XU Chao, WANG Wenzhe, JIANG Zenghua, et al. BD‑YOLO: A deep learning‑based model for blade damage detection in aero‑engine borescope images[J]. Journal of Nanjing University of Aeronautics & Astronautics(Natural Science Edition),2026, 58(2):372⁃379.
XU Chao, WANG Wenzhe, JIANG Zenghua, et al. BD‑YOLO: A deep learning‑based model for blade damage detection in aero‑engine borescope images[J]. Journal of Nanjing University of Aeronautics & Astronautics(Natural Science Edition),2026, 58(2):372⁃379. DOI: 10.16356/j.2097-6771.2026.02.013.
BD‑YOLO: A Deep Learning‑Based Model for Blade Damage Detection in Aero‑engine Borescope Images
针对航空发动机孔探图像叶片损伤方向任意且细长损伤易引入过多背景干扰、导致定位精度下降的问题,本文提出一种基于改进You only look once version 8(YOLOv8)的旋转目标检测模型BD‑YOLO。首先,设计融合跨阶段连接(Cross stage partial,CSP)与密集连接扩张卷积模块(Receptive field enhancement module,RFEM)的小目标检测跨阶段局部感受野增强模块(Cross stage partial receptive field enhancement module,CSRFEM),增强对细小损伤的特征提取能力。其次,在颈部网络引入改进的双向特征金字塔网络SimBiFPN,实现多尺度特征的高效融合。最后,在头部网络增设专用小目标检测头,提升小尺寸损伤的识别精度。实验结果表明,BD‑YOLO的平均精度均值(Mean average precision,mAP)50、mAP75和mAP50~95分别达到98.6%、84.3%和63.3%,检测速度为34帧/s,能够实现叶片损伤的高精度实时检测。
Abstract
To address the issues in aero-engine borescope images, such as the arbitrary orientation of blade damage and the tendency for slender damage to introduce excessive background interference, leading to reduced localization accuracy, this paper proposes a rotated object detection model, BD-YOLO, based on an improved You Only Look Once version 8 (YOLOv8). Firstly, a small object detection module named cross stage partial receptive field enhancement module (CSRFEM), which integrates the cross stage partial (CSP) and receptive field enhancement module (RFEM), is designed to enhance feature extraction capabilities for minor damages. Secondly, an improved bidirectional feature pyramid network, SimBiFPN, is introduced into the neck network to achieve efficient multi-scale feature fusion. Finally, a dedicated small object detection head is added to the head network to improve the recognition accuracy of small-sized damages. Experimental results demonstrate that BD-YOLO achieves mean average precision (mAP)50, mAP75, and mAP50-95 values of 98.6%, 84.3%, and 63.3%, respectively, with a detection speed of 34 frames per second, enabling high-precision real-time detection of blade damage.
WANG Xiaobing . Research on civil aviation engine maintenance interval optimization based on borescope inspection [D]. Tianjin : Civil Aviation University of China , 2018 .
LU Hongtao , LUO Mukun . Survey on new progresses of deep learning based computer vision [J]. Journal of Data Acquisition and Processing , 2022 , 37 ( 2 ): 247 - 278 .
SUN Han , LIU Yishan , LIN Yuhan . Deep learning based salient object detection: A survey [J]. Journal of Data Acquisition and Processing , 2023 , 38 ( 1 ): 21 - 50 .
KUANG Kejia . Research on deep learning and its application on the defects detection for aeroengine [D]. Guangzhou : South China University of Technology , 2017 .
CAO Siyan , LIU Junqiang , SONG Gaoteng , et al . Borehole image detection of aero-engine based on self-attention semantic segmentation model [J]. Journal of Beijing University of Aeronautics and Astronautics , 2023 , 49 ( 6 ): 1504 - 1515 .
HE Yuhao , CAO Xueguo , LIU Xinliang , et al . Real-time detection of aircraft engine blade damage based on SW-YOLO model [J]. Journal of Propulsion Technology , 2024 , 45 ( 2 ): 197 - 203 .
HUANG Rui , CHENG Xuyi , WANG Ruofei , et al . Adaptive parameters based borescope image-related damage detection of aeroengine [J]. Journal of Image and Graphics , 2023 , 28 ( 10 ): 3136 - 3148 .
SHANG H , WU J , SUN C , et al . Global prior transformer network in intelligent borescope inspection for surface damage detection of aeroengine blade [J]. IEEE Transactions on Industrial Informatics , 2023 , 19 ( 8 ): 8865 - 8877 .
LLERENA J M , ZENI L F , KRISTEN L N , et al . Gaussian bounding boxes and probabilistic intersection-over-union for object detection [J]. IEEE Transactions on Image Processing , 2024 , 33 : 671 - 681 .
WANG C Y , LIAO H Y M , YEH I H , et al . CSPNet: A new backbone that can enhance learning capability of CNN [EB/OL]. ( 2019-11-27 ). https://arxiv.org/abs/1911.11929 https://arxiv.org/abs/1911.11929 .
HAN Y , LIAO J , LU T , et al . KCPNet: Knowledge-driven context perception networks for ship detection in infrared imagery [J]. IEEE Transactions on Geoscience and Remote Sensing , 2023 , 61 : 1 - 19 .
YANG L , ZHANG R Y , LI L , et al . SimAM: A simple, parameter-free attention module for convolutional neural networks [C]// Proceedings of the 38th International Conference on Machine Learning . [S.l.] : PMLR , 2021 : 11863 - 11874 .
TAN M , PANG R , LE Q V . EfficientDet: Scalable and efficient object detection [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . [S.l.] : IEEE , 2020 : 10781 - 10790 .