Vehicle component defect detection method based on transformer and multi-scale feature fusion.

Area of Science:

• Automotive Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Vehicle component defect detection is crucial for automotive safety and reliability.
• Traditional methods face challenges with complex, multi-scale defects and capturing both local and global features.
• Existing models often exhibit suboptimal performance in real-world automotive scenarios.

Purpose of the Study:

• To propose a novel defect detection method for vehicle components.
• To address limitations of traditional methods in handling complex and multi-scale defect features.
• To improve the accuracy, robustness, and real-time processing of defect detection in automotive systems.

Main Methods:

• A novel defect detection method combining Transformer architecture and multi-scale feature fusion.
• Leveraging Transformer's global feature learning capability.
• Integrating multi-scale features to enhance detection of varying defect sizes and complexities.

Main Results:

• The proposed model demonstrates superior performance compared to traditional defect detection methods.
• Achieved higher accuracy and robustness in detecting complex, multi-scale defects across various vehicle parts.
• Showcased significant improvements in real-time processing capabilities for practical deployment.

Conclusions:

• Transformer-based architectures combined with multi-scale feature fusion offer a powerful solution for defect detection.
• The developed method provides a more efficient and accurate approach for vehicle maintenance and safety.
• This study validates the potential for enhanced defect detection in automotive quality control.