摘要
针对PCB电路板缺陷检测对高速度、高准确率及轻量化的需求,提出改进YOLOv8算法(Sim B-YOLO)。该算法引入轻量级注意力机制Sim AM增强关键特征关注,采用加权双向特征金字塔网络(Bi FPN)优化多尺度特征融合,并通过模块缝合,结合C2f与自校准卷积(Self-Calibrated Convolutions)提升特征图有效性。实验表明,Sim BYOLO在北大PCB数据集上的m AP@0.5较YOLOv8s提升1.7%,计算量降低3.8GFLOPs,性能优于主流目标检测算法。
In view of the demand for high speed,high accuracy and light-weight processing on defect detection of the PCB circuit board,an improved YOLOv8 algorithm(SimB-YOLO)is proposed.This algorithm introduces a light-weight processing attention mechanism,SimAM,to enhance the focus on key features,employs the weighted bidirectional feature pyramid network(BiFPN)to optimize multi-scale feature fusion and utilizes module stitching combined with C2f and self-calibrated convolutions to enhance effectiveness of feature maps.Experimental results show that mAP@0.5 of SimB-YOLO on the PCB dataset of Peking University is 1.7%higher than that of YOLOv8s,and it is with a reduction in computational load of 3.8 GFLOPs.Performance of the improved algorithm is superior to that of mainstream object detection algorithms.
作者
赵慕东
赵望宇
董建新
Zhao Mudong;Zhao Wangyu;Dong Jianxin
出处
《一重技术》
2025年第4期67-70,共4页
CFHI Technology
关键词
YOLOv8
缺陷检测
轻量化
YOLOv8
defect detection
light-weight processing
作者简介
赵慕东,沈阳理工大学机械工程学院硕士研究生,辽宁沈阳110159;赵望宇,沈阳理工大学机械工程学院硕士研究生,辽宁沈阳110159;董建新,沈阳理工大学机械工程学院硕士研究生,辽宁沈阳110159。
