Damage location and area measurement of aviation functional surface via neural radiance field and improved Yolov8 network

Abstract

To realize high-precision intelligent detection, location and area measurement of aviation functional surface damage, a damage location and area measurement method combining neural radiance field and improved Yolov8 network is proposed in this paper. The high-fidelity NeRF (Neural Radiance Field) and 3DGS (3D Gaussian Splatting) models are trained by acquired multi-view optical images of damaged functional surfaces. The rendered new-view images are used as a new data augmentation method to enhance the training effect of Yolov8 network. The network architecture of Yolov8 model is improved. The backbone is replaced with the latest StarNet feature extraction network, and the context feature fusion module (CFFM) proposed in this paper is used for feature fusion enhancement. The improved context-guide multi-head self-attention (CG-MHSA) is added to the detection Head. The comparison and ablation experiment results show that the improved module proposed in this paper has a good effect on the improvement of Yolov8 model, and improves the damage detection ability and location ability of the model. The application experiment results verify the effectiveness of the proposed method for calculating the damage area of a plane/camber surface, and the accuracy of the damage area measurement is high.