Performance Evaluation of Edge-Based Segmentation Methods for Electrical Tree Image Analysis in High-Voltage Experiments

Abstract

This research evaluates the performance of edge-based segmentation methods in analysing two-dimensional (2D) electrical tree images obtained during high-voltage (HV) electrical tree experiments. Non-uniform illumination in 2D optical images poses challenges in accurately extracting and measuring the original treeing image. Edge segmentation emerges as a promising solution to precisely distinguish tree structures from the insulation background within the image. Cross-linked polyethylene (XLPE) samples were subjected to HV stress for real-time propagation observation, followed by extraction and segmentation of treeing images using edge-based operators. The findings emphasize the superiority of the Roberts edge operator in accurately detecting electrical trees, showcasing the highest average accuracy of 97.01% and 99.58% specificity, while also demonstrating relatively high sensitivity. Moreover, the Roberts method provide much precisely measures the propagation length and width than conventional measurement method, closely approximating the actual tree measurements. This research emphasizes the significance of accurate segmentation for investigating electrical tree propagation in XLPE materials and provides recommendations for future research, especially in HV XLPE cable manufacturing.