The overhead high-voltage transmission lines are the key infrastructures that connect China''s clean energy centers and load centers. Ensuring their safe and stable operation is vital to achieving the emission peak and carbon neutrality goal on schedule. Defect detection of power transmission lines based on drone inspection has strong practical value. Since there are a huge number of defect classes, the long-tailed distribution is caused by the unbalanced category distribution in datasets. This paper introduces a method based on the offline Gaussian model for defect detection in power line images. This method enhances the classifier in Mask R-CNN by using the offline Gaussian model, thereby improving the classification performance of the classifier on the tail data categories. The offline Gaussian model does not require additional training steps and is robust to data distribution. This method is simple, effective and highly scalable. It does not require additional model structures and hyperparameters. It can be directly used on existing object detection and segmentation models, which can effectively alleviate the impact of the long-tailed distributions of data categories on the classifier.