An Advanced YOLOv8-Nano Model with Attention Neck Network for PCB Bubble Detection

Abstract

The occurrence of bubbles during the coating process of PCBs (Printed Circuit Boards) can significantly degrade product reliability. Therefore, the detection of bubbles during the production process is essential. In this study, we propose a model that combines the YOLOv8 architecture with NAMAttention and Cross Stage Fusion (C2F) structures, leveraging the excellent performance of the YOLOv8 model for image detection. The model effectively integrates the normalization-based attention mechanism of NAMAttention with the multi-scale features of C2F, leveraging their respective advantages. This model demonstrates a 12% improvement in detecting smaller bubbles compared to the original YOLOv8. Our research makes a significant contribution to real-time quality control in PCB manufacturing processes.