Detection of SMD components on PCBs using neural networks: A comparative study of Roboflow 3.0 and YOLO v11

Abstract

The inspection of surface-mount device (SMD) components on printed circuit boards (PCBs) is crucial to ensuring quality in electronic manufacturing. Conventional methods often lack precision and speed, resulting in defects and higher costs. This study compared two advanced neural networks, Roboflow 3.0 and YOLO v11, to address these challenges. Using a dataset of 1,300 images that included capacitors, resistors, and transistors under varying conditions, the models were trained and evaluated based on metrics such as mAP50, mAP50:95, precision, and recall. The results showed that Roboflow 3.0 achieved superior performance with a mAP50 of 95.6% and a mAP50:95 of 64.9%, along with consistent improvements in precision and recall during incremental training. In contrast, YOLO v11 demonstrated stability but achieved lower metrics, with a mAP50 of 90.2% and a mAP50:95 of 61.3%. While both models offered robust detection capabilities, Roboflow 3.0 excelled in adapting to diverse variations in lighting and geometry. This study highlights the potential of convolutional neural networks to transform PCB inspection in quality environments, offering greater precision and efficiency, thereby reducing human errors and associated costs while optimizing production processes.