Multiple Regression Methods Based Core Rod Localization for Composite Insulators Images

This study aims to detect the temperature rise of the core rod of composite insulators through infrared image thermography to assess their operating condition in the power system. Initially, the infrared images of composite insulators were preprocessed using image erosion to reduce the error of the skirt mask and highlight the core rod. Subsequently, linear regression methods were applied to the eroded images to separate the core rod and obtain its temperature distribution. The study compared four regression methods: ordinary least squares, ridge regression, Lasso regression, and Elastic Net regression, and evaluated the significance of the models using an F-test.