Image Visibility Patch Aided Hydrophobic Class Detection of Silicone Rubber Insulators Employing Bi-LSTM Network

Precise detection of hydrophobicity grades (HGs) of silicone rubber insulator is necessary to prevent premature tripping of transmission lines. Keeping this in mind, in this paper, a novel HG detection framework employing image visibility graph is proposed for remote health diagnosis of silicone rubber insulators. To this end, an experiment has been designed to acquire surface images from 11kV silicone rubber suspension insulators representing different HGs. Following this, the captured images were suitably pre-processed and were further analyzed using image visibility graph (IVG). The IVG converts the captured images to a topological network preserving their pixel information. In this study, image visibility patches (s) were extracted from different IVG transformed HG images to quantify intricate pixel-level alterations taking place between different HGs on a local scale. The extracted VPS were utilized as inputs to a bi-directional long short-term memory (bi-LSTM) network for classification of HGs. Comparing against existing methods revealed that the proposed HG detection methodology is superior indicating its potential application for remote health diagnosis of outdoor insulators.