Performance Analysis of the Fine-Tuned YOLOv8n Model for Detection and Classification of Printed Circuit Boards Defects

This paper presents a detailed analysis of results obtained after transfer learning and fine-tuning the YOLOv8n with various parameters, to detect and classify printed circuit boards defects. The model identifies six types of defects: mouse bite, spur, missing hole, short circuit, open circuit, and spurious copper. A publicly available dataset containing labeled images of these defects was used for training. By varying the initial learning rate, batch size and optimization algorithm, the best results, in terms of Accuracy, Precision, Recall and F1-score, were obtained for 0.01 initial learning rate, 32 batch size, with SGD optimizer.