Transformers are crucial components in power transmission systems, directly influencing the reliability of the power transmission and distribution network. Addressing the issue of inadequate precision in transformer fault diagnosis during power maintenance, this paper presents a method for transformer fault diagnosis leveraging a convolutional neural network (CNN) augmented with channel attention mechanisms and data augmentation techniques. Firstly, we integrate a 1D-CNN neural network with channel attention mechanisms to enhance feature extraction and generalization capabilities. Secondly, to tackle the challenge of limited fault samples, we employ a sample augmentation approach based on generative adversarial networks to generate a large volume of simulated samples. Lastly, through comparative simulation experiments involving the construction of 1D-CNN and LSTM neural models, our results demonstrate that the SA-CNN model achieves an average improvement of 15.2% and 10.8% in the histograms of AUC-ROC and AUC-PR respectively, effectively enhancing diagnostic accuracy.