Diamond-DETR: lightweight real-time quality evaluation algorithm for synthetic diamonds.

Area of Science:

• Materials Science
• Computer Vision
• Artificial Intelligence

Background:

• Synthetic diamond manufacturing can introduce subtle defects affecting quality and value.
• Accurate defect detection is challenging due to diamond's complex geometry and varied defect scales.
• Existing methods may struggle with accuracy and efficiency in resource-constrained environments.

Purpose of the Study:

• To develop an advanced target detection algorithm for synthetic diamond quality evaluation.
• To improve defect detection accuracy and model generalization in resource-limited scenarios.
• To create a computationally efficient model for real-time industrial inspection.

Main Methods:

• Proposed Diamond-DETR algorithm optimizing backbone with lightweight multi-scale feature extraction and RepFasterNet blocks.
• Incorporated an encoder with a screening-feature fusion pyramid network and channel attention for multi-scale feature fusion.
• Introduced a cross-stage fusion module with dilated convolutions to enhance long-range dependency perception.

Main Results:

• Diamond-DETR demonstrated superior parameter efficiency, inference speed, and detection accuracy compared to RT-DETR.
• The model is well-suited for deployment in resource-constrained inspection scenarios.
• Achieved competitive performance on an external industrial dataset, showing cross-dataset applicability.

Conclusions:

• Diamond-DETR offers an effective solution for accurate and efficient synthetic diamond defect detection.
• The algorithm's performance makes it suitable for industrial applications with limited computational resources.
• The study highlights the potential for AI-driven quality control in the synthetic diamond industry.