Surface Defect Detection: An Approach Utilising Feature-Based Transfer Learning

Surface defects in manufacturing processes pose significant challenges, affecting product quality and safety. Traditional labour-based inspection is deemed to be ineffective and has led the shift to computer vision-based solutions and to a certain extent, the employment of artificial intelligence. In the present study, we leverage the capability of a pre-trained convolutional neural networks model, i.e. VGG19, in extracting the features from a set of surface defect dataset that comprises six unique defect categories. The ability of different machine learning models, namely Logistic Regression (LR), Random Forest (RF), k-Nearest Neighbour (kNN) and Support Vector Machine (SVM), to classify the defects was investigated. It was demonstrated from the study that the VGG-19 + LR combination is the optimal pipeline. This study suggests that the feature-based transfer learning approach is an attractive approach to be employed for surface defect detection.