The Application of Machine Learning to Paint Condition Assessment Using Hyperspectral Data

Paint provides a shield that protects our structures, particularly steel structures, from corrosion and damage due to harsh environmental conditions such as sunlight and humidity. It is essential to ensure that paint coating works effectively and remains in good structural conditions (without peeling or fracture). Assessing the paint condition is complex, especially on large structures such as bridges, and can be time-consuming and costly. The traditional visual inspection methods depend on the experience and evaluation of inspectors in the field and require hard labor and time to perform. This study aims to employ hyperspectral imaging and image classification approaches to develop a structural condition monitoring system that rapidly assesses paint conditions and degradation. Several classification algorithms, such as Decision Tree, Support Vector Machines, Logistic Regression, and Naïve Bayes, were trained and tested. The results obtained showed that the Decision Tree classifier outperformed the rest of the classifiers, achieving a highly accurate assessment of the paint condition and degradation levels with a detection accuracy of 0.98.