Improving the lightweight VTG-YOLOv7-tiny for steel surface defect detection
Information Sciences|更新时间:2024-05-08
|
Improving the lightweight VTG-YOLOv7-tiny for steel surface defect detection
“Significant breakthroughs have been made in the field of steel defect detection. Researchers have proposed a lightweight VTG-YOLOv7 tiny detection algorithm to address the diverse forms and complex structures of surface defects in steel. This algorithm enhances feature fusion capability by designing a VoVGA FPN network; Building a triple coordinate attention mechanism to enhance feature extraction capability; Introducing ghosting mixed convolution to reduce the number of model parameters and computational complexity; Add a large target detection layer to improve detection accuracy. Experimental verification shows that the improved algorithm improves mAP by 5.7% and 8.5% on NEU-DET and Severstal datasets, reduces parameter and computational complexity by 0.61M and 4.2G, respectively, and significantly improves accuracy and recall. This achievement provides a new solution for steel defect detection and is expected to provide efficient and accurate detection capabilities for edge terminal equipment.”
Optics and Precision EngineeringVol. 32, Issue 8, Pages: 1227-1240(2024)