HS-YOLO:A Hierarchical Partial Multi-scale Convolution Model for PCB Small Object Detection
WANG Zhizhong
SUN Kelei
Abstract：To address the issue that the slight differences among the tiny defects(such as missing holes,rat bites,open circuits,etc.)occurring during the manufacturing of printed circuit board(PCB)made it difficult to distinguish them,the hierarchical partial multi-scale convolution and small targets strengthen pyramid-you only look once(HS-YOLO)model based on YOLO version 11 nano(YOLOv11n)was proposed.Firstly,a hierarchical partial multi-scale convolution(HPMSConv)module was introduced,which adopted a progressive mixed feature fusion strategy to enhance the model's adaptability to various defect types.Secondly,a small target enhancement pyramid was proposed,in which a crossover all nucleus module was designed,and improvements were made to the path aggregation feature pyramid network(PAFPN).By focusing on the global information of defects,the detection capability for small targets was significantly improved.The experimental results on the dataset of the Peking University market powered by printed circuit board(PKU-Market-PCB)showed that the mean average precision of the HS-YOLO model under the intersection over union thresholds ranging from 0.50 to 0.95 with a step size of 0.05 was increased by 3.7 percent compared with that of the YOLOv11n model,and the recall rate was increased by 4.1 percent.The HS-YOLO model not only enhanced the detection accuracy of minute PCB defects but also effectively addressed the issues of low differentiation and small target detection capability in PCB defect detection,providing a high-performance solution for automated PCB inspection.
Keywords：PCBsmall defectsfeature fusionsmall target enhancement pyramidcrossover all nucleus
Publication Date：2025-09-20
Online Publishing Date：2025-09-24(First online date of this platform, not the publication date of the document)
Pages：8( 411-417,424 )