Metal Surface Defect Detection using Object Detection Models

Metals are widely used in the manufacturing industry to manufacture a variety of products and it is important to visually inspect the end products. Companies incur financial losses and lose customer trust if defective products are delivered by them. Moreover, current inspection processes are highly manual and time-consuming. To overcome these issues, we propose a method to identify the various types of defects like creases, oil stains, etc. as well as locate them accurately. We have created our custom dataset for defect detection on metal surfaces by merging and processing images from two datasets, NEV-DET, and GC10-DET. State-of-the-art object detection models like YOLOv5, YOLOv7, and Detectron2 (Faster RCNN) have been implemented in this paper. We were able to achieve mAP scores of 77.4%, 77.7%, and 43.4% respectively.