Research on Approaches for Computer Aided Detection of Casting Defects in X-ray Images with Feature Engineering and Machine Learning

Abstract

X-ray testing has been adopted as the principal non-destructive testing approach to identify defects within a casting component. However, manual detection for X-ray images carried out by operator or expert always tends to be time-consuming, subjective and error-prone. Intelligent inspection techniques based on computer vision, which have been broadly employed in object recognition with promising results in optical natural images, provides a new idea for computer aided detection of casting defects in X-ray images. In this paper, we compare and evaluate several methods, most of which have not been researched for computer aided detection of casting defects in X-ray images and are based on different feature engineering methods and machine learning models, including local binary patterns-SVM, Gabor-SGD, histogram of oriented gradient-random forest and combination among them, to pursue an approach with better performance on detection of casting defects in X-ray images from the our InteCAST dataset. The experimental results demonstrate that the best performance was acquired by LBP feature and an ensemble learning model, which indicates that the approach proposed provides valuable reference for solving the problems in manual detection.