Enhancing Precision with an Ensemble Generative Adversarial Network for Steel Surface Defect Detectors (EnsGAN-SDD)

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Steel product quality is significantly impacted by defects, which are often vague and small.
• Developing accurate defect detectors for steel is challenging due to these characteristics.

Purpose of the Study:

• To propose a novel approach, EnsGAN-SDD, for enhanced steel defect detection.
• To improve the accuracy and efficiency of identifying subtle defects in steel products.

Main Methods:

• Utilized ensemble of enhanced super-resolution generative adversarial networks (ESRGAN) for image preprocessing.
• Employed De-tectoRS, a state-of-the-art feature pyramid network, for multi-scale feature extraction.
• Incorporated Side-Aware Boundary Localization for precise defect prediction.

Main Results:

• The EnsGAN-SDD approach demonstrated superior performance compared to existing state-of-the-art methods.
• Achieved improved accuracy in detecting vague and small steel defects.
• Showcased enhanced processing efficiency over the original ESRGAN.

Conclusions:

• EnsGAN-SDD offers a significant advancement in steel defect detection technology.
• The proposed method has the potential to substantially contribute to improved steel production quality.
• This innovation addresses key challenges in identifying subtle defects, enhancing industrial inspection capabilities.