International Research Journal of Education and Technology

This project presents a PCB defect detection system based on the YOLOv8 deep learning model, enhanced with advanced feature fusion techniques. The primary objective is to improve the balance between detection speed and accuracy, a challenge faced by traditional defect detection algorithms. To achieve this, the system employs the GhostConv module for efficient feature extraction, reducing computational complexity while preserving accuracy. The system also integrates a multi-scale semantic pyramid fusion structure (SPPFCS), which enhances the fusion of deep, multi-dimensional semantic information, improving the model's capacity to detect various PCB defects, including small-scale anomalies.

Additionally, the introduction of the A2 attention mechanism focuses on improving the detection of smaller targets by enhancing the network’s ability to process complex and high-dimensional semantic information. The system utilizes the Wise-IoU loss function during training, which strengthens the model’s ability to fit and generalize across different defect types. Experimental evaluations on open- source PCB defect datasets demonstrate that the proposed model achieves a significant improvement in both detection speed and accuracy compared to previous approaches, making it ideal for real-time industrial PCB inspection applications. With these optimizations, our system operates at up to 125 FPS while maintaining high accuracy, achieving a mAP such as open circuits, missing components, and short circuits. This makes it a robust solution for enhancing the quality control process in PCB manufacturing environments, where real-time and precise defect detection is crucial.

Key Words: .PCB -defect detection, Yolo , deep learning, Semantic feature fusion, Ghost convolution, SPPFCS structure, A2 attention mechanism.