Flaw Detection in White Porcelain Wine Bottles Based on Improved YOLOv4 Algorithm.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Materials Science

Background:

• Traditional white porcelain wine bottle flaw detection methods suffer from low accuracy and slow speeds.
• Automated defect detection is crucial for quality control in ceramic manufacturing.

Purpose of the Study:

• To develop an improved flaw detection algorithm for white porcelain wine bottles.
• To enhance both the accuracy and speed of defect identification.

Main Methods:

• An improved flaw detection algorithm based on YOLOv4 was proposed.
• Coordinate Attention was integrated into the backbone for better feature extraction.
• Deformable convolution was employed for precise flaw localization.
• Efficient Intersection over Union (EIoU) replaced Complete Intersection over Union (CIoU) in the loss function.

Main Results:

• The proposed algorithm effectively detects surface flaws on white porcelain wine bottles.
• The model achieved a mean Average Precision (mAP) of 92.56%.
• The detection speed reached 37.17 frames per second (fps).

Conclusions:

• The enhanced YOLOv4 algorithm significantly improves white porcelain wine bottle flaw detection.
• The integration of Coordinate Attention, deformable convolution, and EIoU contributes to higher accuracy and faster processing speeds.