Solder Joint Defect Detection Based on Image Segmentation and Deep Learning

Image feature extraction and machine learning methods are used to detect and identify PCB solder joints. The normal/abnormal classification of solder joints are realized. What is more, the maximal class variance method (OTSU) is used to distinguish solder joints from background, and the Morphology was used to extract the suspected solder joints. The positive and negative samples were labeled and then the convolution neural network (CNN) was used to classify the test samples. Compared with the support vector machine (SVM) method, which extracts HOG features, and the fisher discriminant method based on the circular degree feature, the neural network has better classification effect.