| 姜媛媛,蔡梦南.轻量化的印刷电路板缺陷检测网络 Multi-CR YOLO[J].电子测量与仪器学报,2023,37(11):217-224 |
| 轻量化的印刷电路板缺陷检测网络 Multi-CR YOLO |
| Lightweight PCB defect detection network Multi-CR YOLO |
| |
| DOI： |
| 中文关键词: Multi-CR YOLO 缺陷检测 印刷电路板 SDDT-FPN PCR C5ECA |
| 英文关键词:Multi-CR YOLO defect detection PCB SDDT-FPN PCR C5ECA |
| 基金项目:安徽省重点研究与开发计划( 202104g01020012)、安徽理工大学环境友好材料与职业健康研究院研发专项基金( ALW2020YF18)项目资助 |
|
|
|
|
| 摘要点击次数: 252 |
| 全文下载次数: 417 |
| 中文摘要: |
| 针对印刷电路板表面缺陷目标小,检测精度低问题,设计了印刷电路板表面缺陷检测网络 Multi-CR YOLO,满足实时检
测速度的前提下,有效提高了检测精度。 首先,由 3 个 Multi-CR 块组成的主干特征提取网络 Multi-CR backbone 对印刷电路板
小目标缺陷进行特征提取。 其次,SDDT-FPN 特征融合模块使层级高的特征层向层级低的特征层进行特征融合,同时为小目标
预测头 YOLO Head-P3 所在特征融合层加强特征融合,进一步增强低层特征层的表达能力。 PCR 模块加强主干特征提取网络
与 SDDT-FPN 特征融合模块不同尺度的特征层的特征融合机制,且防止模块之间进行特征融合时信息丢失。 C5ECA 模块负责
自适应调节特征权重和自适应注意小目标缺陷信息的要求,进一步提高了特征融合模块的自适应特征提取能力。 最后,3 个
YOLO-Head 负责针对不同尺度的小目标缺陷进行预测。 实验表明,Multi-CR YOLO 网络模型检测 mAP 达到 98. 55%,模型大小
为 8. 90 MB,达到轻量化要求,检测速度达到了 95. 85 fps,满足小目标缺陷实时检测的应用需求。 |
| 英文摘要: |
| Aiming at the problem of small target and low detection accuracy of printed circuit board surface defects, Multi-CR YOLO, a
printed circuit board surface defect detection network, is designed to meet the premise of real-time detection speed and effectively
improve the detection accuracy. Firstly, the backbone feature extraction network Multi-CR backbone, which consists of three Multi-CR
residual blocks, performs feature extraction for small target defects on printed circuit boards. Secondly, the SDDT-FPN feature fusion
module enables the feature fusion from the high level feature layer to the low level feature layer, and at the same time strengthens the
feature fusion for the feature fusion layer where the small target prediction head YOLO Head-P3 is located, to further enhance the
expressive ability of the low level feature layer. The PCR module strengthens the feature fusion mechanism of the different scales of the
backbone feature extraction network and the feature layer of the SDDT-FPN feature fusion module, and prevents the fusion mechanism
between the modules. The C5ECA module is responsible for adaptively adjusting the feature weights and adaptively paying attention to
the requirement of small target defect information, which further improves the adaptive feature extraction capability of the feature fusion
module. Finally, the three YOLO-Head are responsible for predicting small target defects for different scales. The experiments show that
the detection mAP of the Multi-CR YOLO network model reaches 98. 55%, the model size is 8. 90 MB, which meets the lightweight
requirement, and the detection speed reaches 95. 85 fps, which meets the application requirements of real-time detection of small-target
defects. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
