Steel surface defect detection method based on improved YOLOv9.

Area of Science:

• Materials Science and Engineering
• Computer Vision and Artificial Intelligence
• Industrial Automation

Background:

• Steel surface defect detection is vital for quality control in industrial automation.
• Diverse defect types and sizes, especially small ones, present significant detection challenges, leading to high error rates.
• Existing methods struggle with subtle, small defects, impacting production efficiency and product quality.

Purpose of the Study:

• To develop an improved steel surface defect detection algorithm based on YOLOv9.
• To enhance the detection accuracy and efficiency of small-sized defects.
• To reduce computational complexity and improve multi-scale target detection capabilities.

Main Methods:

• Implemented Depthwise Separable Convolution (DSConv) to reduce model complexity.
• Integrated the C3 module for effective multi-level feature fusion and multi-scale target detection.
• Incorporated a bidirectional feature pyramid network (BiFPN) and DySample upsampling for improved small target feature extraction and localization.

Main Results:

• Achieved a mean average precision (mAP) of 78.2%, an increase of 1.8% over the baseline.
• Reached an accuracy of 82.5%, a 7.4% improvement compared to the baseline model.
• Reduced the number of model parameters by 8.9% while enhancing performance.

Conclusions:

• The improved YOLOv9 algorithm effectively addresses challenges in steel surface defect detection, particularly for small targets.
• The proposed enhancements lead to significant improvements in detection accuracy, localization, and computational efficiency.
• This research offers practical value for advancing quality control in intelligent manufacturing and steel production.