In power systems, insulator fault detection is crucial for ensuring the safe operation of the grid. This paper proposes an improved YOLOv5 model for efficient insulator fault detection. The improvements primarily involve two aspects: replacing the backbone network of YOLOv5 with EfficientViT (Efficient Vision Transformer) and adding a SimAM (Simulated Attention Module) attention module to the neck of the network. EfficientViT, as an efficient vision transformer, significantly reduces computational complexity and parameter count while maintaining model performance. The SimAM attention module enhances feature extraction accuracy and model robustness by simulating the attention mechanism. In the experiments, we evaluated the improved model using a public dataset containing various types of insulator faults. The results show that the improved model achieves a 2.1% increase in mean Average Precision (mAP) and a 1.4% increase in recall rate under the same hardware conditions. Additionally, the model's parameter count is reduced by 4%, floating-point operations are reduced by 9%, and the final model size is decreased by 4.5%.