Comparative Study of CNN and ResNet Models in the Classification of Corrosion and Damage in Tunnel Lighting Fixtures

Abstract

This study aims to automate the assessment of corrosion and damage in tunnel lighting fixtures using Artificial Intelligence (AI). Traditional inspection methods have predominantly relied on human visual inspection, leading to inconsistencies due to varying expertise levels among inspectors. In our research, we employed Convolutional Neural Networks (CNN), a subset of deep learning, along with the advanced ResNet model, to assess corrosion and damage in lighting fixtures. Conventional assessment techniques required annotating training data with bounding boxes around objects, which proved to be impractical in terms of workload. Therefore, our study developed a methodology that enables the assessment of corrosion and damage in lighting fixtures without the necessity for object detection preprocessing, successfully allowing AI to recognize the shape of lighting fixtures.