An Aero-Engine Defect Detection Method via Sequential Matching and Historical Experience-Guided Transfer

Key Points

• Detection precision improves significantly with the proposed transformer-based method, leading to better performance in challenging conditions.
• The method shows a 5% improvement in mean Average Precision at multiple thresholds, compared to existing baseline models like Align-DETR.
• This analysis employs a unique matching consistency loss and adaptive cross attention for enhanced defect detection and convergence stability.
• The findings highlight the model's robustness and generalization in complex environments, suggesting its practical application in engineering solutions.

Abstract