Insulator age estimation using limited unlabeled UAV data samples

Electrical insulators in power systems undergo aging, which can compromise their performance and increase the likelihood of faults. Timely assessment of insulator aging is critical for implementing preventive maintenance strategies; however, traditional inspection techniques often rely on expert judgment and may involve costly laboratory tests or require labeled datasets. This paper presents an unsupervised method for estimating the aging condition of electrical insulators using standard RGB images captured by unmanned aerial vehicles (UAVs). The proposed approach combines object detection with self-supervised contrastive learning and incorporates illumination-invariant features such as discrete wavelet transforms (DWT), local binary patterns (LBP), and Lab a/b color channels. The method is evaluated on a dataset of 100 UAV-acquired insulator images. Among five tested scenarios, the best-performing configuration (LBP + DWT) achieves a Silhouette score of 0.49, a Davies–Bouldin Index of 0.92, and a Calinski–Harabasz score of 273.8, demonstrating strong clustering quality. A 2D t-SNE visualization further reveals a smooth progression of cluster groupings that reflect visually coherent aging stages, highlighting the effectiveness of the proposed framework in capturing unsupervised degradation patterns.