A frequency-aware lightweight network with wavelet efficient layer aggregation for defect detection of aero-engine blade

Borescope inspection is a commonly used method for detecting aero-engine blade damage, which still heavily relies on manual operation. Although existing intelligent borescope inspection methods have achieved notable improvements in detection accuracy, their high computational complexity limits practical deployment in resource-constrained scenarios. To address this issue, this paper proposes a Wavelet-based Convolution Module (WConv) derived from discrete wavelet transform theory, which enables efficient multi-scale feature extraction by jointly capturing low-frequency structural information and high-frequency detail features. In addition, a Wavelet Efficient Layer Aggregation Network (WELAN) is designed based on gradient path design and three-channel image representation, effectively enhancing gradient diversity while reducing computational complexity and model parameter size. By integrating these modules with the YOLO object detection framework, we construct a lightweight defect detection model, YOLOv9t-WELAN. To evaluate the effectiveness of the proposed method, we build a borescope dataset of aero-engine blade damage and conduct comparative experiments, ablation studies, visualization analyses and module placement analysis. Experimental results demonstrate that YOLOv9t-WELAN achieves a detection accuracy of 0.98 in mAP@0.5 while reducing model size by 25% compared with the baseline, transfer experiments on a public real-damage dataset further confirm its cross-domain generalization ability, validating its effectiveness in realizing lightweight design without sacrificing performance.