摘要
提出一种基于Faster RCNN(Faster Region with Convolutional Neural Networks)的电路板缺陷图像自动检测方法。该方法首先应用ResNet50网络作为主干网络以提取缺陷图像特征;然后针对电路板图像中缺陷的极端长宽比特点,提出基于特征金字塔的区域推荐网络,得到多尺度特征融合图并生成建议框;最后,通过对感兴趣区域进行池化处理,并结合后续网络实现对电路板图像上缺陷的快速检测。试验证明,所提算法能够快速定位出电路板图像中的各种缺陷,并能实现准确的自动分类识别。
An automatic detection method of PCB defect image based on fast RCNN(Fast Region with Convolution Neural Networks)is proposed.Firstly,ResNet50 network is used as the backbone network to extract the feature of defect image.Secondly,according to the extreme aspect ratio of defects in circuit board image,a region proposal network based on feature pyramid is proposed to obtain multi-scale feature fusion map and generate suggestion box.Finally,the region of interest is pooled and combined with the subsequent network to realize the recognition of circuit board image rapid detection of defects.Experiments show that the proposed algorithm can quickly locate various defects in the circuit board images,and can achieve accurate automatic classifi cation and recognition.
作者
何威
吴伟
张加波
汪锐
黄子健
HE Wei;WU Wei;ZHANG Jiabo;WANG Rui;HUANG Zijian(Aperture Array and Space Exploration Laboratory of Anhui Province,The 38th Research Institute of CETC,Hefei 230031,China)
出处
《电子工艺技术》
2023年第6期7-11,20,共6页
Electronics Process Technology
基金
安徽省重点研发计划项目(202004a05020049)。