ENGINEERING JOURNAL CHIANG MAI UNIVERSITY

Volume 26, No. 03, Month DECEMBER, Year 2019, Pages 224 - 235

Comparison of image segmentation techniques for pcb quality inspection

Jakkrit Onshaunjit and Jakkree Srinonchat

Abstract Download PDF

Quality processes control plays an important role not only in Electronics industry but also in the electronics piece parts manufacturing. Printed Circuit Board (PCB) fabrication is one of the beginning processes to produce the electronics piece parts that is required a sophisticated quality control inspection technique. This is to ensure that defective PCBs can be rejected in early stages which it is resulting in higher manufacturing cost than that beginning rejected PCB screening out. An automation of Printed Circuit Board’s quality control inspection with a digital image processing technique is one of the introduced methods that it is not only solving that mentioned higher cost but also eliminating an inspection error from human eyes inspection method because it provides higher efficiency, accuracy and less time. One of steps in Digital Image Processing that plays an important role, and also determines the accuracy of PCB quality inspection is Image Segmentation. This research presents a comparison of different image segmentation techniques for PCB quality inspection, in which there are 4 methods as follows 1) Global Thresholding, 2) the Adaptive Thresholding, 3) Kmeans Clustering and 4) the Fuzzy C-means Clustering. There are a total of 500 images of PCB’s that used as the input of this experiment. In this study, the Intersection over Union, Accuracy, Precision, Recall and Elapse Time parameters are used to evaluate the performance of those 4 methods mentioned above. The results show that the Fuzzy C-means Clustering and K-means Clustering methods provide the accuracy of 98.57 % and 97.43% respectively, in which they are better than other methods. Even though the Fuzzy C-means Clustering method gives the best accuracy but it takes much longer processing time than others. Therefore, an improving of K-means Clustering method that provides the second best in accuracy is initiated. In order to improve the accuracy of K-means Clustering method, a dividing of cluster with two iterations under K-means Clustering method is evaluated. The results show that mentioned improving method provides better accuracy of 98.38%, while processing time is slightly increased. Hence, Improve K-means clustering method is considered as an appropriated method for further study and analysis as well as to use for PCB’s quality inspection in piece part electronics manufacturing.

Keywords

Image segmentation, Image processing, PCB Inspection, Threshold, Clustering

---