ODNet: A High Real-Time Network Using Orthogonal Decomposition for Few-Shot Strip Steel Surface Defect...

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Strip steel surface defect classification is vital for industrial production.
• Deep learning models face challenges with limited, redundant defect data.
• Existing methods struggle with real-time accuracy in few-shot scenarios.

Purpose of the Study:

• To develop a high real-time network for few-shot strip steel surface defect classification.
• To address data acquisition and redundancy issues in defect detection.
• To enhance classification accuracy and efficiency.

Main Methods:

• Introduced ODNet (Orthogonal Decomposition Network) using ResNet backbone.
• Employed orthogonal decomposition to reduce feature redundancy.
• Integrated skip connections to retain essential sample correlations.
• Utilized Euclidean distance for optimized parameter efficiency.

Main Results:

• ODNet demonstrated superior real-time performance and accuracy on the FSC-20 benchmark.
• The network effectively handled challenges of few-shot defect classification.
• Orthogonal decomposition reduced redundant information while preserving key correlations.

Conclusions:

• ODNet offers an effective solution for real-time, accurate few-shot strip steel surface defect classification.
• The proposed methods overcome limitations of data scarcity and redundancy.
• ODNet shows strong generalization capabilities compared to existing approaches.