A Novel Fault Inspection Method of Steel Plate Surface

Abstract

Defect detection is an essential requirement for quality control in steel plate manufacturing. Traditional defect detection methods use the classic image process, high labor cost, and inefficient detection ability. This paper proposes a novel image detection model for steel plate surface defect detection called ID-RCNN. This model builds a new network based on Dynamic RCNN. In detail, we use a ResNet50 with an attention module as the backbone for feature extraction to better detect surface defects in steel plate production. Then, we proposed a novel attention module called CM, which is improved for Convolutional Block Attention Module (CBAM). The experimental results show that this model is more suitable for use in production than other steel plate surface defect detection methods, with the mean Average Precision of defect detection with 99.1% accuracy. We have tested in the DAGM defects dataset, and experiments have shown that this model is equally valid. This research can greatly improve the ability of Deep Learning models in industrial defect detection.