Abstract: In the inspection of power transmission system, due to the influence of various complex weather, the imaging of insulator can easily be obscured and blurred. Aiming at this problem, in order to better reflect the real patrol scenes, firstly, this study constructed a dataset of insulator surface defects under various complex weather conditions, including normal weather, sunlight exposure, rainy weather, foggy weather, and snowy weather by using the Albumentations image augmentation framework. Secondly, based on the YOLOv5s model, two deep dense feature-weighted fusion neck networks are proposed to replace the original neck networks so as to strengthen the ability to learn weak and vague features in various complex weather. Finally, the Focal-Efficient IOU Loss is introduced as the loss function in the head network of the model to deal with the problem of positive and negative sample unbalances of boundary box regression due to complex weather interference. The experiments show that when the intersection over union threshold is 0.5, the mean average accuracy of the proposed model in complex weather insulator data sets reaches 99.3%, and the detection precision and recall rate reach 100% and 98.1%, respectively. Compared with other detection models, this model shows good performance in various indicators, and can better meet the high-precision detection tasks of blurred self-explosion defects on insulator surfaces under challenging weather conditions.