Small-object defect detection method for printed circuit boards based on dual-branch feature extraction

doi:10.11772/j.issn.1001-9081.2025081034

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (9): 2977-2986.DOI: 10.11772/j.issn.1001-9081.2025081034

• Multimedia computing and computer simulation • Previous Articles

Small-object defect detection method for printed circuit boards based on dual-branch feature extraction

Yufeng OU¹, Chia-Hung WANG¹^,²()

^1.School of Computing and Data Science，Fujian University of Technology，Fuzhou Fujian 350118，China
^2.Fujian Provincial Key Laboratory of Big Data Mining and Applications （Fujian University of Technology），Fuzhou Fujian 350118，China

Received:2025-09-08 Revised:2025-10-20 Accepted:2025-10-23 Online:2025-11-07 Published:2026-09-10
Contact: Chia-Hung WANG
About author:OU Yufeng， born in 2000， M. S. candidate. His research interests include artificial intelligence， simulation optimization.
WANG Chia-Hung， born in 1979， Ph. D.， associate professor. His research interests include artificial intelligence， big data analytics.
Supported by:
Fujian Provincial Natural Science Foundation(2021J011070)

基于双分支特征提取的印刷电路板小目标缺陷检测方法

欧宇枫¹, 王嘉宏¹^,²()

^1.福建理工大学计算机与数据科学学院，福州 350118
^2.福建省大数据挖掘与应用技术重点实验室（福建理工大学），福州 350118

通讯作者: 王嘉宏
作者简介:欧宇枫（2000—），男，福建福州人，硕士研究生，CCF会员，主要研究方向：人工智能、仿真优化
王嘉宏（1979—），男，台湾台南人，副教授，博士，CCF会员，主要研究方向：人工智能、大数据分析。
基金资助:
福建省自然科学基金资助项目(2021J011070)

Abstract

Abstract:

To address the challenges of detecting small-object defects on Printed Circuit Board （PCB） surfaces， such as low contrast， irregular shapes， and susceptibility to background interference， a dual-branch feature detection model named Dual branch feature Extraction-Spatial attention Detection （DES-Det） was proposed. First， a dual-branch collaboration was used to extract global contextual and local detailed features， thereby enhancing the discriminative capability for low-pixel defects. Second， a deformable attention mechanism was introduced to achieve adaptive feature focusing， thereby improving the localization accuracy for blurred edges and irregular defects. At the same time， a detection head based on spatial attention and depthwise separable convolution was designed to reduce model complexity significantly while maintaining detection performance. Experimental results on two public PCB defect datasets demonstrate that DES-Det achieves precisions of 93.6% and 98.4%， respectively， representing improvements of 5.7 and 1.4 percentage points， respectively， compared to the baseline model YOLOv8， with the number of parameters reduced by 26.7%. Cross-scenario generalization experimental results further validate the method's effectiveness and robustness in complex industrial environments.

Key words: Printed Circuit Board (PCB), small-object defect detection, attention mechanism, multi-scale feature extraction, depthwise separable convolution

摘要：

针对印刷电路板（PCB）表面小目标缺陷检测中存在的低对比度、形态不规则及易受背景干扰等难题，提出一种双分支特征检测模型DES-Det（Dual branch feature Extraction-Spatial attention Detection）。首先，通过双分支结构协同提取全局上下文与局部细节特征，增强对低像素缺陷的判别能力；其次，引入可变形注意力机制实现自适应特征聚焦，提升对模糊边缘与不规则缺陷的定位精度；并设计基于空间注意力与深度可分离卷积的检测头，在保障检测性能的同时显著降低模型复杂度。在2个公开的PCB缺陷数据集上的实验结果表明，DES-Det的精确率分别达到93.6%与98.4%，相较于基线模型YOLOv8分别提升了5.7与1.4个百分点，参数量下降了26.7%。跨场景泛化实验的结果进一步验证了该方法在复杂工业环境中的有效性与鲁棒性。

关键词: 印刷电路板, 小目标缺陷检测, 注意力机制, 多尺度特征提取, 深度可分离卷积

CLC Number:

TP391.4

Yufeng OU, Chia-Hung WANG. Small-object defect detection method for printed circuit boards based on dual-branch feature extraction[J]. Journal of Computer Applications, 2026, 46(9): 2977-2986.

欧宇枫, 王嘉宏. 基于双分支特征提取的印刷电路板小目标缺陷检测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2977-2986.

Figures/Tables 14

References 31

[1]	孙成路. 基于机器视觉的PCB异型插装元件AOI检测系统［D］. 武汉：武汉纺织大学， 2023： 1-76.
	Sun Chenglu. An AOI inspection system for PCB irregular plug-in components based on machine vision ［D］. Wuhan： Wuhan Textile University， 2023： 1-76.
[2]	杨思念，曹立佳，杨洋，等. 基于机器视觉的PCB缺陷检测算法研究综述［J］. 计算机科学与探索， 2025， 19（4）： 901-915.
	Yang Sinian， Cao Lijia， Yang Yang， et al. Review of PCB defect detection algorithm based on machine vision ［J］. Journal of Frontiers of Computer Science and Technology， 2025， 19（4）： 901-915.
[3]	Hu B， Wang J. Detection of PCB surface defects with improved Faster-RCNN and feature pyramid network ［J］. IEEE Access， 2020， 8： 108335-108345.
[4]	龚陈博，南卓江，陶卫. 基于改进Faster RCNN的PCB表面缺陷检测研究［J］. 自动化仪表， 2024， 45（7）： 99-103， 109.
	Gong Chenbo， Zhuojiang Nan， Tao Wei. Research on PCB surface defect detection based on improved Faster RCNN ［J］. Process Automation Instrumentation， 2024， 45（7）： 99-103， 109.
[5]	Li Q， Wu L， Xiao H， et al. PCB-DETR： a detection network of PCB surface defect with spatial attention offset module ［J］. IEEE Access， 2024， 12： 158436-158445.
[6]	王玲，王明慧，王鹏，等. 融合多尺度特征的知识蒸馏工业缺陷检测算法［J］. 计算机工程与应用， 2025， 61（21）： 276-286.
	Wang Ling， Wang Minghui， Wang Peng， et al. Industrial defect detection algorithm with knowledge distillation integrating multi-scale features ［J］. Computer Engineering and Applications， 2025， 61（21）： 276-286.
[7]	毛坤，朱学军，赖惠鸽，等. SGW-YOLO：多尺度融合优化的PCB缺陷检测算法［J/OL］. 激光与光电子学进展（2025-02-13）［2025-03-10］..
	Mao Kun， Zhu Xuejun， Lai Huige， et al. Multi scale fusion algorithm for PCB defect detection based on SGW-YOLO ［J/OL］. Laser and Optoelectronics Progress （2025-02-13）［2025-03-10］..
[8]	Tong Z， Chen Y， Xu Z， et al. Wise-IoU： bounding box regression loss with dynamic focusing mechanism ［PP/OL］. V3. arXiv （2023-04-08）［2025-03-12］..
[9]	Bai L， Xu W H. Improved printed circuit board defect detection scheme ［J］. Scientific Reports， 2025， 15： No.2389.
[10]	Yuan T， Jiao Z， Diao N. YOLO-SSW： an improved detection method for printed circuit board surface defects ［J］. Mathematics， 2025， 13（3）： No.435.
[11]	王健，肖迪，冯李航，等. 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， et al. PCB small object defect detection model based on improved YOLOv8s ［J］. Computer Engineering and Applications， 2025， 61（15）： 288-297.
[12]	李耀龙，陈晓林，林浩，等. DySnake-YOLO：改进的YOLOv9c电路板表面缺陷检测方法［J］. 计算机工程与应用， 2025， 61（3）： 242-252.
	Li Yaolong， Chen Xiaolin， Lin Hao， et al. DySnake-YOLO： improved detection of surface defects on YOLOv9c circuit board ［J］. Computer Engineering and Applications， 2025， 61（3）： 242-252.
[13]	郭艳，王智文，赵润星. YOLO-POD：基于多维注意力机制的高精度PCB微小缺陷检测算法［J］. 电子学报， 2024， 52（7）： 2515-2528.
	Guo Yan， Wang Zhiwen， Zhao Runxing. YOLO-POD： high-precision PCB tiny-defect detection algorithm based on multi-dimensional attention mechanism ［J］. Acta Electronica Sinica， 2024， 52（7）： 2515-2528.
[14]	王海群，王炳楠，葛超. 改进ESP-YOLO的PCB缺陷检测算法［J］. 计算机工程与科学， 2025， 47（2）： 317-326.
	Wang Haiqun， Wang Bingnan， Ge Chao. A PCB defect detection algorithm based on improved ESP-YOLO ［J］. Computer Engineering and Science， 2025， 47（2）： 317-326.
[15]	张淑卿，孟昊，葛超. 改进YOLOv8n的轻量级PCB缺陷检测算法［J］.现代电子技术， 2024， 47（15）： 115-121.
	Zhang Shuqing， Meng Hao， Ge Chao. Lightweight PCB defect detection algorithm for improved YOLOv8n ［J］. Modern Electronic Technique， 2024， 47（15）： 115-121.
[16]	Zhang T， Zhang J， Pan P， et al. YOLO-RRL： a lightweight algorithm for PCB surface defect detection ［J］. Applied Sciences， 2024， 14（17）： No.7460.
[17]	Nandal P， Mann P， Bohra N， et al. Tropical cyclone intensity estimation based on YOLO-NAS using satellite images in real time［J］. Alexandria Engineering Journal， 2025， 113： 227-241.
[18]	Ultralytics. YOLOv8［EB/OL］. ［2024-07-01］..
[19]	Xia Z， Pan X， Song S， et al. Vision Transformer with deformable attention ［C］// CVPR 2022. Piscataway： IEEE， 2022： 4794-4803.
[20]	Zheng Z， Wang P， Liu W， et al. Distance-IoU loss： faster and better learning for bounding box regression ［J］. Proceedings of the AAAI Conference on Artificial Intelligence， 2020， 34（7）： 12993-13000.
[21]	Zhang Y F， Ren W， Zhang Z， et al. Focal and efficient IOU loss for accurate bounding box regression ［J］. Neurocomputing， 2022， 506： 146-157.
[22]	Govorgyan Z. SIoU loss： more powerful learning for bounding box regression ［PP/OL］. arXiv （2022-05-25）［2025-03-24］..
[23]	Wang J， Xu C， Yang W， et al. A normalized Gaussian Wasserstein distance for tiny object detection ［PP/OL］. V2. arXiv （2022-06-14）［2025-03-18］..
[24]	Zhao Y， Lv W， Xu S， et al. DETRs beat YOLOs on real-time object detection ［C］// CVPR 2024. Piscataway： IEEE， 2024： 16965-16974.
[25]	Woo S， Debnath S， Hu R， et al. ConvNext V2： co-designing and scaling ConvNets with masked autoencoders ［C］// CVPR 2023. Piscataway： IEEE， 2023： 16133-16142.
[26]	Wang C Y， Yeh I H， Mark Liao H Y. YOLOv9： learning what you want to learn using programmable gradient information ［C］// ECCV 2024， LNCS 15089. Cham： Springer， 2025： 1-21.
[27]	Wang A， Chen H， Liu L， et al. YOLOv10： real-time end-to-end object detection ［C］// NeurIPS 2024. Red Hook： Curran Associates Inc.， 2024： 107984-108011.
[28]	Khanam R， Hussain M. YOLOv11： an overview of the key architectural enhancements［PP/OL］. arXiv （2024-10-23）［2025-03-14］..
[29]	Tian Y， Ye Q， Doermann D. YOLOv12： attention-centric real-time object detectors［PP/OL］. arXiv （2025-02-18）［2025-03-10］..
[30]	Liu Z， Hao Z， Han K， et al. GhostNetV3： exploring the training strategies for compact models ［PP/OL］. V2. arXiv （2024-04-22）［2025-03-10］..
[31]	Wang C， He W， Nie Y， et al. Gold-YOLO： efficient object detector via gather-and-distribute mechanism ［C］// NeurIPS 2023. Red Hook： Curran Associates Inc.， 2023： 51094-51112.

参数名称	参数设置
图片尺寸（Image Size）	640×640
学习率（Learning Rate）	1E-2
学习率衰减策略	随机梯度下降法
训练轮次（Train Epoch）	200
优化器（Optimizer）	SGD
权重衰减（Weight Delay）	5E-4
批大小（Batch Size）	16
数据加载的子进程数（Workers）	8
DBdown分支通道占比	50%∶50%
Deformable Attention的采集点个数	4
Deformable Attention的注意力头数	8

参数名称	参数设置
图片尺寸（Image Size）	640×640
学习率（Learning Rate）	1E-2
学习率衰减策略	随机梯度下降法
训练轮次（Train Epoch）	200
优化器（Optimizer）	SGD
权重衰减（Weight Delay）	5E-4
批大小（Batch Size）	16
数据加载的子进程数（Workers）	8
DBdown分支通道占比	50%∶50%
Deformable Attention的采集点个数	4
Deformable Attention的注意力头数	8

损失函数	P	F1	mAP@0.5
CIoU	89.6	68.8	66.7
EIoU	91.3	72.1	71.8
SIoU	90.4	72.4	72.3
NWD	91.7	72.5	72.3
WIoUv1	89.4	70.5	68.5
WIoUv2	93.6	73.5	72.5
WIoUv3	89.9	73.1	74.1

损失函数	P	F1	mAP@0.5
CIoU	89.6	68.8	66.7
EIoU	91.3	72.1	71.8
SIoU	90.4	72.4	72.3
NWD	91.7	72.5	72.3
WIoUv1	89.4	70.5	68.5
WIoUv2	93.6	73.5	72.5
WIoUv3	89.9	73.1	74.1

模型	P/%	F1/%	mAP@0.5/%	参数量/10⁶	浮点运算量/GFLOPs	帧率/（frame·s^-1）
YOLOv8（baseline）	87.9	69.2	68.0	3.0	8.1	456.8
YOLOv8+DBdown	91.2	72.6	72.3	2.7	7.7	449.9
YOLOv8+SDDH	88.9	71.2	70.0	2.6	5.6	423.9
YOLOv8+C2f_DAT	92.2	72.9	71.7	3.1	8.1	439.4
YOLOv8+ DBdown +C2f_DAT	92.1	71.6	72.2	2.7	7.6	439.3
YOLOv8+ DBdown + SDDH	91.7	70.6	70.2	2.1	5.5	426.1
YOLOv8+C2f_DAT+ SDDH	90.2	72.3	71.3	2.4	5.7	434.9
YOLOv8+ DBdown +C2f_DAT+ SDDH	93.6	73.5	72.5	2.2	5.3	435.7

Small-object defect detection method for printed circuit boards based on dual-branch feature extraction

基于双分支特征提取的印刷电路板小目标缺陷检测方法

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 31

Related Articles 15

Recommended Articles

Metrics

数据集	模型	P	R	F1	mAP@0.5
实景拍摄PCB缺陷数据集	YOLOv8	77.3±0.3	78.1±0.2	77.6±0.2	78.7±0.2
	GhostNet	62.6±0.4	60.4±0.5	61.4±0.4	62.8±0.4
	Gold-YOLO	58.3±0.3	72.3±0.4	64.4±0.3	45.9±0.3
	ConvNeXtV2	73.3±0.2	44.6±0.3	55.3±0.2	47.9±0.2
	YOLO-NAS	74.8±0.1	57.6±0.2	64.9±0.1	37.8±0.1
	DES-Det	77.6±0.2	80.4±0.1	80.0±0.2	79.6±0.2
铝材缺陷数据集	YOLOv8	75.2±0.3	85.9±0.2	80.2±0.2	70.1±0.2
	GhostNet	70.0±0.3	53.2±0.4	60.5±0.4	58.4±0.4
	Gold-YOLO	74.4±0.2	57.4±0.1	62.9±0.1	62.8±0.2
	ConvNeXtV2	67.4±0.3	59.0±0.2	62.9±0.2	61.1±0.3
	YOLO-NAS	71.1±0.2	55.8±0.1	62.5±0.1	61.3±0.1
	DES-Det	79.7±0.1	86.9±0.2	83.1±0.2	72.8±0.2

[1]	Lixia XIE, Chaoyue CAO, Hongyu YANG. ADS-B attack detection method based on improved DeepLabV3+ [J]. Journal of Computer Applications, 2026, 46(9): 2867-2876.
[2]	Zhangjian JI, Siyuan WANG. Lightweight human pose estimation framework based on wavelet attention mechanism with enhanced low frequency [J]. Journal of Computer Applications, 2026, 46(9): 2787-2792.
[3]	Yifan WANG, Hu HAN, Dong LI, Yating FAN, Lin LI. Heterogeneous enhancement and multi-source knowledge fusion for aspect-based sentiment analysis [J]. Journal of Computer Applications, 2026, 46(9): 2752-2760.
[4]	Ming LIU, Dongqi SHEN, Ziyang MENG. Point cloud registration network with dual-branch multi-level feature fusion [J]. Journal of Computer Applications, 2026, 46(8): 2584-2593.
[5]	Fengchun LIU, Xinying SHAO, Chunying ZHANG, Liya WANG, Jing REN. FCMdepth： monocular depth estimation framework with multi-scale feature optimization [J]. Journal of Computer Applications, 2026, 46(8): 2603-2611.
[6]	Wei LIU, Weigang LI, Zhiqiang TIAN. Representation learning method of hierarchical rotation-invariant geometric structure for point cloud classification and segmentation [J]. Journal of Computer Applications, 2026, 46(8): 2594-2602.
[7]	Jielong BAI, Chenyun FANG, Zhiwei QIAO. Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion [J]. Journal of Computer Applications, 2026, 46(8): 2660-2667.
[8]	Guanghui LI, Licheng QU. Dynamic dilated convolution and hierarchical local-global attention model based on improved Lite-Mono architecture [J]. Journal of Computer Applications, 2026, 46(8): 2567-2576.
[9]	Yuanhang HUANG, Na RONG. Model-free photovoltaic hosting capacity assessment method using attention mechanism and deep reinforcement learning [J]. Journal of Computer Applications, 2026, 46(8): 2708-2715.
[10]	Xiaojin GUO, Xuyang SUI, Kenan ZHOU. Channel estimation of STAR-RIS-assisted hybrid-field communication system based on deep learning [J]. Journal of Computer Applications, 2026, 46(8): 2548-2554.
[11]	Jianqiang YAN, Beibei DONG, Boting QU, Chen PENG. Bi-directional diffusion dynamic graph convolutional network with multi-source information fusion and graph-level attention for traffic flow prediction [J]. Journal of Computer Applications, 2026, 46(7): 2327-2333.
[12]	Wenling LIU, Yong LI, Jiahui LI, Houkang ZHANG. Unsupervised dehazing method for remote sensing images with non-uniform haze [J]. Journal of Computer Applications, 2026, 46(7): 2297-2306.
[13]	Xiuli DU, Xing GAO, Xiaoyu ZHANG, Chengsheng PAN, Qijie ZOU. Video snapshot compressive imaging reconstruction method based on dense spatio-temporal deformable attention [J]. Journal of Computer Applications, 2026, 46(7): 2288-2296.
[14]	Guoyou ZHANG, Hongyu NIE, Lihu PAN, Rundong LEI. Point-MLPBLS： point cloud semantic segmentation network based on MLP cascaded broad learning system [J]. Journal of Computer Applications, 2026, 46(7): 2259-2266.
[15]	Yasong ZHANG, Bihui CONG, Shuang XU. Graph neural network node classification model incorporating clustering coefficients [J]. Journal of Computer Applications, 2026, 46(6): 1855-1862.

模型	P/%	F1/%	mAP@0.5/%	不同缺陷类型的AP/%						参数量/10⁶	浮点运算量 /GFLOPs	帧率/（frame·s^-1）
模型	P/%	F1/%	mAP@0.5/%	缺孔	鼠咬	开路	短路	毛刺	杂铜	参数量/10⁶	浮点运算量 /GFLOPs	帧率/（frame·s^-1）
RT-DETR	72.4	59.1	52.6	86.6	67.2	64.7	74.2	75.5	66.3	284.5	100.6	40.3
YOLOv8	87.9	69.2	68.0	95.8	87.1	89.9	87.1	87.8	79.5	3.0	8.1	456.8
YOLOv9	90.2	69.4	66.8	94.5	84.3	94.0	89.7	83.2	95.4	2.1	7.6	328.2
YOLOv10	86.7	66.9	65.5	91.0	80.6	89.2	89.3	84.8	85.3	2.5	8.2	358.8
YOLOv11	91.3	67.3	65.1	89.0	86.6	95.6	91.6	97.6	87.4	2.5	6.3	339.1
YOLOv12	90.2	68.5	63.4	86.9	90.1	91.5	88.7	94.7	89.0	2.5	6.3	217.5
GhostNet-YOLO	85.5	65.3	62.3	91.0	78.4	86.3	87.7	86.1	83.8	2.0	5.8	363.1
Gold-YOLO	88.3	68.9	66.4	93.9	82.3	87.4	88.8	89.7	87.6	5.9	10.2	309.0
ConvNeXtV2	57.2	46.8	39.4	64.7	52.3	49.0	60.4	57.5	59.6	56.6	14.1	158.0
YOLO-NAS	86.3	69.1	66.5	93.6	79.7	91.3	84.7	90.7	78.4	2.4	6.6	423.5
YOLO-RRL	91.3	71.7	71.1	95.8	85.7	94.3	89.6	93.1	89.1	2.4	5.8	470.5
YOLO-SSW	90.8	73.3	73.7	94.5	89.3	89.7	93.3	89.5	88.4	4.3	11.2	218.1
DES-Det	93.6	73.5	72.5	96.7	91.3	96.1	90.7	94.1	92.9	2.2	5.3	435.7

模型	P/%	F1/%	mAP@0.5/%	不同缺陷类型的AP/%						参数量/10⁶	浮点运算量 /GFLOPs	帧率/（frame·s^-1）
模型	P/%	F1/%	mAP@0.5/%	缺孔	鼠咬	开路	短路	毛刺	杂铜	参数量/10⁶	浮点运算量 /GFLOPs	帧率/（frame·s^-1）
RT-DETR	72.4	59.1	52.6	86.6	67.2	64.7	74.2	75.5	66.3	284.5	100.6	40.3
YOLOv8	87.9	69.2	68.0	95.8	87.1	89.9	87.1	87.8	79.5	3.0	8.1	456.8
YOLOv9	90.2	69.4	66.8	94.5	84.3	94.0	89.7	83.2	95.4	2.1	7.6	328.2
YOLOv10	86.7	66.9	65.5	91.0	80.6	89.2	89.3	84.8	85.3	2.5	8.2	358.8
YOLOv11	91.3	67.3	65.1	89.0	86.6	95.6	91.6	97.6	87.4	2.5	6.3	339.1
YOLOv12	90.2	68.5	63.4	86.9	90.1	91.5	88.7	94.7	89.0	2.5	6.3	217.5
GhostNet-YOLO	85.5	65.3	62.3	91.0	78.4	86.3	87.7	86.1	83.8	2.0	5.8	363.1
Gold-YOLO	88.3	68.9	66.4	93.9	82.3	87.4	88.8	89.7	87.6	5.9	10.2	309.0
ConvNeXtV2	57.2	46.8	39.4	64.7	52.3	49.0	60.4	57.5	59.6	56.6	14.1	158.0
YOLO-NAS	86.3	69.1	66.5	93.6	79.7	91.3	84.7	90.7	78.4	2.4	6.6	423.5
YOLO-RRL	91.3	71.7	71.1	95.8	85.7	94.3	89.6	93.1	89.1	2.4	5.8	470.5
YOLO-SSW	90.8	73.3	73.7	94.5	89.3	89.7	93.3	89.5	88.4	4.3	11.2	218.1
DES-Det	93.6	73.5	72.5	96.7	91.3	96.1	90.7	94.1	92.9	2.2	5.3	435.7

模型	P/%	F1/%	mAP@0.5/%	不同缺陷类型的AP/%						帧率/（frame·s^-1）
模型	P/%	F1/%	mAP@0.5/%	孔洞	开路	短路	鼠咬	毛刺	杂铜	帧率/（frame·s^-1）
RT-DETR	92.0	90.3	94.9	91.3	91.2	94.3	92.4	90.8	92.1	45.1
YOLOv8	97.0	95.9	98.4	96.9	94.5	97.3	97.5	99.3	96.5	467.2
YOLOv9	97.5	96.4	98.5	97.6	97.1	97.9	97.3	97.7	97.2	301.7
YOLOv10	96.6	95.5	98.1	96.2	95.1	97.4	96.7	97.2	96.9	389.3
YOLOv11	96.6	96.0	98.1	96.1	96.4	97.6	95.9	96.9	96.8	342.9
YOLOv12	95.5	94.9	97.6	96.5	93.9	95.9	93.5	97.4	96.0	233.8
GhostNet-YOLO	97.5	94.9	98.0	97.9	97.2	97.1	96.1	99.1	97.5	440.6
Gold-YOLO	96.1	95.2	98.2	97.7	96.3	97.1	97.2	99.0	89.2	302.0
ConvNeXtV2	92.2	88.5	93.5	95.0	85.3	93.1	92.5	92.4	95.8	95.0
YOLO-NAS	90.7	93.6	96.7	91.5	89.2	90.3	91.1	89.2	93.3	98.5
YOLO-RRL	97.3	95.7	97.5	95.8	85.7	94.3	89.6	93.1	89.1	418.3
YOLO-SSW	97.4	96.4	98.4	94.5	89.3	89.7	93.3	89.5	88.4	235.6
DES-Det	98.4	97.4	98.7	98.7	97.6	99.0	98.4	99.2	97.8	446.8

模型	P/%	F1/%	mAP@0.5/%	不同缺陷类型的AP/%						帧率/（frame·s^-1）
模型	P/%	F1/%	mAP@0.5/%	孔洞	开路	短路	鼠咬	毛刺	杂铜	帧率/（frame·s^-1）
RT-DETR	92.0	90.3	94.9	91.3	91.2	94.3	92.4	90.8	92.1	45.1
YOLOv8	97.0	95.9	98.4	96.9	94.5	97.3	97.5	99.3	96.5	467.2
YOLOv9	97.5	96.4	98.5	97.6	97.1	97.9	97.3	97.7	97.2	301.7
YOLOv10	96.6	95.5	98.1	96.2	95.1	97.4	96.7	97.2	96.9	389.3
YOLOv11	96.6	96.0	98.1	96.1	96.4	97.6	95.9	96.9	96.8	342.9
YOLOv12	95.5	94.9	97.6	96.5	93.9	95.9	93.5	97.4	96.0	233.8
GhostNet-YOLO	97.5	94.9	98.0	97.9	97.2	97.1	96.1	99.1	97.5	440.6
Gold-YOLO	96.1	95.2	98.2	97.7	96.3	97.1	97.2	99.0	89.2	302.0
ConvNeXtV2	92.2	88.5	93.5	95.0	85.3	93.1	92.5	92.4	95.8	95.0
YOLO-NAS	90.7	93.6	96.7	91.5	89.2	90.3	91.1	89.2	93.3	98.5
YOLO-RRL	97.3	95.7	97.5	95.8	85.7	94.3	89.6	93.1	89.1	418.3
YOLO-SSW	97.4	96.4	98.4	94.5	89.3	89.7	93.3	89.5	88.4	235.6
DES-Det	98.4	97.4	98.7	98.7	97.6	99.0	98.4	99.2	97.8	446.8