Steel surface defect detection algorithm based on ESI-YOLOv8

Steel surface defect detection algorithm based on ESI-YOLOv8

14 May 2024 | Xinrong Zhang, Yanlong Wang, Huaisong Fang
This paper introduces an enhanced version of the YOLOv8 algorithm, named ESI-YOLOv8, to improve the precision of detecting defects on steel plate surfaces. The ESI-YOLOv8 algorithm incorporates several novel modules and optimizations to address the limitations of the original YOLOv8 model, such as low detection accuracy, high computational requirements, and slow convergence. Key contributions include the introduction of the EP module, which reduces redundant computations and model parameters, and the SPPF-LSKA module, which integrates the Large Separable Kernel Attention module and the Spatial Pyramid Pooling module to enhance multi-scale feature fusion. Additionally, the INNER-CIOU loss function is introduced to improve the scale size of the auxiliary border, thereby enhancing detection speed and accuracy. Experimental results show that the ESI-YOLOv8 model achieves an average accuracy of 78%, a 3.7% improvement over the original YOLOv8, with reduced model parameters and computational requirements. The model's performance is validated on the CoCo dataset, achieving an average accuracy of 77.8%. The paper also includes a comparative analysis with other mainstream models, demonstrating the ESI-YOLOv8 model's superior performance in terms of accuracy, parameter count, and computational efficiency.This paper introduces an enhanced version of the YOLOv8 algorithm, named ESI-YOLOv8, to improve the precision of detecting defects on steel plate surfaces. The ESI-YOLOv8 algorithm incorporates several novel modules and optimizations to address the limitations of the original YOLOv8 model, such as low detection accuracy, high computational requirements, and slow convergence. Key contributions include the introduction of the EP module, which reduces redundant computations and model parameters, and the SPPF-LSKA module, which integrates the Large Separable Kernel Attention module and the Spatial Pyramid Pooling module to enhance multi-scale feature fusion. Additionally, the INNER-CIOU loss function is introduced to improve the scale size of the auxiliary border, thereby enhancing detection speed and accuracy. Experimental results show that the ESI-YOLOv8 model achieves an average accuracy of 78%, a 3.7% improvement over the original YOLOv8, with reduced model parameters and computational requirements. The model's performance is validated on the CoCo dataset, achieving an average accuracy of 77.8%. The paper also includes a comparative analysis with other mainstream models, demonstrating the ESI-YOLOv8 model's superior performance in terms of accuracy, parameter count, and computational efficiency.
Reach us at info@study.space
[slides] Steel surface defect detection algorithm based on ESI-YOLOv8 | StudySpace