July 30, 2025

Benchmarking Ai-based Corrosion Classification: A Case Study on Marine Corrosion Image Dataset

Key Points

ResNet-50 achieved a remarkable accuracy of 99% in classifying marine corrosion images.
The study utilized a dataset of 5,105 images sourced from three public repositories for training.
Four CNN architectures were evaluated, with ResNet-50 outperforming others in precision and F1-score.
Results indicate the potential for enhanced real-time monitoring systems in marine engineering applications.

Abstract

Abstract Marine corrosion detection is crucial and must be time-saving and accurate to avoid offshore and coastal infrastructure destruction. This research article presents a deep learning-based approach for classifying corrosion in marine environments using an integrated dataset compiled from three publicly available sources: GitHub repositories by Jackson and Cheng, and a Kaggle dataset by Ebrahim007. The final dataset comprises 5,105 images spanning corrosion and healthy structural conditions. Four convolutional neural network (CNN) architectures—VGG16, VGG19, ResNet-50, and InceptionV3—were trained and evaluated using the KAI software platform. The experimental setup included data augmentation, a consistent training protocol, and stratified dataset splits. ResNet-50 achieved the highest overall performance among the tested models, with 99% accuracy, 98% precision, and an F1-score of 98% on the test set. The experiment results demonstrate the feasibility and quality of using CNNs for binary corrosion classification and highlight the value of combining multi-source datasets to improve model generalization. This work establishes a strong foundation for the future development of real-time corrosion monitoring systems in marine engineering applications.

KI fragen

Bookmark