Abstract: The goal of this study is to classify the welds joint defects capture by CCD camera into three categories which are good welds, excess welds and insufficient welds in three weld joint shapes; straight lines, curve and tooth saw. Firstly, we extract the features characteristic from the input images represent in 2D gray values of coocurrence matrix and gray absolute histogram of edge amplitude consists of energy, correlation, homogeneity and contrast. Then zooming input image by 0.5 to calculated the next characteristic feature values. Furthermore, use the Support Vector Machine (SVMs) classifier to classify the welds joint defect according to the feature vector belongs to the same categories as the training data. The experimental result taken from 45 welds joints samples in three welds joint shapes; straight lines, curve and tooth saw where 3 samples as training set and 2 samples as testing set show that the proposal approach able to classify the welds joint defects effective automatically.
Hairol Nizam Mohd Shah, Marizan Sulaiman and Ahmad Zaki Shukor, 2017. Automatic Classification Vision Based System for Welds Joints Defects Using Support Vector Machine (SVMs). International Journal of Soft Computing, 12: 66-71.