Abstract: Bushing faults have become the main cause of transformer failures in recent years, which can be detected online by using thermal images. However, the abnormal states data of thermal images of transformer bushing are rare in practice, which brings difficulties to the model training of online intelligent anomaly identification for transformer bushings. To solve this problem, this paper proposes an online identification method for transformer bushing anomaly based on data domain transformation of thermal images and Inception-CNN network. Firstly, abnormal thermal image samples are obtained by simulating the temperature distribution of bushings under a variety of typical abnormal conditions. Then, we complete the mapping between the simulated image domain and the thermal image domain based on CycleGAN to realize the enhancement of the abnormal data. Finally, an Inception-CNN-based bushing abnormality classification network is established to identify the bushing abnormal states. The experimental results show that the F₁ value of the proposed method exceeds 97% for each phase of bushing abnormality classification, outperforming the results in ablation experiment without data domain transformation.It is revealed that the proposed method can be adopted to effectively identify the normal and abnormal operating states of transformer bushings and provide a new idea for the enhancement of abnormal data.