A novel method for the automatic classification of defects using magnetic flux leakage inspection is presented. A technique based on geometric measures to distinguish between different defects due to petro-chemical tank corrosion is presented. In order to characterize a defect, a process of feature extraction is proposed. Principal component analysis is then used to select the most powerful set of features.

The performance is compared using two different methods: k-nearest neighbor and support vector machine. The results show an accuracy of 91% with which automatic classification is possible on unseen test examples on steel plates.