ILCS: An Improved Lightweight Convolution Structure and Mixed Interactive Attention for Steel Surface Defect...

Area of Science:

• Materials Science
• Computer Science
• Artificial Intelligence

Background:

• Steel surface defect classification is crucial for industrial quality control.
• Existing deep learning models face challenges with limited computing resources in production environments.
• Developing lightweight models for rapid and accurate classification is essential.

Purpose of the Study:

• To propose an improved lightweight convolution structure (LCS) for efficient steel surface defect classification.
• To enhance classification accuracy by integrating attention mechanisms.
• To reduce model parameters and computational load for edge deployment.

Main Methods:

• Developed a novel lightweight convolution structure (LCS) using separable convolutions, depthwise convolutions, and point-wise convolutions.
• Integrated spatial and channel attention mechanisms to mitigate accuracy loss from lightweight convolutions.
• Introduced a mixed interactive attention module (MIAM) to further boost feature extraction.

Main Results:

• The proposed method significantly reduces the number of model parameters and computational complexity.
• Achieved higher recognition accuracy compared to traditional deep learning models.
• Demonstrated the effectiveness of the LCS and MIAM for lightweight steel surface defect classification.

Conclusions:

• The developed lightweight model successfully balances efficiency and accuracy for steel surface defect classification.
• The approach is well-suited for deployment on edge devices with limited computational power in industrial settings.
• This work contributes to advancing automated quality control in steel manufacturing.