孙雪怡,王一群,陈雯柏,耿悦,汪三元

• 1:北京信息科技大学自动化学院

摘要(Abstract)：

为解决印制电路板（printed-circuit board, PCB）焊点缺陷检测领域存在的小目标特征易丢失、检测精度低以及模型参数量过多的问题，提出了一种基于改进YOLOv8算法的PCB焊点检测算法YOLO-CH。在YOLOv8特征金字塔与路径聚合网络基础上新增高分辨率P2检测层，将下采样倍数降至4倍，结合渐进式特征融合与双向特征交互策略，有效保留小目标的纹理、边缘等细节；设计了一种通道与边缘上下文聚合（channel and edge-wise context aggregation, CECA）双阶段注意力模块，采用通道注意力与坐标注意力级联的结构，增强小目标特征表示的同时保持模型轻量化；设计三维复合加权损失函数，缓解大小目标优化不平衡问题。基于自制PCB焊点缺陷数据集的实验结果表明，相较于YOLOv8算法，所提算法的mAP50提升了4.5百分点，mAP50-95提升了4.6百分点，召回率提升了5.3百分点，浮点运算量（floating-point operations, FLOPs）降低至7.09×10~(10)，有效降低了模型复杂度，提升了缺陷检测效果。

关键词(KeyWords)： 焊点缺陷检测;小目标;注意力机制;损失函数

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金重大研究计划项目（92267110）;国家自然科学基金面上项目（62276028）;; 北京市自然科学基金-小米联合基金重点研究专题（L233006）

作者(Author): 孙雪怡,王一群,陈雯柏,耿悦,汪三元

DOI: 10.16508/j.cnki.11-5866/n.2026.04.005

参考文献(References)：

• [1]Wang Qingyi,Guo Lingyun,Chen Bo. A real-time detection model for underwater dam cracks via machine vision[J].Engineering Applications of Artificial Intelligence,2026,167(Part 3):113942.
• [2]Peng Sheng,Periwal P. Deep learning enhanced analyses for pore type-based pore characteristics in shales using highresolution large-area SEM[J]. Fuel,2026,415:138407.
• [3]李帆，周迅，张岩，等.基于多尺度上下文增强的轻量级工业缺陷检测网络[J/OL].电子测量技术，2026-02-02. https：//link. cnki. net/urlid/11. 2175. tn. 20260130. 1739. 004.Li Fan,Zhou Xun,Zhang Yan,et al. A lightweight industrial defect detection network using multi-scale context enhancement[J/OL]. Electronic Measurement Technology,2026-02-02.https：//link. cnki. net/urlid/11. 2175. tn. 20260130. 1739. 004.(in Chinese)
• [4]苏佳，罗都，贾欣雨，等.基于YOLO-G的PCB缺陷检测算法[J].微电子学与计算机，2024,41(10):35-44.Su Jia,Luo Du,Jia Xinyu,et al. PCB defect detection algorithm based on YOLO-G[J]. Microelectronics&Computer,2024,41(10):35-44.(in Chinese)
• [5]胡兰兰，邓超.基于YOLO-MCG的PCB缺陷检测算法[J].仪表技术与传感器，2024(4):100-106.Hu Lanlan,Deng Chao. PCB defect detection algorithm based on YOLO-MCG[J]. Instrument Technique and Sensor,2024(4):100-106.(in Chinese)
• [6]Zheng Haoyu,Peng Jinmin,Yu Xinyi,et al. FDDC-YOLO:an efficient detection algorithm for dense small-target solder joint defects in PCB inspection[J]. Journal of Real-Time Image Processing,2025,22(2):83.
• [7]杨瑞君，赵雯，程燕，等.基于特征关系增强的PCB表面缺陷检测算法[J].激光杂志，2025,46(5):117-123.Yang Ruijun,Zhao Wen,Cheng Yan,et al. PCB surface defect detection algorithm based on feature relationship enhancement[J]. Laser Journal,2025,46(5):117-123.(in Chinese)
• [8]Ancha V K,Gonuguntla V,Vaddi R. TRSBi-YOLO:transformer based lightweight and high-performance model for PCB defects detection[J]. The Journal of Supercomputing,2025,81(13):1277.
• [9]Liu Bin,Lian Zheng,Yu Hui,et al. Research on the signal quantization method for in-line inspection of composite defects in pipeline welds using weak magnetic method[J]. Mechanical Systems and Signal Processing,2025,236:112947.
• [10]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:highprecision PCB tiny-defect detection algorithm based on multidimensional attention mechanism[J]. Acta Electronica Sinica,2024,52(7):2515-2528.(in Chinese)
• [11]Liu Wencong,Li Yehang,Huang Shunsong,et al. A small target detection algorithm for unmanned aerial vehicles incorporating global information modeling and multi-scale feature interaction[J]. Digital Signal Processing,2026,168(Part A):105465.
• [12]张雨，罗素云. YOLOv8-DSM：高精度的小目标交通检测算法[J/OL].信息与控制，2026-02-03. https：//doi. org/10. 13976/j. cnki. xk. 2025. 3213.Zhang Yu,Luo Suyun. YOLOv8-DSM:high-precision small target traffic detection algorithm[J/OL]. Information and Control,2026-02-03. https：//doi. org/10. 13976/j. cnki. xk.2025. 3213.(in Chinese)
• [13]Cheng Yulin,Liu Wenyu,Huang Tianqiang,et al. BCA-IML:bidirectional cross-attention guided multi-scale feature fusion for image manipulation localization[J]. Expert Systems with Applications,2026,309:131218.
• [14]Selvi R S,Nair R R,Suresh G R. Deep learning-based speech enhancement via adaptive Trans-UNet with novel loss function using enhanced aquila optimization algorithm[J]. The European Physical Journal Plus,2026,141(1):89-89.
• [15]孙中毅，王栋，曹国刚，等. Harmony-YOLO11：基于高频增强与特征引导的轻量级小目标检测算法[J].计算机工程与应用，2026,62(10):111-122.Sun Zhongyi,Wang Dong,Cao Guogang,et al. HarmonyYOLO11:lightweight small object detector via high-frequency enhancement and feature guidance[J]. Computer Engineering and Applications,2026,62(10):111-122(in Chinese)
• [16]邱永峰，罗开西，王志兵，等.自适应全局-局部注意力PCB缺陷检测算法AGS-YOLO[J].半导体技术，2026,51(2):190-199.Qiu Yongfeng,Luo Kaixi,Wang Zhibing,et al. AGS-YOLO:adaptive global-local self-attention algorithm for PCB defect detection[J]. Semiconductor Technology,2026,51(2):190-199.(in Chinese)
• [17]Xia Kewen,LüZhongliang,Liu Kang,et al. Global contextual attention augmented YOLO with ConvMixer prediction heads for PCB surface defect detection[J]. Scientific Reports,2023,13(1):9805.
• [18]施智华.基于机器视觉技术的交通监控图像细节信息增强处理方法[J].电子设计工程，2026,34(3):181-185.Shi Zhihua. Method for enhancing detail information of traffic monitoring images based on machine vision technology[J].Electronic Design Engineering,2026,34(3):181-185.(in Chinese)
• [19]Song Feihu,Bao Kaicheng,Deng Minglei,et al. Steel surface defect detection based on PSO-Gabor and an improved faster RCNN[J]. Nondestructive Testing and Evaluation,2026,41(5):2607-2630.
• [20]Wang Shoubin,Li Kun,Peng Guili,et al. APMLF-YOLO:a progressive multi-level feature fusion algorithm for industrial PCB tiny defect detection[J]. Signal,Image and Video Processing,2025,19(14):1195.
• [21]徐祖贵，邵辉，高崇阳，等.基于改进YOLOv5s的施工现场安全帽佩戴检测方法[J].安全与环境工程，2026,33(2):79-87.Xu Zugui,Shao Hui,Gao Chongyang,et al. Safety helmet wearing detection method based on improved YOLOv5s for construction sites[J]. Safety and Environmental Engineering,2026,33(2):79-87.(in Chinese)
• [22]陈孟元，许瑞珩，杨苏朋，等.动态遮挡场景下基于改进YOLOv6s网络的SLAM算法[J].中国惯性技术学报，2025,33(8):802-811.Chen Mengyuan,Xu Ruihang,Yang Supeng,et al. SLAM algorithm based on improved YOLOv6s network in dynamic occlusion scenarios[J]. Journal of Chinese Inertial Technology,2025,33(8):802-811.(in Chinese)
• [23]Zhang Jiayue,Liu Shanhui,Chen Minghui,et al. Surface defect detection in liquid crystal display polariser coating manufacturing based on an enhanced YOLOv10-N approach[J]. Coatings,2026,16(4):451-451.
• [24]Wang Ziteng,Mu Dengcong,Li Zheng,et al. A safety helmet detection algorithm based on improved YOLOv11n for complex coal mine scenarios[J]. Digital Signal Processing,2026,176:106082.
• [25]魏嘉敏，续欣莹，韩晖，等.基于YOLO-HorL的轻量级PCB缺陷检测算法[J].计算机工程与设计，2025,46(11):3283-3291.Wei Jiamin,Xu Xinying,Han Hui,et al. Lightweight PCB defect detection algorithm based on YOLO-HorL[J]. Computer Engineering and Design,2025,46(11):3283-3291.(in Chinese)
• [26]苏源，孙丹枫，赵建勇，等.基于结构重参数化和共享卷积的实时工业缺陷检测方法[J].现代制造工程，2025(12):83-96.Su Yuan,Sun Danfeng,Zhao Jianyong,et al. Real-time industrial defect detection method based on structural reparameterization and shared convolution[J]. Modern Manufacturing Engineering,2025(12):83-96.(in Chinese)