基于MCB-FAH-YOLOv8的钢材表面缺陷检测算法

doi:10.11996/JG.j.2095-302X.2024010112

摘要/Abstract

摘要：

针对现有基于深度学习的钢材表面缺陷检测算法存在误检、漏检和检测精度低等问题，提出一种基于改进CBAM(modified CBAM，MCB)和可替换四头ASFF预测头(four-head ASFF prediction head，FAH)的YOLOv8钢材表面缺陷检测算法，简记为MCB-FAH-YOLOv8。通过加入改进后的卷积注意力机制模块(CBAM)对密集目标更好的确定；通过将FPN结构改为BiFPN更加高效的提取上下文信息；通过增加自适应特征融合(ASFF)自动找出最适合的融合特征；通过将SPPF模块替换为精度更高的SimCSPSPPF模块。同时，针对微小物体检测，提出了四头ASFF预测头，可根据数据集特点进行替换。实验结果表明，MCB-FAH-YOLOv8算法在VOC2007数据集上检测精度(mAP)达到了88.8%，在NEU-DET钢铁缺陷检测数据集上检测精度(mAP)达到了81.8%，较基准模型分别提高了5.1%和3.4%，该算法在牺牲较少检测速度的情况下取得较高的检测精度，很好的平衡了算法的精度和速度。

华北电力大学崔克彬博士等针对实际生产中钢铁缺陷检测存在误检、漏检等问题，提出了基于改进YOLOv8n的钢铁缺陷检测算法MCB-FAH-YOLOv8。通过加入改进后的CBAM注意力机制模块对密集目标更好的确定；通过将FPN结构改为BiFPN更加高效的提取上下文信息；通过增加ASFF自动找出最适合的融合特征；通过将SPPF模块替换为精度更高的SimCSPSPPF模块。同时，针对微小物体检测，提出了四头ASFF预测头，可根据数据集特点进行替换。

关键词: MCB-FAH-YOLOv8, 缺陷检测, 注意力机制, 四头ASFF预测头, 特征融合

Abstract:

To address the problems of misdetection, omission, and low detection accuracy in existing deep learning-based algorithms for detecting defects on steel surfaces, a YOLOv8 steel surface defect detection algorithm was proposed based on a modified CBAM (MCB) and replaceable four-head ASFF prediction head (FAH), abbreviated as MCB-FAH-YOLOv8. By integrating the modified convolutional attention mechanism module (CBAM), the algorithm could achieve better determination of the densely populated targets. By changing the FPN structure to BiFPN, it could extract context information more efficiently. It also incorporated adaptive feature fusion (ASFF) for the automatic identification of the most suitable fusion features. The algorithm also boosted its precision by replacing the SPPF module with the SimCSPSPPF module. Meanwhile, for tiny object detection, a four-head ASFF prediction head was proposed, designed to be replaceable based on the dataset characteristics. The experimental results demonstrated that the MCB-FAH-YOLOv8 algorithm could achieve a detection accuracy (mAP) of 88.8% on the VOC2007 dataset and 81.8% on the NEU-DET steel defect detection dataset, outperforming the benchmark model by 5.1% and 3.4%, respectively. This new algorithm achieved a higher detection accuracy with less loss of detection speed, thus ensuring a good balance between accuracy and speed.

Key words: MCB-FAH-YOLOv8, defect detection, attention mechanism, four-head ASFF prediction head, feature fusion

中图分类号:

TP391

崔克彬, 焦静颐. 基于MCB-FAH-YOLOv8的钢材表面缺陷检测算法[J]. 图学学报, 2024, 45(1): 112-125.

CUI Kebin, JIAO Jingyi. Steel surface defect detection algorithm based on MCB-FAH-YOLOv8[J]. Journal of Graphics, 2024, 45(1): 112-125.

图/表 28

图1 MCB-FAH-YOLOv8网络结构

Fig. 1 MCB-FAH-YOLOv8 network structure

图2 MCB模块

Fig. 2 MCB module

图3 SimCSPSPPF模块

Fig. 3 SimCSPSPPF module

图4 BiFPN

Fig. 4 BiFPN

图5 ASFF

Fig. 5 ASFF

图6 改进后的算法结构

Fig. 6 Structure of the improved algorithm

图7 部分带标签的VOC2007数据集

Fig. 7 Partially labeled VOC2007 dataset

图8 部分带标签的NEU-DET数据集

Fig. 8 Partially labeled NEU-DET dataset

表1 加入改进CBAM模块在VOC2007数据集实验

Table 1 Incorporating Improved CBAM Module in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n	5.96	3.01	8.9	83.7	116
YOLOv8n-CA	6.34	3.06	8.3	84.2	103
YOLOv8n-CBAM	6.81	3.29	8.5	85.5	93
YOLOv8n-MCB	9.07	4.41	9.6	86.5	101

表2 加入BiFPN模块在VOC2007数据集实验

Table 2 Incorporate improve the BiFPN module in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n-MCB	9.07	4.41	9.6	86.5	101
YOLOv8n-MCB-B2	9.07	4.41	9.6	87.0	94
YOLOv8n-MCB-B3	9.10	4.43	9.7	87.2	97

表3 加入ASFF模块在VOC2007数据集实验

Table 3 Incorporate the ASFF module in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n-MCB-B3	9.10	4.43	9.7	87.2	97
YOLOv8n-MCB-BA	11.70	5.73	11.8	88.5	81

表4 改为四头预测头在VOC2007数据集实验

Table 4 Changed to four-head prediction head in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n-MCB-BA	11.7	5.73	11.8	86.1	81
YOLOv8n-MCB-FAH_1	12.0	5.82	17.2	86.8	64

图9 使用四头ASFF预测头对比图

Fig. 9 Four-head ASFF prediction head comparison chart

表5 YOLOv3.0更改为SimCSPSPFF模块在VOC2007数据集实验

Table 5 YOLOv3.0 changed to SimCSPSPFF module in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n-MCB-BA	11.7	5.73	11.8	88.5	81
MCB-FAH-YOLOv8	12.4	6.06	12.1	88.8	80

表6 YOLOv5.0更改为SimCSPSPFF模块在VOC2007数据集实验

Table 6 YOLOv5.0 changed to SimCSPSPFF module in VOC2007 Dataset Experiments

算法模型	体积/MB	参数量/M	计算量/GFLOPs	mAP@0.5:0.95/%	FPS
YOLOv8n-MCB-FAH_1	12.0	5.82	17.2	86.8	64
MCB-FAH-YOLOv8_FD	12.7	6.15	17.4	87.0	60

图10 YOLOv8n算法F1分数曲线

Fig. 10 YOLOv8n algorithm F1 score curve

图11 MCB-FAH-YOLOv8n算法F1分数曲线

Fig. 11 MCB-FAH-YOLOv8n algorithm F1 score curve

图12 MCB-FAH-YOLOv8n_FD算法F1分数曲线

Fig. 12 MCB-FAH-YOLOv8n_FD algorithm F1 score curve

表7 NEU-DET数据集实验对比

Table 7 Experimental comparison of NEU-DET dataset

算法模型	体积/MB	参数量/M	mAP@0.5:0.95/%	FPS
YOLOv8n	5.96	3.01	78.4	116
MCB-FAH-YOLOv8	12.40	6.06	81.8	80

表8 NEU-DET上检测效果

Table 8 Detection of effects on NEU-DET

算法模型	mAP@0.5/%	Sc	Pa	Cr	In	Ps	Rs
Faster-RCNN	77.5	96.1	93.9	33.5	86.3	90.0	65.4
SSD	74.7	73.4	95.4	44.4	82.7	88.7	63.7
YOLOv3	73.4	82.0	91.3	37.5	74.3	84.8	70.2
YOLOv4	74.6	90.3	93.7	37.9	84.0	76.8	64.9
YOLOv5	75.1	89.9	94.0	38.4	81.7	82.6	63.8
YOLOv8n	97.7	98.8	99.4	94.4	98.8	96.8	98.3
MCB-FAH-YOLOv8	98.6	99.2	99.5	97.3	98.7	97.7	99.2

表9 加入改进CBAM模块在NEU-DET数据集实验

Table 9 Incorporating Improved CBAM Module in NEU-DET Dataset Experiments

算法模型	Sc	Pa	Cr	In	Ps	Rs	mAP@0.5:0.95/%
YOLOv8n	79.4	87.1	70.4	74.8	79.5	79.4	78.4
YOLOv8n-CBAM	79.4	86.9	70.5	75.2	80.2	79.4	78.5
YOLOv8n-MCB	78.5	87.5	72.2	75.2	81.1	80.7	79.2

表10 YOLOv8n-MCB更改为SimCSPSPFF模块在NEU-DET数据集实验

Table 10 YOLOv8n-MCB changed to SimCSPSPFF module in NEU-DET Dataset Experiments

算法模型	Sc	Pa	Cr	In	Ps	Rs	mAP@0.5:0.95/%
YOLOv8n-MCB	78.5	87.5	72.2	75.2	81.1	80.7	79.2
YOLOv8n-MCB-Sim	80.6	88.6	78.4	77.8	81.9	84.8	82.0

表11 加入BiFPN模块在NEU-DET数据集实验

Table 11 Incorporate improve the BiFPN module in NEU-DET Dataset Experiments

算法模型	Sc	Pa	Cr	In	Ps	Rs	mAP@0.5:0.95/%
YOLOv8n-MCB-Sim	80.6	88.6	78.4	77.8	81.9	84.8	82.0
MCB-FAH-YOLOv8	79.7	88.5	77.2	77.9	83.3	84.3	81.8

图13 YOLOv8n算法P-R曲线

Fig. 13 YOLOv8n algorithm P-R curve

图14 MCB-FAH-YOLOv8算法P-R曲线

Fig. 14 MCB-FAH-YOLOv8 algorithm P-R curve

图15 YOLOv8n混淆矩阵

Fig. 15 YOLOv8n confusion matrix

图16 MCB-FAH-YOLOv8混淆矩阵

Fig. 16 MCB-FAH-YOLOv8 confusion matrix

图17 NEU-DET上检测结果((a)原始图片；(b) YOLOv8n；(c) MCB-FAH-YOLOv8)

Fig. 17 Results on NEU-DET ((a) Original image; (b) YOLOv8n; (c) MCB-FAH-YOLOv8)

参考文献 26

[1]	张艳, 冯锋. 带钢表面缺陷检测技术探析[J]. 信息与电脑: 理论版, 2021, 33(11): 19-22.
	ZHANG Y, FENG F. Analysis of strip surface defect detection technology[J]. China Computer & Communication, 2021, 33(11): 19-22 (in Chinese).
[2]	李维刚, 叶欣, 赵云涛, 等. 基于改进YOLOv3算法的带钢表面缺陷检测[J]. 电子学报, 2020, 48(7): 1284-1292. DOI
	LI W G, YE X, ZHAO Y T, et al. Strip steel surface defect detection based on improved YOLOv3 algorithm[J]. Acta Electronica Sinica, 2020, 48(7): 1284-1292 (in Chinese).
[3]	梁日强, 胡燕林, 蒋占四. 基于改进的残差收缩网络的带钢表面缺陷识别[J]. 组合机床与自动化加工技术, 2022(6): 82-85. DOI
	LIANG R Q, HU Y L, JIANG Z S. Strip surface defect identification based on improved residual shrinkage network[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2022(6): 82-85 (in Chinese).
[4]	彭晏飞, 袁晓龙, 陈炎康, 等. 改进YOLOv5s的带钢表面缺陷检测方法[J/OL]. 机械科学与技术(2023-06-02) [2023-07- 01]. https://kns.cnki.net/kcms2/article/abstract?v=sf24_f5fySZD1- Ycih16WJXSfIVveZZx-CIfY7zktvjqzQn6B0DTDcZgNVLA0NIuChhowEzC5HzZY9pPVuTLKBnfsQHrTmDr0UAjhaYBQI51aIHBb882L1pGFuSBOaqzFrS6UO45b_s=&uniplatform=NZKPT&language=CHS.
	PENG Y F, YUAN X L, CHEN Y K, et al. Improved YOLOv5s strip surface defect detection method[J/OL]. Mechanical Science and Technology for Aerospace Engineering (2023-06-02) [2023-07-01]. https://kns.cnki.net/kcms2/article/abstract?v=sf24_f5fySZD1- Ycih16WJXSfIVveZZx-CIfY7zktvjqzQn6B0DTDcZgNVLA0NIuChhowEzC5HzZY9pPVuTLKBnfsQHrTmDr0UAjhaYBQI51aIHBb882L1pGFuSBOaqzFrS6UO45b_s=&uniplatform=NZKPT&language=CHS. (in Chinese).
[5]	唐东林, 杨洲, 程衡, 等. 浅层卷积神经网络融合Transformer的金属缺陷图像识别方法[J]. 中国机械工程, 2022, 33(19): 2298-2305, 2316.
	TANG D L, YANG Z, CHENG H, et al. Metal defect image recognition method based on shallow CNN fusion transformer[J]. China Mechanical Engineering, 2022, 33(19): 2298-2305, 2316 (in Chinese). DOI
[6]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]// European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[7]	LI C Y, LI L L, GENG Y F, et al. YOLOv6 v3.0: a full-scale reloading[EB/OL]. [2023-06-29]. https://arxiv.org/abs/2301.05586.pdf.
[8]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 936-944.
[9]	TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 10781-10790.
[10]	LIU S T, HUANG D, WANG Y H. Learning spatial fusion for single-shot object detection[EB/OL]. [2023-06-29]. https://arxiv.org/abs/1911.09516.pdf.
[11]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2023-06-29]. https://arxiv.org/abs/2004.10934.pdf.
[12]	ZHANG H Y, CISSE M, DAUPHIN Y N, et al. Mixup: beyond empirical risk minimization[EB/OL]. [2023-06-29]. https://arxiv.org/abs/1710.09412.pdf.
[13]	YUN S, HAN D, CHUN S, et al. CutMix: regularization strategy to train strong classifiers with localizable features[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2020: 6022-6031.
[14]	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. New York: IEEE Press, 2014: 580-587.
[15]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[C]// European Conference on Computer Vision. Cham: Springer, 2014: 346-361.
[16]	REN S Q, HE K M, 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, 2017, 39(6): 1137-1149. DOI PMID
[17]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[M]//Computer Vision - ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[18]	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. New York: IEEE Press, 2016: 779-788.
[19]	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 6517-6525.
[20]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. [2023-06-29]. https://arxiv.org/abs/1804.02767.pdf.
[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. New York: IEEE Press, 2022: 9992-10002.
[22]	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16 x16 words: transformers for image recognition at scale[EB/OL]. [2023-06-29]. https://arxiv.org/abs/2010.11929.pdf.
[23]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 7132-7141.
[24]	FU J, LIU J, TIAN H J, et al. Dual attention network for scene segmentation[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 3141-3149.
[25]	GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. [2023-06-29]. https://arxiv.org/abs/2107.08430.pdf.
[26]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[EB/OL]. [2023-06-29]. https://arxiv.org/abs/2207.02696.pdf.