Corrosion Image Classification Using Deep Learning

The slow deterioration of metal brought on by its chemical interactions with the environment is known as corrosion. Material corrosion affects significant economic sectors and industries, including infrastructure, transportation, defense, environment, and health, leading to a massive loss in gross domestic product (GDP). However, it can be challenging to identify corrosion early on. Conventional techniques for identifying and categorizing corrosion often depend on manual visual inspection, which can be expensive, time-consuming, and prone to human error, thus posing a challenge for early detection of corrosion. In recent years, there has been a shift from traditional techniques to automating corrosion detection using computer vision. This shift has been primarily driven by advancements in deep learning, particularly convolutional neural networks (CNNs), which have demonstrated remarkable potential in identifying and classifying corrosion types from images even in the early stages. These automation techniques utilizing deep learning paved the way for reducing the need for manual inspection, thus helping improve corrosion detection accuracy and reliability. In this study, different deep learning models and techniques are utilized to classify corrosion images based on their rate or degree of corrosion. Different experiments using pretrained VGG16 and ResNet18 are showcased which showed promising results. The study employed transfer learning and feature extraction to reduce computational complexity. Further approaches like dimensionality reduction and data augmentation techniques resulted in matching state-of-art classification accuracies.