一种用于PCB缺陷检测的DRM-YOLOv8n算法  

A DRM-YOLOv8n algorithm for PCB defect detection

在线阅读下载全文

作  者:周静 黄丽雯 唐鑫 王博思 ZHOU Jing;HUANG Liwen;TANG Xin;WANG Bosi(School of Electrical and Electronic Engineering,Chongqing University of Technology,Chongqing 400000,China;KingZon(Chongqing)Package Technology Co.,Ltd.,Chongqing 400000,China;China Merchants Zhixing(Chongqing)Technology Co.,Ltd.,Chongqing 400000,China)

机构地区:[1]重庆理工大学电气与电子工程学院,重庆400000 [2]鼎兆(重庆)包装科技有限公司,重庆400000 [3]招商智行(重庆)科技有限公司,重庆400000

出  处:《激光杂志》2025年第3期65-70,共6页Laser Journal

基  金:国家自然科学基金面上项目(No.51876018);重庆市教委项目(No.KJQN201801126、No.KJQN201905604);重庆市科技局技术创新与应用发展专项重点项目(No.cstc2019jscx-mbdxX0002)。

摘  要:为了提高图像检测算法对印刷电路板(PCB)缺陷检测的精确度,提出一种DRM-YOLOv8n小目标检测算法。首先针对小尺度特征提取问题,采用可变形卷积模块改进骨干网络,提高网络对关键特征的提取能力,提升检测准确性;其次,针对复杂场景问题,在Neck结构中引入一种感受野注意力卷积(RFAConv),使模型定位更准确,提高网络性能和效率;最后,使用MPDIoU损失函数优化原网络损失函数,提高边界框回归的准确性与效率及模型的收敛能力。通过对比实验得出,提出的算法与YOLOv8n相比,平均精度值(mAP)有了明显提升,mAP50%从87.2%提高到94.5%,mAP50:95%从60.9%提高到65.8%,较YOLOV8n分别提高了7.3%和4.9%,证明了改进算法的有效性。The paper proposes a DRM-YOLOv8n small target detection algorithm to enhance the accuracy of image detection for PCB defect detection.Firstly,the deformable convolutional module is employed to address the issue of small-scale feature extraction,thereby enhancing the backbone network's capability in extracting crucial features and improving detection accuracy.Secondly,a receptive field attention Convolution(RFAConv) is introduced into the Neck structure to improve model positioning accuracy in complex scenes and enhance network performance and efficiency.Lastly,the MPDIoU loss function is utilized to optimize the original network loss function,resulting in improved boundary box regression accuracy and model convergence ability.Compared with YOLOv8n,our proposed algorithm achieves significant improvements in average precision value(mAP),with mAP50% increasing from 87.2% to 94.5% and mAP50:95% increasing from 60.9% to 65.8%,respectively-representing a 7.3% and 4.9% improvement over YOLOV8n.

关 键 词:深度学习 YOLOv8 小目标检测 可变形卷积 损失函数 

分 类 号:TN249[电子电信—物理电子学]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象