Steel surface defect detection algorithm based on improved YOLOv10.

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Steel surface defect detection is a critical area in industrial quality control.
• Existing deep learning methods face challenges with high computational complexity and feature loss in noisy industrial environments.
• Real-time monitoring requires efficient and accurate defect detection systems.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for steel surface defect detection.
• To address the limitations of high computational cost and feature loss in current methods.
• To enable real-time defect monitoring in complex industrial settings.

Main Methods:

• Proposed a novel latent-space attention multi-scale YOLOv10n (LAM-YOLOv10n) model.
• Integrated a lightweight ghost module to reduce computational complexity and parameters.
• Introduced a spatial multi-scale attention (SMA) module for enhanced feature extraction.
• Implemented a multi-branch feature fusion network (MFFN) for improved multi-scale feature aggregation.

Main Results:

• The LAM-YOLOv10n model achieved a 3.47% improvement in precision over the baseline YOLOv10n.
• Demonstrated superior performance compared to state-of-the-art object detection models in accuracy and efficiency.
• Validated the model's effectiveness in detecting various steel surface defects.

Conclusions:

• The proposed LAM-YOLOv10n model offers an effective and practical solution for real-time steel surface defect detection.
• The model's lightweight design and enhanced feature extraction capabilities make it suitable for industrial applications.
• This research contributes to advancing machine vision techniques for quality control in steel manufacturing.