YOLOv8-AGFan Efficient Lightweight Model for Weld Defect Detection based on YOLOv8n

Weld surface defect detection is critical for industrial automation, ensuring product quality and safety. To address the computational constraints of embedded platforms, this study proposes YOLOv8-AGF, a lightweight algorithm based on YOLOv8n. YOLOv8-AGF introduces the Adaptive Ghost Filter module to reduce parameters and computational complexity, and an AIoU loss function to enhance small defect detection by dynamically adjusting loss components during training. Validated on public and self-built datasets, YOLOv8-AGF achieves high accuracy with only 1.28M parameters and 3 GFLOPs, reducing parameters by 57.33% and GFLOPs by 63% compared to the original model. This algorithm is highly suitable for deployment on industrial mobile and embedded systems, offering an efficient solution for weld defect detection. The self-built dataset is available at https://github.com/WangAiFei111/Weld-Defect-Dataset.git.