Abstract: Aiming at the problem that it is difficult to detect the defects and tilt angle of the grading ring of transmission lines, an improved YOLOv7 grading ring defect and rotated object detection model is proposed. The backbone network of YOLOv7 is designed as a two-branch network, which combines the local features of CNN and the global representation of Transformer to enhance the expression ability of object features. A progressive feature fusion structure is used to make the model more effective in fusing shallow and deep features. A Gaussian modeling representation of rotated box detection is introduced in the detection head to improve the accuracy of angle detection. Finally, based on the improved rotated object detection model, a grading ring tilt angle detection algorithm is designed to realize the detection of the grading ring tilt angle under any viewing angle. The experimental results show that the average angular error of the improved algorithm is reduced by 0.97° to only 5.32°, and the mAP increased by 4.1% to 91.5%. The algorithm significantly improves the mean average precision in the defect detection of grading ring, and provides an effective solution for the quantitative detection of grading ring defects and tilt angle.