In view of the likely missed and false detection in the process of insulator detection due to the complex background and large differences in the scale of the detection targets in aerial insulator images, a CHD-YOLO algorithm model has thus been proposed for a more accurate detection of defective insulators. Firstly, in the neck network, a lightweight cross scale feature fusion module (CCFM) is adopted for a better utilization of the feature information, thus reducing the computational complexity and overhead of the network. Next, the HAttention (Hybrid Attention Transformer) attention mechanism is integrated into the YOLOv8n algorithm so as to obtain more detailed features and improve the ability of the model to extract and fuse target features. Finally, the Dynamic Head module is introduced to enhance the performance of the detection head, as well as improve the recognition ability of the model for small defect areas. Experimental results show that the mAP value of the improved YOLOv8 model reached as high as 93.0%, an increase of 2.3% compared to the original YOLOv8. The proposed method achieves a balance between accuracy and speed, meeting the accurate detection requirements of insulator defects in power systems.