Abstract: Infrared thermography-based fault diagnosis of printed circuit boards (PCBs) is gaining attention in field applications such as emergency repairs and equipment inspections due to its rapid detection, simple operation, and equipment versatility. However, existing methods often require strict image acquisition conditions and are not suitable for portable devices. To overcome these limitations, a novel fault diagnosis method for PCBs using portable devices is proposed, based on a convolutional neural network (CNN) for infrared thermogram classification. A pre-trained Residual Neural Network (ResNet) was employed as the backbone to automatically extract diagnostic features from the images. In addition, a Spatial Transformer Network (STN) was integrated at the input stage to enhance the network’s robustness to viewpoint variations, thereby reducing the dependence on strict imaging conditions. Five-fold cross-validation was performed on a self-built dataset containing multi-view images. The results demonstrated that the proposed method achieved an average accuracy of 93%. This study provides a new approach for the development of portable infrared fault diagnosis devices for PCBs.