In response to the need for manual recognition of infrared images of transmission line insulators and the difficulty in constructing a complete anomaly sample database, an intelligent detection algorithm is proposed based on the combination of domain transform infrared image processing module and unsupervised feature reconstruction anomaly detection module. The domain transform module is used to enhance the matching degree between input data and the intelligent model, while the unsupervised anomaly detection module utilizes feature reconstruction, knowledge distillation, and a memory matrix to discriminate anomalies by comparing input and reconstructed images. The domain transform module improves the detection accuracy of the unsupervised anomaly detection model by 2%~4% on a self-made insulator infrared image dataset, demonstrating success in various background comparison experiments. Furthermore, the AUCROC index of the unsupervised feature reconstruction anomaly detection module (FRAD) on MVTec AD dataset and self-made dataset reaches 88.78% and 88.56% respectively, which is in the leading position compared to models in the same domain.