基于改进YOLOv8s的PCB小目标缺陷检测模型

doi:10.3778/j.issn.1002-8331.2406-0020

摘要/Abstract

摘要： 针对目前PCB缺陷检测中面临缺陷的形态复杂、目标小，难以对其进行精确的捕捉，导致识别检出率、准确率低等问题，提出了基于改进YOLOv8s的PCB小目标缺陷检测算法优化。提出增设小目标检测层，同时添加小物体检测头提高在小目标情况下的检测效果；在骨干网络中引入可选择空洞卷积（S-Conv）与CAFM（context-aware feature modulation）卷积和注意力融合模块，扩大感受野，在提升特征表示能力的同时增强对各尺度特征进行融合；使用可变形卷积和空间信息增强模块设计更加灵活和有效的空间金字塔池化层，提高模型对目标特征的表征能力和检测精度；融合信息聚集-分发机制对颈部结构进行改进。改进损失函数，以VFWD-CIoU代替原损失函数，提升密集小目标检测。改进后的模型算法在四张拼凑的PCB小目标数据集上进行相关对比实验。结果表明，改进后算法模型的平均精度（mAP）为99.1%。相比于Faster R-CNN、YOLOv5、YOLOv7等网络模型，检测精度得到很大的提升，表明该算法可以运用于实际生产环境中的PCB小目标缺陷检测。

关键词: PCB, 小目标缺陷检测, YOLOv8s, 特征融合, 注意力机制, 损失函数

Abstract: In response to the complex morphology and small targets of defects in current PCB defect detection, which make it difficult to accurately capture them, resulting in low recognition and detection rates, this paper proposes an optimized PCB small target defect detection algorithm based on improved YOLOv8s. Firstly, a small object detection layer and a small object detection head is added to improve the detection performance in the case of small targets. Secondly, optional dilated convolution (S-Conv), CAFM (context aware feature modulation) convolution and attention fusion modules are introduced into the backbone network to expand the receptive field, enhance feature representation ability, and enhance the fusion of features at various scales. Using deformable convolution and spatial information enhancement modules, more flexible and effective spatial pyramid pooling layers are designed to improve the representation ability and detection accuracy of target features of the model. This paper introduces an information aggregation distribution mechanism to improve the neck structure. Finally, the loss function is improved by replacing the original loss function with VFWD-CIoU to enhance the detection of dense small targets. The improved model algorithm is used for comparative experiments on four pieced PCB small target datasets. The results show that the average accuracy (mAP) of the improved algorithm model is 99.1%. Compared with Faster R-CNN, YOLOv5, and YOLOv7 network models, the detection accuracy has been greatly improved, indicating that this algorithm can be applied to PCB small target defect detection in actual production environments.

Key words: PCB, small target defect detection, YOLOv8s, feature fusion, attention mechanism, loss function

王健, 肖迪, 冯李航, 沈成. 基于改进YOLOv8s的PCB小目标缺陷检测模型[J]. 计算机工程与应用, 2025, 61(15): 288-297.

WANG Jian, XIAO Di, FENG Lihang, SHEN Cheng. PCB Small Object Defect Detection Model Based on Improved YOLOv8s[J]. Computer Engineering and Applications, 2025, 61(15): 288-297.

参考文献

[1] 李娟. 基于机器学习的PCB缺陷检测与分类方法研究[J]. 印制电路信息, 2024, 32(3): 57-59.
LI J. Research on PCB defect detection and classification based on machine learning[J]. Printed Circuit Information, 2024, 32(3): 57-59.
[2] 程立英, 张文雅, 程强, 等. 基于深度学习的PCB缺陷检测技术[J]. 沈阳师范大学学报(自然科学版), 2024, 42(2): 151-156.
CHENG L Y, ZHANG W Y, CHENG Q, et al. PCB defect detection technology based on deep learning[J]. Journal of Shenyang Normal University (Natural Science Edition), 2024, 42(2): 151-156.
[3] 瞿栋, 汪鹏宇, 黄允, 等. 基于卷积神经网络的PCB缺陷图像识别[J]. 计量与测量技术, 2021, 48(8): 21-23.
QU D, WANG P Y, HUANG Y, et al. Image recognition of PCB defects based on convolutional neural network PCB defect image recognition based on convolutional neural network[J]. Metrology & Measurement Technique, 2021, 48(8): 21-23.
[4] BETTI SORBELLI F, PALAZZETTI L, PINOTTI C M. YOLO-based detection of Halyomorpha halys in orchards using RGB cameras and drones[J]. Computers and Electronics in Agriculture, 2023, 213: 108228.
[5] REDMON J, FARHADI A. YOLOv3: an incremental improvement[J]. arXiv:1804.02767, 2018.
[6] GIRSHICK R. Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2015: 1440-1448.
[7] 耿振宇. 基于Faster R-CNN的PCB表面缺陷检测[J]. 科学技术创新, 2021(11): 86-87.
GENG Z Y. PCB surface defect detection based on Faster R-CNN[J]. Scientific and Technological Innovation, 2021(11): 86-87.
[8] 程霖, 张涛, 葛平淑, 等. 基于改进YOLOv5的PCB裸板缺陷检测[J]. 大连民族大学学报, 2023, 25(5): 403-407.
CHENG L, ZHANG T, GE P S, et al. PCB bare board defect detection based on improved YOLOv5[J]. Journal of Dalian Minzu University, 2023, 25(5): 403-407.
[9] 严舒, 郭颖, 黄骏. FCM-YOLO: 一种基于特征增强和多尺度融合的PCB缺陷检测方法[J]. 控制与决策, 2024, 39(10): 3181-3189.
YAN S, GUO Y, HUANG J. FCM-YOLO: a PCB defect detection method based on feature enhancement and multi-scale fusion[J]. Control and Decision, 2024, 39(10): 3181-3189.
[10] 张旭, 陈慈发, 董方敏. 基于改进YOLOv7的PCB缺陷检测算法[J]. 计算机工程, 2024, 50(12): 318-328.
ZHANG X, CHEN C F, DONG F M. PCB defect detection algorithm based on improved YOLOv7[J]. Computer Engineering, 2024, 50(12): 318-328.
[11] YU F， KOLTUN V. Multi-scale context aggregation by dilated convolutions[J]. arXiv:1511.07122, 2015.
[12] HU S, GAO F, ZHOU X W, et al. Hybrid convolutional and attention network for hyperspectral image denoising[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 5504005.
[13] ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2020: 12993-13000.
[14] MA N, SU Y X, YANG L X, et al. Wheat seed detection and counting method based on improved YOLOv8 model[J]. Sensors, 2024, 24(5): 1654.
[15] SINGH G, STEFENON S F, YOW K C. Interpretable visual transmission lines inspections using pseudo-prototypical part network[J]. Machine Vision and Applications, 2023, 34(3): 41.
[16] LOU H T, DUAN X H, GUO J M, et al. DC-YOLOv8: small-size object detection algorithm based on camera sensor[J]. Electronics, 2023, 12(10): 2323.
[17] 党张敏, 喻崇仁, 殷双飞, 等. 基于CLIP与注意力机制的跨模态哈希检索算法[J]. 计算机工程与设计, 2024, 45(3): 852-858.
DANG Z M, YU C R, YIN S F, et al. Cross modal hash retrieval algorithm based on CLIP and attention mechanism[J]. Computer Engineering and Design, 2024, 45(3): 852-858.
[18] LI S C, HUANG H P, MENG X Y, et al. A glove-wearing detection algorithm based on improved YOLOv8[J]. Sensors, 2023, 23(24): 9906.
[19] OUYANG W, LUO P, ZENG X, et al. DeepID-Net: multi-stage and deformable deep convolutional neural networks for object detection[J]. arXiv:1409.3505, 2014.
[20] WANG C, HE W, NIE Y, et al. Gold-YOLO: efficient object detector via gather-and-distribute mechanism[C]//Advances in Neural Information Processing Systems, 2024.
[21] HU D A, YU M, WU X Y, et al. DGW-YOLOv8: a small insulator target detection algorithm based on deformable attention backbone and WIoU loss function[J]. IET Image Processing, 2024, 18(4): 1096-1108.
[22] REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 658-666.