Binary Classification of Marine Corrosion Using Deep Learning: A Benchmark Study on the Full Marine_corrosion_dataset

Abstract Corrosion poses a persistent challenge to marine infrastructure, leading to structural degradation, costly maintenance, and safety risks. While deep learning techniques have gained traction in automating corrosion detection, comprehensive benchmark studies on large, high-resolution datasets remain limited. This study presents a binary classification framework using deep convolutional neural networks (CNNs) to distinguish between corroded and healthy marine surfaces based on the publicly available marine_corrosion_dataset, comprising 9,000 unaltered images (512×512 pixels). Eight distinct corrosion types were aggregated into a single “All_forms_of_corrosion” class and evaluated against “Healthy_structures.” Under standardized conditions, the KAI platform implemented and trained four state-of-the-art CNN models—VGG16, VGG19, InceptionV3, and ResNet50. Among them, ResNet50 achieved perfect classification metrics (accuracy, precision, recall, and F1-score = 100%), followed closely by InceptionV3 (F1-score = 99%) and the VGG models (F1-scores = 96–97%). These findings affirm the robustness of deep learning for binary marine corrosion detection and offer a reliable reference for developing intelligent inspection tools in offshore and shipboard applications.