intelligentizing mobile edge computing, caching and communication by federated learning[J]. IEEE Network, 2019, 33(5): 156 − 165. doi: 10.1109/MNET.2019.1800286
[13] Zheng Z, Wang P, Liu W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[C]//Association for the Advancement of Artificial Intelligence. Proceedings of the AAAI Conference on Artificial Intelligence. New York, USA, 2020, 34(7): 12993−13000.
[14] Mery D, Riffo V, Zscherpel U, et al. GDXray: The database of X-ray images for nondestructive testing[J]. Journal of Nondestructive Evaluation, 2015, 34(4): 1 − 12.
计量
- 文章访问数: 665
- HTML全文浏览量: 361
- PDF下载量: 112
