The polymeric materials used in insulator sheds are prone to ageing under different environmental conditions

which can lead to flashovers and grid outages. The ageing process in composite insulators primarily involves changes in the microscopic properties of the materials. Ageing levels were determined based on the hydrophobicity of artificially accelerated ageing samples

and spectral lines ranging from 400 to 1000 nm were analysed at various ageing stages. This study employed sample entropy as the objective function and used Beluga whale optimization to determine the optimal number of decomposition layers

K

and the penalty parameter

α

for the variational mode decomposition

thereby achieving effective denoising in the spectral domain. The channel attention mechanism was able to better extract ageing status information from the spectral bands. By employing a long short-term memory network with an adaptively weighted squeeze-and-excitation network

a high global accuracy of 95.48% was attained

outperforming traditional classification models. This approach successfully addressed the challenge of low evaluation accuracy resulting from the high spectral line similarity observed in severely aged levels. The trained model enabled the visualisation of ageing distribution at the pixel level within the sheds' structure

overcoming difficulties associated with detecting microscopic ageing states and providing comprehensive assessments of composite insulators. This research study significantly contributes to ensuring the safe operation of power grids and offers valuable insights for guiding the material selection and design of composite insulators in diverse environments

thereby presenting promising applications in various fields of electrical equipment condition monitoring.