Steel surface defect detection based on lightweight model

Abstract

Steel surface defect detection, with the goal of enhancing the quality of steel within the industry, is a vital process in steel production. Over the years, a number of techniques have been developed to meet the challenge of object detection. Nonetheless, designing lightweight models that can detect defects quickly and accurately remains challenging due to a significant amount of environmental interference and limited computational resources of the edge devices used in steel factories. In this paper, In this paper, we propose an improved Lightweight Detection Network(LDN) for steel surface defect detection. First, to extract features, the lightweight MobileNetV2 model is employed as the backbone. Second, to improve the detection accuracy, a feedback mechanism from the feature pyramid is added to the backbone network to integrate the features from shallow and deep layers. Additionally, the Efficient Intersection over Union(EIoU) loss is utilized to make the regression more accurate. The experimental results indicate that the method's accuracy in recognition surpasses that of previous models, and the parameters and computational requirements have decreased significantly, leading to a rapid and precise detection of steel surface defects.