In the operation of 10 kV distribution lines

insulator surface discharge is one of the common defect types

and effective monitoring of such defects is of great significance. Surface discharge is accompanied by high-frequency current pulses

and traditional detection methods based on feature extraction using digital signal processing and preset rules are difficult to adapt to complex and variable operating conditions. Therefore

this study proposes a method for identifying insulator surface discharge defects by applying the YOLOv8 object detection model to perform inference on two-dimensional images of discharge pulse waveforms. A dedicated high-frequency current sensor was designed to collect insulator discharge waveforms under different pollution levels

based on which training and validation datasets were constructed. To improve the generalization capability of the model

interference components were artificially added to the waveforms. PyTorch was used as the deep learning framework

and the trained model was deployed on an edge computing device for testing. The experimental results show that the YOLOv8 model can be used to identify pollution-induced surface discharge defects and their severity with relatively high accuracy. After training and testing

the model achieves a mean average precision of 98.2% at an intersection over union (IoU) threshold of 0.5

with a recall of 83%. The model was further deployed on a Raspberry Pi 4B edge computing device for validation. The image inference process required less than 400 MB of memory and approximately 0.5 s

demonstrating the feasibility of its future application to on-site online monitoring. This study provides a technical solution for the online monitoring of insulator surface discharge defects. The model accuracy can be further improved through continuous iterative training

indicating promising application prospects.