Uncertainty-Aware Measurement of Corona Discharge Area on Insulators from Solar-Blind Ultraviolet Imagery

The area of the corona discharge region on an insulator is reported as a diagnostic indicator, yet it is almost always obtained by thresholding a segmentation probability map at the conventional 0.5 level and is reported without any uncertainty statement. Using 1600 solar-blind ultraviolet frames, we show that this practice is the dominant error source. Selecting the operating point on a validation split reduces the median relative area error from 20.8% to 3.3% and, unexpectedly, increases Dice from 0.874 to 0.903; a 7.76 M-parameter U-Net then reaches 3.26 ± 0.39% area error, within 0.07 percentage points of an oracle threshold. Dice is not a valid proxy for area: a two-pixel registration shift changes the area by 0.03% but Dice by 0.159, because the annotations are fragmented into about 31 sub-ten-pixel blobs. Split conformal prediction yields intervals whose empirical coverage (93.7–97.8%) meets or exceeds the nominal level even when checkpoint selection, threshold selection and calibration use disjoint data, with a domain of validity of roughly −5% to +10% gain deviation. A GUM budget attributes 15.7% relative standard uncertainty to annotation-boundary ambiguity, against 2.1% for the model and acquisition combined, so the measurement is annotation-limited. The protocol makes the reported image-domain area fraction an interpretable measurand with a stated uncertainty.

Authors

Institutions

Publication Details

Journal
Energies
Published
2026-11-13
DOI
https://doi.org/10.3390/en19225481
Primary Topic
High voltage insulation and dielectric phenomena
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Uncertainty-Aware Measurement of Corona Discharge Area on Insulators from Solar-Blind Ultraviolet Imagery

Yunmin Xie, 郑泉云, Bing Zeng, Xiaofan Xie et al.
1 links
Energies
High voltage insulation and dielectric phenomena
article

Uncertainty-Aware Measurement of Corona Discharge Area on Insulators from Solar-Blind Ultraviolet Imagery

Yunmin Xie, 郑泉云, Bing Zeng, Xiaofan Xie, Hui Liao, Zhen Yao, Duo Xu, Tianjiao Cao, Mengyao Liang
article en

Abstract

The area of the corona discharge region on an insulator is reported as a diagnostic indicator, yet it is almost always obtained by thresholding a segmentation probability map at the conventional 0.5 level and is reported without any uncertainty statement. Using 1600 solar-blind ultraviolet frames, we show that this practice is the dominant error source. Selecting the operating point on a validation split reduces the median relative area error from 20.8% to 3.3% and, unexpectedly, increases Dice from 0.874 to 0.903; a 7.76 M-parameter U-Net then reaches 3.26 ± 0.39% area error, within 0.07 percentage points of an oracle threshold. Dice is not a valid proxy for area: a two-pixel registration shift changes the area by 0.03% but Dice by 0.159, because the annotations are fragmented into about 31 sub-ten-pixel blobs. Split conformal prediction yields intervals whose empirical coverage (93.7–97.8%) meets or exceeds the nominal level even when checkpoint selection, threshold selection and calibration use disjoint data, with a domain of validity of roughly −5% to +10% gain deviation. A GUM budget attributes 15.7% relative standard uncertainty to annotation-boundary ambiguity, against 2.1% for the model and acquisition combined, so the measurement is annotation-limited. The protocol makes the reported image-domain area fraction an interpretable measurand with a stated uncertainty.

EnergiesVol. 19(22)
Jiangxi University of Water Resources and Electric Power (CN), State Grid Corporation of China (China) (CN)
National Natural Science Foundation of China, Natural Science Foundation of Jiangxi Province
High voltage insulation and dielectric phenomena
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon