基于改进YOLOv8的航空铝合金焊缝缺陷检测方法

苏志威; 黄子涵; 邱发生; 郭朝阳; 殷晓芳; 邬冠华

doi:10.13224/j.cnki.jasp.20230414

基于改进YOLOv8的航空铝合金焊缝缺陷检测方法

doi: 10.13224/j.cnki.jasp.20230414

1.
南昌航空大学无损检测技术教育部重点实验室，南昌 330063
2.
江西洪都航空工业集团有限责任公司检验检测中心，南昌 330024

基金项目: 国家自然科学基金（62201241）；江西省自然科学基金青年基金（20224BAB214057）；赣鄱俊才-主要学科学术和技术带头人青年项目（20232BCJ23092）；南昌航空大学博士引进人才项目（EA201908298）

详细信息

作者简介:
苏志威（2002-），男，主要从事无损检测方向的研究

通讯作者:
邱发生（1990-），男，讲师，博士，研究方向为无损检测。E-mail：qiufasheng2019@nchu.edu.cn

中图分类号: V260.5；TP391.4
计量
- 文章访问数: 1434
- HTML浏览量: 646
- PDF量: 218
- 被引次数: 0
出版历程
- 收稿日期: 2023-06-25
- 网络出版日期: 2024-01-03

Weld defect detection of aviation aluminum alloy based on improved YOLOv8

1.
Key Laboratory of Nondestructive Testing，Ministry of Education，Nanchang Hangkong University，Nanchang 330063，China
2.
Centre for Inspection and Testing，Jiangxi Hongdu Aviation Industry Group Limited Liability Company，Nanchang 330024，China

摘要

摘要:
为了提高航空铝合金焊接缺陷数字射线成像自动检测效率和准确度，提出了一种改进YOLOv8智能检测方法。针对样本数据不足和缺陷不清晰的问题，采用Retinex图像增强算法和引导滤波算法对原始图像进行图像增强处理，然后采用旋转和翻转等方式扩充数据集。在模型改进中，使用GhostBottleneck模块替换C2f中的Bottleneck模块，完成模型的轻量化，减少了额外的冗余参数并降低了计算量。同时，引入空间注意力机制，获得缺陷更多的空间信息，并调整预测框的回归范围，提升了模型的精度。通过铝合金焊接件中常见几类缺陷进行测试和验证，改进YOLOv8算法平均精度均值（mAP50）达到92.9%，优于传统的Faster-RCNN、SSD和YOLOv8算法，能够有效适用于焊缝缺陷的自动识别。
- 数字射线 /
- 图像增强 /
- 自动识别 /
- YOLOv8算法 /
- 焊缝缺陷
Abstract:
In order to improve the efficiency and accuracy of automatic detection, an improved YOLOv8 detection method was proposed. Retinex image enhancement algorithm combing guided filtering was used to improve the contrast of digital radiograph images. Then, the digital radiography images was rotated and flipped to extend the data-set. In the process of model improvement, the Bottleneck module in C2f was replaced by GhostBottleneck module to reduce additional redundant parameters, so the lightweight model was acquired. In addition, spatial attention mechanism was introduced to obtain more spatial information of the defect. The regression range of the prediction box was adjusted to improve the detection accuracy of the proposed model. Several common aluminum alloy weld defects were used for experimental testing and verification. It was verified that the mAP of the improved YOLOv8 was 92.9%, which was better than Faster-RCNN, SSD and YOLOv8. The proposed model can be used for detecting the weld defect.
- digital radiography /
- image enhancement /
- automatic identification /
- YOLOv8 algorithm /
- weld defect

HTML

图 1 航空铝合金焊缝缺陷智能识别方法

Figure 1. Intelligent identification of weld defect in aviation aluminum alloy

下载: 全尺寸图片幻灯片

图 2 图像增强效果细节对比

Figure 2. Comparison of image enhancement results

下载: 全尺寸图片幻灯片

图 3 YOLOv8改进模块示意图

Figure 3. Schematic diagram of improved module in YOLOv8

下载: 全尺寸图片幻灯片

图 4 最终改进YOLOv8模型框图

Figure 4. Schematic diagram of improved YOLOv8 algorithm

下载: 全尺寸图片幻灯片

图 5 瓶颈层和幻影瓶颈层

Figure 5. Bottleneck and GhostBottleneck

下载: 全尺寸图片幻灯片

图 6 幻影卷积

Figure 6. Ghost convolution

下载: 全尺寸图片幻灯片

图 7 空间注意力机制

Figure 7. Spatial attention mechanism

下载: 全尺寸图片幻灯片

图 8 目标框与最大预测框

Figure 8. Target box and maximum prediction box

下载: 全尺寸图片幻灯片

图 9 标签框尺寸特点统计

Figure 9. Statistics of label box size characteristics

下载: 全尺寸图片幻灯片

图 10 训练过程中改进模型与YOLOv8n模型的F_1-score折线图

Figure 10. F_1-score chart in the training process of the proposed model and YOLOv8n model

下载: 全尺寸图片幻灯片

图 11 改进模型各类缺陷精确度、召回率、mAP50结果

Figure 11. Precision, recall and mAP50 result of the proposed model for defect detection

下载: 全尺寸图片幻灯片

图 12 检测效果图

Figure 12. Detection results

下载: 全尺寸图片幻灯片

表 1 训练和测试参数

Table 1. Training and testing parameters

训练设备	设备参数	训练环境	环境参数
GPU	RTX3090（24 GB）	CUDA	11.3
CPU测试	AMD R7-4800U	torch	1.11.0

下载: 导出CSV

表 2 改进模型与YOLOv8n精确率、召回率、mAP50对比

Table 2. Comparison of precision, recall, mAP50 between the proposed model and YOLOv8n

模型	精确度	召回率	mAP50
YOLOv8n	0.892	0.782	0.854
改进模型	0.916	0.877	0.929

下载: 导出CSV

表 3 改进模型与YOLOv8n模型参数量与检测时间对比

Table 3. Comparison of parameters and detection time between improved model and YOLOv8n model

模型	参数量/10⁶	每张图像检测时间/ms
YOLOv8n	3.1	324.8
改进模型	2.3	295

下载: 导出CSV

表 4 多种模型性能对比

Table 4. Performance of different models

模型	mAP50	F	参数量/10⁶	检测时间/ms	权重文件/MB
改进模型	0.929	0.896	2.3	295.0	4.73
YOLOv8n	0.854	0.833	3.1	324.8	6.29
YOLOv5s	0.846	0.844	7.0	509.4	14.1
YOLOv6-V3.0	0.854	0.850	18.5	730.3	38.7
YOLOv7-tiny	0.805	0.764	6.0	441.1	11.8
Faster-RCNN	0.794	0.8	41.36	3333.3	319.0
SSD-Lite	0.604	0.646	3.07	303.0	24.3

下载: 导出CSV

参考文献(24)

[1]	臧金鑫,陈军洲,韩凯,等. 航空铝合金研究进展与发展趋势[J]. 中国材料进展,2022,41(10): 769-777,807. ZANG Jinxin,CHEN Junzhou,HAN Kai,et al. Research progress and development tendency of aeronautical aluminum alloys[J]. Materials China,2022,41(10): 769-777,807. (in Chinese ZANG Jinxin, CHEN Junzhou, HAN Kai, et al. Research progress and development tendency of aeronautical aluminum alloys[J]. Materials China, 2022, 41(10): 769-777, 807. (in Chinese)
[2]	杜亮,闫福旭. 铝合金焊接常见缺陷及预防措施研究[J]. 中国金属通报,2021(7): 71-72. DU Liang,YAN Fuxu. Study on common defects in aluminum alloy welding and preventive measures[J]. China Metal Bulletin,2021(7): 71-72. (in Chinese DU Liang, YAN Fuxu. Study on common defects in aluminum alloy welding and preventive measures[J]. China Metal Bulletin, 2021(7): 71-72. (in Chinese)
[3]	胡文刚,陆云鹏,郭世雄,等. 基于DR数字射线成像技术的铝合金焊缝缺陷检测[J]. 焊接,2021(2): 46-51,64. HU Wengang,LU Yunpeng,GUO Shixiong,et al. Weld defect detection of aluminum alloy based on digital radiography[J]. Welding & Joining,2021(2): 46-51,64. (in Chinese HU Wengang, LU Yunpeng, GUO Shixiong, et al. Weld defect detection of aluminum alloy based on digital radiography[J]. Welding & Joining, 2021(2): 46-51, 64. (in Chinese)
[4]	邬冠华,熊鸿建. 中国射线检测技术现状及研究进展[J]. 仪器仪表学报,2016,37(8): 1683-1695. WU Guanhua,XIONG Hongjian. Radiography testing in China[J]. Chinese Journal of Scientific Instrument,2016,37(8): 1683-1695. (in Chinese WU Guanhua, XIONG Hongjian. Radiography testing in China[J]. Chinese Journal of Scientific Instrument, 2016, 37(8): 1683-1695. (in Chinese)
[5]	冯雄博,陈曦,闵慧娜,等. 基于改进CLAHE的航空发动机导向叶片DR图像增强[J]. 航空动力学报,2022,37(7): 1425-1436. FENG Xiongbo,CHEN Xi,MIN Huina,et al. DR image enhancement of aero-engine guide vane based on improved CLAHE[J]. Journal of Aerospace Power,2022,37(7): 1425-1436. (in Chinese FENG Xiongbo, CHEN Xi, MIN Huina, et al. DR image enhancement of aero-engine guide vane based on improved CLAHE[J]. Journal of Aerospace Power, 2022, 37(7): 1425-1436. (in Chinese)
[6]	丁卫良,常华峰,潘龙龙,等. X射线无损检测的应用及发展趋势[J]. 科技创新与应用,2020(36): 161-162. DING Weiliang,CHANG Huafeng,PAN Longlong,et al. Application and development trend of X-ray nondestructive testing[J]. Technology Innovation and Application,2020(36): 161-162. (in Chinese DING Weiliang, CHANG Huafeng, PAN Longlong, et al. Application and development trend of X-ray nondestructive testing[J]. Technology Innovation and Application, 2020(36): 161-162. (in Chinese)
[7]	王克广,岳乔,缪伟,等. 发动机小直径管X射线底片数字化图像增强技术[J]. 失效分析与预防,2022,17(2): 86-95. WANG Keguang,YUE Qiao,MIAO Wei,et al. Digital image enhancement on X-ray film of aeroengine small diameter tube[J]. Failure Analysis and Prevention,2022,17(2): 86-95. (in Chinese WANG Keguang, YUE Qiao, MIAO Wei, et al. Digital image enhancement on X-ray film of aeroengine small diameter tube[J]. Failure Analysis and Prevention, 2022, 17(2): 86-95. (in Chinese)
[8]	王睿,胡云雷,刘卫朋,等. 基于边缘AI的焊缝X射线图像缺陷检测[J]. 焊接学报,2022,43(1): 79-84,118. WANG Rui,HU Yunlei,LIU Weipeng,et al. Defect detection of weld X-ray image based on edge AI[J]. Transactions of the China Welding Institution,2022,43(1): 79-84,118. (in Chinese WANG Rui, HU Yunlei, LIU Weipeng, et al. Defect detection of weld X-ray image based on edge AI[J]. Transactions of the China Welding Institution, 2022, 43(1): 79-84, 118. (in Chinese)
[9]	GAO Jinling,FEZZAA K,CHEN Weinong. Multiscale dynamic experiments on fiber-reinforced composites with damage assessment using high-speed synchrotron X-ray phase-contrast imaging[J]. NDT & E International,2022,129: 102636.
[10]	GIRSHICK R,DONAHUE J,DARRELL T,et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway,US: IEEE,2014: 580-587.
[11]	ZHAO Wenqing,XU Minfu,CHENG Xingfu,et al. An insulator in transmission lines recognition and fault detection model based on improved faster RCNN[J]. IEEE Transactions on Instrumentation and Measurement,2021,70: 1-8.
[12]	蔡彪,沈宽,付金磊,等. 基于Mask R-CNN的铸件X射线DR图像缺陷检测研究[J]. 仪器仪表学报,2020,41(3): 61-69. CAI Biao,SHEN Kuan,FU Jinlei,et al. Research on defect detection of X-ray DR images of casting based on Mask R-CNN[J]. Chinese Journal of Scientific Instrument,2020,41(3): 61-69. (in Chinese CAI Biao, SHEN Kuan, FU Jinlei, et al. Research on defect detection of X-ray DR images of casting based on Mask R-CNN[J]. Chinese Journal of Scientific Instrument, 2020, 41(3): 61-69. (in Chinese)
[13]	陈乐. 基于深度学习的小径管环焊缝DR图像缺陷智能识别[J]. 化工装备技术,2022,43(4): 30-35. CHEN Le. Intelligent recognition of small diameter pipe girth weld DR image defects based on deep learning[J]. Chemical Equipment Technology,2022,43(4): 30-35. (in Chinese CHEN Le. Intelligent recognition of small diameter pipe girth weld DR image defects based on deep learning[J]. Chemical Equipment Technology, 2022, 43(4): 30-35. (in Chinese)
[14]	LIU Wei,ANGUERLOV D,ERHAN D E,et al. SSD: single shot multibox detector [C]// Proceedings of Computer Vision–ECCV 2016: 14th European Conference. Amsterdam: Springer,2016: 21-37.
[15]	李兰,奚舒舒,张才宝,等. 基于改进SSD模型的工件表面缺陷识别算法[J]. 计算机工程与科学,2020,42(9): 1608-1615. LI Lan,XI Shushu,ZHANG Caibao,et al. A surface defect recognition algorithm based on improved SSD model[J]. Computer Engineering & Science,2020,42(9): 1608-1615. (in Chinese LI Lan, XI Shushu, ZHANG Caibao, et al. A surface defect recognition algorithm based on improved SSD model[J]. Computer Engineering & Science, 2020, 42(9): 1608-1615. (in Chinese)
[16]	SHAFI I,MAZAHIR A,FATIMA A,et al. Internal defects detection and classification in hollow cylindrical surfaces using single shot detection and MobileNet[J]. Measurement,2022,202: 111836. doi: 10.1016/j.measurement.2022.111836
[17]	REDMON J,DIVVALA S,GIRSHICK R,et al. You only look once: unified,real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway,US: IEEE,2016: 779-788.
[18]	ZHANG Rui,WEN Chuanbo. 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. doi: 10.1002/adts.202100631
[19]	程松,杨洪刚,徐学谦,等. 基于YOLOv5的改进轻量型X射线铝合金焊缝缺陷检测算法[J]. 中国激光,2022,49(21): 136-144. CHENG Song,YANG Honggang,XU Xueqian,et al. Improved lightweight X-ray aluminum alloy weld defects detection algorithm based on YOLOv5[J]. Chinese Journal of Lasers,2022,49(21): 136-144. (in Chinese CHENG Song, YANG Honggang, XU Xueqian, et al. Improved lightweight X-ray aluminum alloy weld defects detection algorithm based on YOLOv5[J]. Chinese Journal of Lasers, 2022, 49(21): 136-144. (in Chinese)
[20]	赵先圣,冯鹏,沈宽,等. 基于深度学习的铁道车辆铸件X射线DR图像缺陷检测算法研究[J]. 中国体视学与图像分析,2021,26(3): 310-320. ZHAO Xiansheng,FENG Peng,SHEN Kuan,et al. Research on a defect detection algorithm for X-ray DR images of railway vehicle castings based on deep learning[J]. Chinese Journal of Stereology and Image Analysis,2021,26(3): 310-320. (in Chinese ZHAO Xiansheng, FENG Peng, SHEN Kuan, et al. Research on a defect detection algorithm for X-ray DR images of railway vehicle castings based on deep learning[J]. Chinese Journal of Stereology and Image Analysis, 2021, 26(3): 310-320. (in Chinese)
[21]	LIU Moyun,CHEN Youping,XIE Jingming,et al. LF-YOLO: a lighter and faster YOLO for weld defect detection of X-ray image[J]. IEEE Sensors Journal,2023,23(7): 7430-7439. doi: 10.1109/JSEN.2023.3247006
[22]	RUIZ-PONCE P,ORTIZ-PEREZ D,GARCIA-RODRIGUEZ J,et al. POSEIDON: a data augmentation tool for small object detection datasets in maritime environments[J]. Sensors,2023,23(7): 3691. doi: 10.3390/s23073691
[23]	周涛,杜玉虎,石道宗,等. 强化特征提取能力的下颌骨骨折检测3M-YOLOv5网络[J]. 光学精密工程,2023,31(21): 3178-3191. ZHOU Tao,DU Yuhu,SHI Daozong,et al. Mandibular fracture detection with 3M-YOLOv5 network based on enhanced feature extraction capability[J]. Optics and Precision Engineering,2023,31(21): 3178-3191. (in Chinese doi: 10.37188/OPE.20233121.3178 ZHOU Tao, DU Yuhu, SHI Daozong, et al. Mandibular fracture detection with 3M-YOLOv5 network based on enhanced feature extraction capability[J]. Optics and Precision Engineering, 2023, 31(21): 3178-3191. (in Chinese) doi: 10.37188/OPE.20233121.3178
[24]	夏烨,雷哓晖,祁雁楠,等. 基于改进Ghost-YOLOv5s-BiFPN算法检测梨树花序[J]. 智慧农业(中英文),2022,4(3): 108-119. XIA Ye,LEI Xiaohui,QI Yannan,et al. Detection of pear inflorescence based on improved ghost-YOLOv5s-BiFPN algorithm[J]. Smart Agriculture,2022,4(3): 108-119. (in Chinese XIA Ye, LEI Xiaohui, QI Yannan, et al. Detection of pear inflorescence based on improved ghost-YOLOv5s-BiFPN algorithm[J]. Smart Agriculture, 2022, 4(3): 108-119. (in Chinese)