Detection And Recognition Of Defects Of Manual Welding PCB

In modem electronic industry, Printed Circuit Board(PCB) is the carrier of different electronic components. The quality of the components mounted on board affects the performance and reliability of electronic products, so it’s quite valuable to inspect the quality of PCB in industrial manufacture. Application of automated inspect technology based on image processing aims at saving human labor and improving reliability, thus promotes the efficiency and mass as well as the reduction of cost, and inspects the defects of PCB with high accuracy and non-contact.The paper study PCB manual welding solders defect detection and recognition algorithm and select the ordinary manual welding PCB as the research object. PCB solder joint images are obtained after scanning and solder joints registration. In order to reduce the solder joint images noise, they need to be preprocessed like gray processing and median filtering. The paper study the solder joint images feature extraction method, proposed threshold image segmentation technique to extract geometric features of solder joint images. According to the results of image threshold segmentation to extract the foreground and background images feature points. After classifying the color images based on SVM, acquire gray images which contain only the foreground image, then extract their wavelet features.In order to improve the method of manual solder joints defect recognition, an combination algorithm of fuzzy C-means clustering(FCM) and relaxed support vector machine(RSVM) based on feature aggregation degree is proposed. Firstly, the samples characteristics were extracted based on FCM algorithm according to the different membership and the feature aggregation degrees were calculated in the same time. Then the slack variable parameter of RSVM algorithm was repaired based on the feature aggregation degree to establish the final classification model. The experimental results show that the proposed algorithm can effectively reduce the effect of noise or blur point on the classification model, and build stronger generalization classification model to improve the accuracy of defect recognition.