基于密集YOLOv3的印刷电路板缺陷识别

Defect Recognition of Printed Circuit Board Based on YOLOv3-Dense

对印刷电路板加工过程中微小缺陷的精准检测是保证电子产品质量的前提。由于电路板缺陷的特征尺寸极小,电路复杂,现有的目标检测方法存在很多不足。针对这一问题,在YOLOv3算法的基础,提出了一种用于印刷电路板缺陷检测的密集YOLOv3目标检测算法。首先,用密集连接卷积网络模块代替YOLOv3算法特征提取网络中的部分残差网络单元,增强网络的特征重用;其次,对损失函数加以改进,用预测框和真实值之间的广义交并比来解决交并比为零时无法继续优化的问题。所提出的密集YOLOv3算法在扩充后的印刷电路板缺陷数据集上得到了有效地验证。实验结果表明,与其他识别算法相比,所提算法在识别精度提高的同时,算法尺寸也有所减小。

Accurate detection of tiny defects in printed circuit board processing is the prerequisite to ensure the quality of electronic products.Due to the small feature size and complex circuit layouts,the existing target detection methods have many shortcomings.To solve this problem,a YOLOv3-dense target detection model is proposed for printed circuit board defect detection based on YOLOv3 algorithm.First,dense connection network modules are used to replace some residual units in the feature extraction network so as to enhance feature reuse of the network.Then,the loss function is improved,and the generalized intersection ratio between the prediction box and the true value is used to solve the problem that the optimization cannot continue when the intersection ratio is zero.The experimental results show that compared with other models,the proposed model can improve the recognition accuracy and reduce the model size.