| 吴必胜,高康松,陈松,徐浩飞,谢凯,贺建飚.CCM-YOLO: 一种改进的电路板密集区域元件检测方法[J].电子测量与仪器学报,2025,39(7):171-179 |
| CCM-YOLO: 一种改进的电路板密集区域元件检测方法 |
| CCM-YOLO: An improved component detection method fordense regions of circuit boards |
| |
| DOI： |
| 中文关键词: 目标检测 深度学习 电子元件 可分离卷积 YOLOv5 |
| 英文关键词:object detection deep learning electronic components separable convolution YOLOv5 |
| 基金项目:国家自然科学基金（62272485,62373372）项目资助 |
|
|
| Author | Institution |
| Wu Bisheng | School of Electronic Information and Electrical Engineering, Yangtze University, Jingzhou 434023, China |
| Gao Kangsong | School of Electronic Information and Electrical Engineering, Yangtze University, Jingzhou 434023, China |
| Chen Song | School of Electronic Information and Electrical Engineering, Yangtze University, Jingzhou 434023, China |
| Xu Haofei | School of Electronic Information and Electrical Engineering, Yangtze University, Jingzhou 434023, China |
| Xie Kai | School of Electronic Information and Electrical Engineering, Yangtze University, Jingzhou 434023, China |
| He Jianbiao | School of Computer Science, Central South University, Changsha 410083, China |
|
| 摘要点击次数: 1070 |
| 全文下载次数: 569 |
| 中文摘要: |
| 以YOLOv5为基准模型，针对电子元件在电路板上检测容易出现漏检、精度较低的问题。通过使用卷积块注意力模块（convolutional block attention module, CBAM）在进行特征提取过程中提升检测精度，改进边界回归损失函数的方法来改善模型在检测中出现的漏检问题。首先，利用卷积层提取元件的特征信息；其次，在路径聚合网络（path aggregation network, PANet）中引入了CBAM模块，既丰富了元件特征信息，又改善了模型精度较低的问题；最后，通过多尺度预测和自适应的锚框来实现对不同尺度元件的准确检测。实验结果表明，改进后的CCM-YOLO算法在自制的数据集上的均值平均精度（mean average precision, mAP）值达到96.8%，它的漏检率达到4.5%，相较于YOLOv5网络均值平均精度的88.8%，提高了8.3%的数值，在漏检率上由原基准模型的13.7%降低了9.2%。因此，该算法有效提高了检测精度，并显著减少了漏检，为元件检测提供了一种有效的检测方案。 |
| 英文摘要: |
| In this paper, YOLOv5 is used as a benchmark model to address the problem of easy leakage and low accuracy of electronic components detection on circuit boards. The leakage problem of the model in detection is improved by using the convolutional block attention module (CBAM) to enhance the detection accuracy in the process of feature extraction and improving the boundary regression loss function. Firstly, the feature information of components is extracted using convolutional layers. Secondly, the CBAM module is introduced into the path aggregation network (PANet), which enriches the feature information of components and improves the problem of lower accuracy of the model. Finally, accurate detection of components of different scales is achieved by multi-scale prediction and adaptive anchor frames. The experimental results show that the improved CCM-YOLO algorithm achieves a mean average precision (mAP) value of 96.8% on the homemade dataset, and it achieves a leakage detection rate of 4.5%, which is an improvement of 8.3% value compared to the 88.8% of the mean average precision of the YOLOv5 network, and the leakage detection rate is reduced from 13.7% of the original baseline model is reduced by 9.2%. Therefore, the algorithm in this paper effectively improves the detection accuracy and significantly reduces the leakage detection, providing an effective detection scheme for component detection. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
