Image segmentation for automatic detection and detectability evaluation of thinning defects using LSM-based induction thermography

Abstract

The infrared thermography (IRT) technique is a highly reliable and attractive technique that can evaluate a large area in real-time without destroying the inspection object. This study used the line scanning method (LSM)-based induction thermography (IT) technique among active IRT to detect thinning defects in S275 steel specimens. Unlike the dynamic mode, LSM was applied to perform sequence processing so that the copper coil shape was removed in the 2D thermal images processed. After image segmentation, a binary image was acquired using the Otsu algorithm. By utilizing image segmentation, the applied threshold for each defect is different when performing the Otsu algorithm. This makes it possible to compare defect detection according to the presence or absence of image Segmentation. Automatic defect detection was performed using a Boundary Tracking algorithm, and 12 defect detections were confirmed. Accuracy was evaluated through pixel calculation, and defects in A, B, and D columns showed high detection accuracy. This study presented the principle of LSM in which includes the relative movement between the IR camera and the specimen. In addition, a mechanism to increase the number of defects detected through image segmentation was presented.