The Steel Defect Detection Algorithm Based on YOLOv5-CGC

To address the issue of poor detection accuracy for small targets in steel surface defect detection, this paper improves upon YOLOv5 by integrating the Global Context (GC) Block with the C3 structure in the Backbone, resulting in a new feature extraction structure called C3GC. Additionally, the original Upsample operator is replaced with the CARAFE upsampling operator to enhance the model's ability to detect small targets. Results show that the improved YOLOv5-CGC algorithm achieved a mAP of 78.4% on the NEU-DET dataset, which is a 3.0% increase over the original YOLOv5. Detection accuracy for small targets, such as cracks and rolled-in scale, improved the most, with increases of 22.2% and 8.4%, respectively. Compared with mainstream detection algorithms, YOLOv5-CGC achieved superior performance.