Stress Concentration Zone Detection of Aircraft Structures Based on LW-YOLOv8

Accurately pinpointing stress concentration zones during aircraft structure fatigue tests contributes significantly to enhancing the efficiency of fatigue crack detection, which represents a pivotal step towards realizing automated vision-based damage monitoring. To achieve high-precision detection of stress concentration zones in aircraft structures on resource-constrained embedded devices, a stress concentration zone detection method based on LW-YOLOv8 is proposed. Firstly, the feature fusion stage of YOLOv8 is optimized based on the C2f_RepViTBlocks, aiming to enhance the detection precision of stress concentration zones while reducing the model parameters and computational complexity. Secondly, EIoU loss is adopted to replace the original loss function, further improving the localization accuracy for stress concentration zones. Ablation studies demonstrate that the proposed model, under the combined effect of two modules, achieves a 1.9% improvement in $\mathrm{mAP}_{0.5: 0.95}$ compared to the baseline, while reducing the parameters and Giga Floating-Point Operations Per Second(GFLOPs) by 1,195,712 and 2.2 respectively. Experimental results illustrate that the presented LW-YOLOv8 possesses robust stress concentration zone detection capabilities under conditions of blurred imaging and complex background interference. Furthermore, its lightweight structural design facilitates easier deployment on embedded devices in subsequent stages.