Open Access

Research on Typical Defect Identification Technology of Composite Insulators for Ultra High Voltage Transmission Lines Based on Spectral Feature Extraction

Peiyong Yu

Peiyong Yu

State Grid Heilongjiang Electric Power Company Limited Extrahigh Voltage CompanyHarbin, China
Search for this author on
Sciendo|Google Scholar
Yu, Peiyong
, 
Haikuan Wang

Haikuan Wang

Northeast Branch of State Grid Corporation of ChinaShenyang, China
Search for this author on
Sciendo|Google Scholar
Wang, Haikuan
, 
Jinfu Han

Jinfu Han

State Grid Heilongjiang Electric Power Company Limited Extrahigh Voltage CompanyHarbin, China
Search for this author on
Sciendo|Google Scholar
Han, Jinfu
, 
Minhao Hu

Minhao Hu

State Grid Heilongjiang Electric Power Company Limited Extrahigh Voltage CompanyHarbin, China
Search for this author on
Sciendo|Google Scholar
Hu, Minhao
 and 
Wei Yang

Wei Yang

State Grid Heilongjiang Electric Power Company Limited Extrahigh Voltage CompanyHarbin, China
Search for this author on
Sciendo|Google Scholar
Yang, Wei
  
Sep 22, 2025
Research on Typical Defect Identification Technology of Composite Insulators for Ultra High Voltage Transmission Lines Based on Spectral Feature Extraction's Cover Image
About this article
Previous Article
Next Article
Cite
Download Cover
Published Online: Sep 22, 2025
Received: Dec 22, 2024
Accepted: Apr 17, 2025
DOI: https://doi.org/10.2478/amns-2025-0956
Keywords
© 2025 Peiyong Yu et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Partition hyperplane
Partition hyperplane

Figure 2.

Optimal hyperplane
Optimal hyperplane

Figure 3.

Spectral data acquisition
Spectral data acquisition

Figure 4.

LBP coding
LBP coding

Figure 5.

Improved LBP coding process
Improved LBP coding process

Figure 6.

Feature series fusion process
Feature series fusion process

Figure 7.

Precision-Recall curve
Precision-Recall curve

Figure 8.

Identification results of the defect degree
Identification results of the defect degree

The two models of the same scenario were compared

Model Type of insulator Correct detection Fail to detect False drop Recall rate/% Accuracy/%
Faster R-CNN Glass type 104 19 16 84.55 86.67
Faster R-CNN Compound type 122 26 23 82.43 84.14
LBP-HOG-SVM Glass type 116 7 5 94.31 95.87
LBP-HOG-SVM Compound type 139 9 4 93.92 97.20

Recognition accuracy under different block and unit sizes

Lumpiness Small (b*b) Cell size(c*c) Step Size d Overlap or not Eigenvector dimension Recognition accuracy(%)
16*16 8*8 16 No 1024 82.75
16*16 8*8 8 Yes 512 87.89
32*32 16*16 32 No 256 89.66
32*32 16*16 32 Yes 128 94.62

Comparison of classification based on HOG feature and LBP-HOG feature

Combined form AB AC AD BC BD CD
SHOG 89.75% 94.81% 96.95% 86.72% 96.53% 95.91%
SLBP-HOG 88.84% 94.12% 96.64% 86.84% 95.75% 95.96%
NHOG 236 228 209 238 219 226
NLBP-HOG 188 184 158 186 164 176
T-train HOG(s) 488.36 488.33 446.06 449.26 448.99 457.37
T-train LBP-HOG(s) 478.04 476.46 476.26 468.35 477.26 477.46
T-test HOG(s) 0.188 0.187 0.186 0.194 0.195 0.187
T-test LBP-HOG(s) 0.147 0.147 0.146 0.164 0.148 0.157

Defect identification results

Algorithm Composite insulator Glass type insulator mAP/% Macro F1/%
AP/% F1/% AP/% F1/%
YOLOV2 91.15 92.82 93.62 93.65 92.17 93.52
SSD300 87.73 88.62 88.33 88.45 88.87 88.57
R-FCN 89.44 88.88 91.77 91.18 91.65 90.37
Faster R-CNN 85.35 84.92 85.88 85.75 85.92 85.66
Ours 92.77 92.85 92.93 93.77 92.88 93.38
Language:
English
Publication timeframe:
1 times per year
Journal Subjects:
Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other
Journal RSS Feed