基于深度学习的电路板焊接异常检测算法研究

焊点的焊接质量决定了电路板的可靠性,而电路板焊接异常的快速检测是大批量生产的先决条件。为了快速地实现焊接异常的精确检测,提出了一种基于深度学习的焊点图像识别算法。该算法通过自适应矩估计配合加速卷积神经网络实现,可对大量焊接图片进行快速分类识别检测。实验选取5 000幅焊接图像训练集测试,并与传统的K-means聚类算法和Canny边缘检测算法对比。实验结果显示,在小球和连桥缺陷中3种方法效果相近,而在虚焊、少锡缺陷中,本算法具有明显优势。在1 000组测试集实验中,其综合检出率及召回率分别达97.92%和98.21%,明显优于传统方法,验证了本算法具有更好的应用前景。

Research on Circuit Board Welding Anomaly Detection Algorithm Based on Deep Learning

The quality of the solder joints determines the reliability of the board,and the rapid detection of abnormal board soldering is a prerequisite for mass production. In order to quickly and accurately detect welding anomalies,a kind of image recognition algorithm for solder joint based on deep learning is proposed. The algorithm is implemented by adaptive moment estimation combined with accelerated convolutional neural network,and it can quickly classify and detect a large number of welding pictures. In the experiment,5 000 welding images were selected for the training set test,and compared with the traditional K-means clustering algorithm and Canny edge detection algorithm. The experimental results show that the three methods have similar effects in the small ball and bridge defects,but the algorithm has obvious advantages in the case of virtual welding and less tin defects. In the 1 000 sets of test set experiments,the comprehensive detection rate and recall rate were 97.92% and 98.21%,respectively. It is obviously superior to the traditional method,and the algorithm has a good application prospect.