LAUSR

Abstract X-ray imaging is proven effective in the visualization of defects inside the Aluminum Conductor Composite Core (ACCC) wires. Although object detection pipelines have been extensively considered in the nondestructive testing tasks, the difficulty in obtaining defect samples has become the main obstacle to the application of such methods in the task of automatic defect detection for ACCC wires X-ray images. In this paper, we conducted a new semi-supervised approach based on anomaly detection. Different from the commonly used supervised methods, the proposed method requires only samples without defects for the learning process, therefore we are no longer limited by the insufficient and unbalanced defect samples. Experimental results show that the accuracy of the proposed method is up to 0.761, which proves the effectiveness of the method in the automatic defect detection of ACCC wires X-ray images.