AEDD-Net: YOLOv5-Based Aeroengine Defect Detection Model with Robust Strategy

Aeroengine defect detection (AEDD) has played important roles in guaranteeing the safety of aircraft, and in this paper, a novel YOLOv5-based AEDD-Net is proposed, which effectively addresses the issues of low efficiency and inaccuracy in the traditional borescope inspection method. In particular, a series of robust strategies are designed to enhance the model performance, including a dual-domain attention mechanism, dense connection structure, and the improved loss function. Evaluation results on a self-build dataset with well-defined annotations show that the proposed AEDD-Net obtains the precision, recall, and mean average precision (at threshold 0.5 of intersection over union) of 97.40%, 97.80% and 98.64%, respectively, which realizes accurate and efficient automated defect detection in high-end equipment inspection scenes.