基于改进YOLOv8的飞机蒙皮缺陷检测算法

章东平; 王杼涛; 夏岳键; 徐云超; 林丽莉

doi:10.13700/j.bh.1001-5965.2023.0744

基于改进YOLOv8的飞机蒙皮缺陷检测算法

doi: 10.13700/j.bh.1001-5965.2023.0744

1.
中国计量大学信息工程学院，杭州 310018
2.
浙江工商大学信息与电子工程学院，杭州 310018

基金项目:

浙江省重点研发计划(2022C01005,2023C01032,2023C01030)

详细信息

通讯作者:
E-mail：06a0303103@cjlu.edu.cn

中图分类号: V223；TP181
计量
- 文章访问数: 1123
- HTML全文浏览量: 310
- PDF下载量: 82
- 被引次数: 0
出版历程
- 收稿日期: 2023-11-20
- 录用日期: 2023-12-01
- 网络出版日期: 2024-01-30
- 整期出版日期: 2026-01-15

Aircraft skin defect detection algorithm based on enhanced YOLOv8

1.
College of Information Engineering，China Jiliang University，Hangzhou 310018，China
2.
School of Information and Electronic Engineering，Zhejiang Gongshang University，Hangzhou 310018，China

Funds:

Zhejiang Key R & D Project of China (2022C01005,2023C01032,2023C01030)

More Information

Corresponding author: E-mail：06a0303103@cjlu.edu.cn

摘要

摘要:
为解决传统飞机蒙皮缺陷检测依靠人眼观察时，因人眼容易疲劳和个体认知有限导致效率降低的问题，提出一种基于改进YOLOv8的飞机蒙皮缺陷检测算法。对数据增强方式进行改进，提出一种切片推理+马赛克的数据增强方法；集成残差块到特征提取网络，增强网络表达能力的同时，提高模型在飞机蒙皮缺陷检测任务中的精度；应用三分支注意力模块改进特征融合网络，减少小目标样本的误检率和漏检率；优化检测头结构，使网络能够更好地将浅层信息与深度信息有效结合。实验结果表明：相比于YOLOv8算法，改进算法在飞机蒙皮缺陷数据集上的平均精度均值(mAP)和查全率分别提高了3.6%和3.7%，在公开数据集VOC2007上的平均精度均值和查全率提高了2.9%和2.2%。
- YOLOv8算法 /
- 表面缺陷检测 /
- 数据增强 /
- 目标检测 /
- 注意力机制
Abstract:
In order to solve the problem that traditional aircraft skin defect detection relies on human eye observation, which leads to reduced efficiency due to easy fatigue of the human eye and limited individual cognition, an aircraft skin defect detection algorithm based on improved YOLOv8 is proposed. Improve the data improvement strategy and propose a new one that combines slice reasoning with mosaic. Integrate the residual block into the feature extraction network to enhance the network expression ability and improve the accuracy of the model in aircraft skin defect detection tasks. Use the triplet attention module to strengthen the feature fusion network and lower the false and missed detection rates of small target samples. Optimize the structure of the detection head so that the network can better effectively combine shallow information with depth information. On the aircraft skin defect data set, experimental results indicate that the revised algorithm’s mean average precision (mAP) and recall rate have increased by 3.6% and 3.7%, respectively, in comparison to the most recent YOLOv8 algorithm. The mAP and recall rate on the public data set VOC2007 increased by 2.9% and 2.2%, respectively.
- YOLOv8 algorithm /
- surface defect detection /
- data augmentation /
- object detection /
- attention mechanism

HTML全文

图 1 改进的数据增强流程

Figure 1. Flow of improved data augmentation

下载: 全尺寸图片幻灯片

图 2 不同数据增强方法对比

Figure 2. Comparison of different data augmentation methods

下载: 全尺寸图片幻灯片

图 3 三分支注意力模块结构

Figure 3. Structure of triplet attention module

下载: 全尺寸图片幻灯片

图 4 第1分支的网络结构

Figure 4. Network structure of the first branch

下载: 全尺寸图片幻灯片

图 5 第2分支的网络结构

Figure 5. Network structure of the second branch

下载: 全尺寸图片幻灯片

图 6 第3分支的网络结构

Figure 6. Network structure of the third branch

下载: 全尺寸图片幻灯片

图 7 残差块的网络结构

Figure 7. Network structure of residual block

下载: 全尺寸图片幻灯片

图 8 改进的检测头网络结构

Figure 8. Network structure of improved detection head

下载: 全尺寸图片幻灯片

图 9 改进的 YOLOv8 网络结构

Figure 9. Structure of improved YOLOv8 network

下载: 全尺寸图片幻灯片

图 10 巡检机器人巡检示意图

Figure 10. Schematic of inspection by inspection robot

下载: 全尺寸图片幻灯片

图 11 飞机巡检路线图

Figure 11. Aircraft inspection route map

下载: 全尺寸图片幻灯片

图 12 各类缺陷样本的示意图

Figure 12. Schematic diagram of various defect samples

下载: 全尺寸图片幻灯片

图 13 改进模型在飞机蒙皮缺陷数据集上的准确率-召回率曲线

Figure 13. Precision-recall curves of improved model on aircraft skin defect dataset

下载: 全尺寸图片幻灯片

图 14 VOC2007数据集平均精度均值对比

Figure 14. Comparison of mAP on VOC2007 dataset

下载: 全尺寸图片幻灯片

表 1 添加不同注意力算法的结果比较

Table 1. Comparison of results in different attention methods %

算法	查全率	mAP0.5
原始的YOLOv8算法	83.2	85.1
改进算法(+Bi-Level Routing^[29])	83.2	85.1
改进算法(+CBAM)	83.9	85.5
改进算法(+SENet)	83.5	86.0
改进算法(+SimAM^[28])	84.2	86.0
改进算法(+三分支注意力)	84.4	86.0

下载: 导出CSV

表 2 添加注意力模块不同位置和数量的结果比较

Table 2. Comparison of results at different positions and quantities for attention module %

添加数量	添加位置	查全率	mAP0.5	mAP0.5:0.95
0	基线模型	83.2	85.1	43.3
1	第3层	79.0	82.9	39.9
1	第4层	82.8	85.7	41.5
1	第5层	84.4	86.0	42.4
2	第3、4层	80.4	85.3	40.5
2	第3、5层	80.6	83.8	40.6
2	第4、5层	82.4	85.3	41.6
3	第3、4、5层	81.6	85.4	41.2

下载: 导出CSV

表 3 添加不同主干网络的结果比较

Table 3. Comparison of results with different backbone networks %

算法	查全率	mAP0.5	mAP0.5:0.95
原始的YOLOv8算法	83.2	85.1	43.3
改进算法(+EfficientViT^[30])	79.9	82.0	36.0
改进算法(+Fasternet^[31])	79.8	80.8	38.4
改进算法(+残差块)	85.7	87.4	48.2

下载: 导出CSV

表 4 不同检测头数量的结果比较

Table 4. Comparison of results with different numbers of head counts

数据集	方法	查全率/%	mAP0.5/%	检测速度/ (帧·s⁻¹)
飞机蒙皮缺陷数据集	基线模型	83.2	85.1	169
	3个改进检测头	83.2	86.2	136
	4个改进检测头	82.5	83.6	90
VOC2007 数据集	基线模型	71.9	77.5	312
	3个改进检测头	72.0	78.0	222
	4个改进检测头	71.7	77.8	126

下载: 导出CSV

表 5 消融实验结果比较

Table 5. Comparison of results of ablation experiment %

残差块	三分支注意力模块	检测头	查全率	mAP0.5	mAP0.5:0.95
			83.2	85.1	43.3
√			85.7	87.4	48.2
	√		84.4	86.0	42.4
		√	83.2	86.2	44.3
√	√		85.7	87.6	48.5
√		√	84.6	86.6	47.3
	√	√	83.4	87.3	45.1
√	√	√	86.9	88.7	49.0

下载: 导出CSV

表 6 与其他算法的结果比较

Table 6. Comparison of results with other algorithms

数据集	模型	查全率/ %	mAP0.5/ %	mAP0.5:0.95/ %	检测速度/ (帧·s⁻¹)
飞机蒙皮缺陷数据集	YOLOv3-tiny^[32]	73.0	75.0	32.8	300
	YOLOv5-s	84.8	80.0	43.0	240
	YOLOv6^[33]	80.0	82.6	37.8	243
	YOLOv7-tiny^[34]	74.4	76.5	30.5	260
	YOLOv8-n	83.2	85.1	43.3	135
	YOLOv8-s	86.5	88.2	48.6	100
	RTD-YOLOv8	86.9	88.7	49.0	60
VOC2007 数据集	YOLOv3-tiny^[32]	55.8	56.8	28.3	322
	YOLOv5-s	71.9	77.0	53.0	263
	YOLOv6^[33]	67.8	70.3	46.8	270
	YOLOv7-tiny^[34]	54.8	54.8	29.3	357.1
	YOLOv8-n	71.9	77.5	55.8	312.5
	YOLOv8-s	73.4	79.9	57.9	196
	RTD-YOLOv8	74.1	80.4	58.0	167

下载: 导出CSV

参考文献(34)

[1]	李婉婷. 基于关键点的飞机表面伤痕检测及后处理算法研究[D]. 北京: 北京邮电大学, 2021. LI W T. Research on aircraft surface damage detection and post-processing algorithm based on key points[D]. Beijing: Beijing University of Posts and Telecommunications, 2021(in Chinese).
[2]	张研, 田军委, 刘青, 等. 基于视觉显著性的飞机蒙皮缺陷检测方法[J]. 激光杂志, 2022, 43(3): 61-64. ZHANG Y, TIAN J W, LIU Q, et al. Aircraft skin defect detection method based on visual saliency[J]. Laser Journal, 2022, 43(3): 61-64(in Chinese).
[3]	TAO X, ZHANG D P, MA W Z, et al. Automatic metallic surface defect detection and recognition with convolutional neural networks[J]. Applied Sciences, 2018, 8(9): 1575.
[4]	GAROUSI V, TASLI S, SERTEL O, et al. Automated testing of simulation software in the aviation industry: an experience report[J]. IEEE Software, 2018, 36(4): 63-75.
[5]	HAGGERTY J, DEFER E, DE LAAT A, et al. Detecting clouds associated with jet engine ice crystal icing[J]. Bulletin of the American Meteorological Society, 2019, 100(1): 31-40.
[6]	SHANG J Z, SATTAR T, CHEN S W, et al. Design of a climbing robot for inspecting aircraft wings and fuselage[J]. Industrial Robot, 2007, 34(6): 495-502.
[7]	DONG X H, TAYLOR C J, COOTES T F. Automatic aerospace weld inspection using unsupervised local deep feature learning[J]. Knowledge-Based Systems, 2021, 221: 106892.
[8]	MEISTER S, WERMES M A M, STÜVE J, et al. Review of image segmentation techniques for layup defect detection in the automated fiber placement process[J]. Journal of Intelligent Manufacturing, 2021, 32(8): 2099-2119.
[9]	TABERNIK D, ŠELA S, SKVARČ J, et al. Segmentation-based deep-learning approach for surface-defect detection[J]. Journal of Intelligent Manufacturing, 2020, 31(3): 759-776.
[10]	赵鹏飞, 谢林柏, 彭力. 融合注意力机制的深层次小目标检测算法[J]. 计算机科学与探索, 2022, 16(4): 927-937. ZHAO P F, XIE L B, PENG L. Deep small object detection algorithm integrating attention mechanism[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16(4): 927-937(in Chinese).
[11]	张宝朋, 康谦泽, 李佳萌, 等. 轻量化的YOLOv4目标检测算法[J]. 计算机工程, 2022, 48(8): 206-214. ZHANG B P, KANG Q Z, LI J M, et al. Lightweight YOLOv4 target detection algorithm[J]. Computer Engineering, 2022, 48(8): 206-214(in Chinese).
[12]	张海涛, 张梦. 引入通道注意力机制的SSD目标检测算法[J]. 计算机工程, 2020, 46(8): 264-270. ZHANG H T, ZHANG M. SSD target detection algorithm with channel attention mechanism[J]. Computer Engineering, 2020, 46(8): 264-270(in Chinese).
[13]	候瑞环, 杨喜旺, 王智超, 等. 一种基于YOLOv4-TIA的林业害虫实时检测方法[J]. 计算机工程, 2022, 48(4): 255-261. HOU R H, YANG X W, WANG Z C, et al. A real-time detection method for forestry pests based on YOLOv4-TIA[J]. Computer Engineering, 2022, 48(4): 255-261(in Chinese).
[14]	蒋镕圻, 彭月平, 谢文宣, 等. 嵌入scSE模块的改进YOLOv4小目标检测算法[J]. 图学学报, 2021, 42(4): 546-555. JIANG R Q, PENG Y P, XIE W X, et al. Improved YOLOv4 small target detection algorithm with embedded scSE module[J]. Journal of Graphics, 2021, 42(4): 546-555(in Chinese).
[15]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. (2020-04-23)[2023-11-01]. https://arxiv.org/abs/2004.10934.
[16]	赵恩兴, 王超. 基于YOLOv8的交通信号灯识别[J]. 人工智能与机器人研究, 2023, 12(3): 246-254. ZHAO E X, WANG C. Traffic signal recognition based on YOLOv8[J]. Research in Artificial Intelligence and Robotics, 2023, 12(3): 246-254(in Chinese).
[17]	LIU H Y, DUAN X H, CHEN H N, et al. DBF-YOLO: UAV small targets detection based on shallow feature fusion[J]. IEEJ Transactions on Electrical and Electronic Engineering, 2023, 18(4): 605-612.
[18]	LIU H Y, SUN F Q, GU J, et al. SF-YOLOv5: a lightweight small object detection algorithm based on improved feature fusion mode[J]. Sensors, 2022, 22(15): 5817.
[19]	旷可嘉. 深度学习及其在航空发动机缺陷检测中的应用研究[D]. 广州: 华南理工大学, 2017. KUANG K J. Deep learning and its application in aero-engine defect detection[D]. Guangzhou: South China University of Technology, 2017(in Chinese).
[20]	ZHANG D Y, WEI P H, TAN M W, et al. Investigation of aircraft surface defects detection based on YOLO neural network[C]//Proceedings of the 7th International Conference on Information Science and Control Engineering. Piscataway: IEEE Press, 2020: 781-785.
[21]	LI Y D, HAN Z Q, XU H Y, et al. YOLOv3-lite: a lightweight crack detection network for aircraft structure based on depthwise separable convolutions[J]. Applied Sciences, 2019, 9(18): 3781.
[22]	JOVANCEVIC I, LARNIER S, ORTEU J J, et al. Automated exterior inspection of an aircraft with a pan-tilt-zoom camera mounted on a mobile robot[J]. Journal of Electronic Imaging, 2015, 24(6): 061110.
[23]	GAO Q J, WANG W J, NIU G C. Design bionic structure and analysis of kinematics for aircraft fuel tank inspection robot[J]. Applied Mechanics and Materials, 2013, 278-280: 594-598.
[24]	AKYON F C, ALTINUC S O, TEMIZEL A. Slicing aided hyper inference and fine-tuning for small object detection[C]//Proceedings of the IEEE International Conference on Image Processing. Piscataway: IEEE Press, 2022: 966-970.
[25]	ALIN A Y, YUANA K A. Data augmentation method on drone object detection with YOLOv5 algorithm[C]//Proceedings of the 8th International Conference on Informatics and Computing. Piscataway: IEEE Press, 2023: 1-6.
[26]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision. Berlin: Springer, 2018: 3-19.
[27]	DAI X Y, CHEN Y P, XIAO B, et al. Dynamic head: unifying object detection heads with attentions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2021: 7369-7378.
[28]	YANG L X, ZHANG R Y, LI L D, et al. SimAM: a simple, parameter-free attention module for convolutional neural network[C]// Proceedings of the International Conference on Machine Learning. San Diego: JMLR, 2021, 139: 11863-11874.
[29]	ZHU L, WANG X J, KE Z H, et al. BiFormer: vision Transformer with bi-level routing attention[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2023: 10323-10333.
[30]	CAI H, LI J Y, HU M Y, et al. EfficientViT: multi-scale linear attention for high-resolution dense prediction [EB/OL]. (2022-05-29)[2023-11-01]. https://arxiv.org/abs/2205.14756.
[31]	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 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2023: 12021-12031.
[32]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2018-04-08)[2023-11-01]. https://arxiv.org/abs/1804.02767.
[33]	LI C, LI L, JIANG H, et al. YOLOV6: a single-stage object detection framework for industrial applications[EB/OL]. (2022-09-07)[2023-11-01]. https://arxiv.org/abs/2209.02976.
[34]	WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2023: 7464-7475.