YOLv5s-DDC: A Rapid Detection Network for Strip Steel Surface Defects Based on Deformable Convolution with Coordinate Attention and Distribution Shifting Convolution

Surface defect detection in strip steel is essential for improving product quality. Current methods using machine vision and image processing struggle with multiscale variations, irregular distributions, and complex shapes. This paper proposes the YOLOv5s-DDC algorithm based on deformable convolution, distribution shifting convolution, and attention mechanism. To tackle complex shapes and small targets in strip surface defects, this paper introduces the C3-DC in the backbone, which integrates deformable convolution and coordinate attention mechanisms. This module can adapt to changes in the shape and size of strip surface defects. Additionally, this paper introduces the C3-DSConv in the neck, integrating YOLOv5s's C3 with distribution shifting convolution. C3-DSConv improves strip surface defect detection without increasing parameters and computations. Experimental results show that YOLOv5s-DDC reaches an [email protected] of 80.4% on the NEU-DET dataset, improving by 4.2% over the original YOLOv5s. The algorithm has a fast convergence speed and meets the needs of the metallurgical industry.