CC-YOLO: An Improved PCB Surface Defect Detection Model for YOLOv7

Abstract

Recently, there has been increasing academic interest in the problem of detecting surface defects on PCBs due to their significance in manufacturing processes. To address the issue of low detection accuracy in existing PCB surface defect detection models, this paper introduces a novel model called CC-YOLO, which enhances YOLOv7. The proposed model incorporates the CARAFE (Content-Aware ReAssembly of FEatures) operator as an up-sampling technique to minimize feature information loss in the input image and enhance the model’s feature extraction capability. Additionally, we propose the C2f-C module, which extends CSPDarknet53 to 2-Stage FPN-Cascade and strengthens it further by incorporating multiple Bottleneck modules to capture more comprehensive gradient flow information. We evaluate the performance of the enhanced YOLOv7-based model, CC-YOLO, on the publicly available PCB surface defect dataset from the Intelligent Robotics Laboratory at Peking University. The experimental results demonstrate that CC-YOLO achieves an accuracy of ${9 8 . 9 \%}, 2.3 \%$ higher than the original YOLOv7 model.