Improved YOLOv8n-based aircraft skin defect detection method in complex environments

Abstract

Aircraft skin defect detection often suffers from low detection accuracy due to variations in lighting, shadows, and complex backgrounds. To address this, this study proposes a lightweight and enhanced YOLOv8n-based algorithm. Firstly, the original C2f structure is replaced by the new C2fGhost module to reduce the floating-point operation volume during the feature channel fusion process. Secondly, the AOM (Attention Occlusion Mechanism) is introduced to enhance feature extraction in complex environments. Finally, a regression loss function combining DFL and WIS-IoU is adopted to improve convergence. The experimental results demonstrate that the proposed algorithm achieves a Precision of 85.4% and an mailto:mAP@0.5 of 83.6%, representing improvements of 2.6% and 1.9% respectively over the baseline YOLOv8n model. Notably, this performance boost is achieved while maintaining a compact model size of merely 6.02 MB. These results signify a substantial advancement in balancing detection accuracy and model efficiency, offering a highly viable solution for real-time, embedded inspection systems in the aerospace industry.