Defect detection in code characters with complex backgrounds based on BBE.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Traditional defect detection methods struggle with high-quality images and simple backgrounds.
• Existing computer vision techniques often fail on complex backgrounds and low-contrast characters.

Purpose of the Study:

• To develop a robust deep learning approach for accurate defect detection of code characters on complex backgrounds.
• To enhance defect detection capabilities for industrial applications, specifically in the plastic container industry.

Main Methods:

• Utilized image processing and data augmentation to create a large, balanced dataset of defect samples.
• Developed a novel object detection network, BBE, leveraging the EfficientNet architecture.
• Implemented quality inspection standards for individual character detection to assess overall code quality.

Main Results:

• Achieved a mean Average Precision (mAP) of 0.9961 and an accuracy of 0.9985.
• Demonstrated high accuracy, speed, robustness, and transferability for defect detection tasks.
• Successfully applied the BBE network to defect inspections in the real plastic container industry.

Conclusions:

• The proposed deep learning method significantly improves defect detection accuracy and efficiency for code characters on complex backgrounds.
• The BBE network offers a robust and transferable solution for industrial defect inspection, validated in the plastic container sector.
• This work represents the first application of the BBE network for defect inspections in the plastic container industry.