Research on Small Sample Power Model for Transmission Inspection Based on Deep Learning

When using small sample transfer learning to train models in the field of power inspection, the total amount of feature information in the small sample is insufficient, and the application scenarios adapted by the model before migration are often significantly different from the power business scenarios, resulting in underfitting and underfitting of the model trained by small sample transfer learning. This paper proposes a small sample power model for power transmission inspection based on improved deep transfer learning. It aims to address the challenges brought by small samples through transfer learning and generative adversarial networks (GAN) and other technologies. First, the generative adversarial network (GAN) is used to generate synthetic samples similar to the real data distribution to expand the data set size and improve the generalization ability of the model. Secondly, the pre-trained deep convolutional neural network (CNN) is used, combined with the attention mechanism, to automatically focus on the key areas in the power equipment to improve the accuracy and efficiency of defect detection. The pre-trained model is fine-tuned through transfer learning to adapt it to the specific field characteristics of power inspection, so as to achieve better detection performance with limited data. Experimental results show that the proposed method can effectively improve the detection accuracy of the small sample power inspection model and has high practical application value.