Online Detection of Surface Defects Based on Improved YOLOV3

Area of Science:

• Computer Vision
• Machine Learning
• Materials Science

Background:

• Traditional product surface defect detection methods suffer from low efficiency and accuracy.
• Automated inspection systems are crucial for quality control in manufacturing.

Purpose of the Study:

• To develop a high-efficiency and high-accuracy online surface defect detection method.
• To improve the performance of existing object detection models for industrial applications.

Main Methods:

• Utilized YOLOV3 architecture with MobileNetV2 as a lightweight backbone for faster feature extraction.
• Proposed an Extended Feature Pyramid Network (EFPN) for multi-size object detection.
• Introduced a novel Feature Fusing Module (FFM) for enhanced feature resolution and detail capture.
• Incorporated an IoU loss function to address bounding box regression inaccuracies.

Main Results:

• The proposed method achieved 86.96% mean Average Precision (mAP) on the NEU-DET dataset.
• The system demonstrated a processing speed of 80.96 Frames Per Second (FPS).
• Outperformed existing algorithms in both accuracy and speed for steel surface defect detection.

Conclusions:

• The developed YOLOV3-based method offers a superior balance between performance and computational cost.
• The approach enables real-time and high-precision inspection of product surface defects.
• The EFPN and FFM modules significantly contribute to improved detection capabilities.