CoralXAI-Net: An Explainable Deep Learning-based Multi-Class Coral Species Recognition for Underwater Ecosystem Monitoring

Abstract

Coral reefs are among the most diverse ecosystems on Earth, supporting nearly 25% of all marine species while providing ecological, economic, and coastal protection benefits. However, global climate change, ocean acidification, and human-induced disturbances have accelerated coral degradation, making timely and accurate coral species identification a pressing requirement for reef conservation and biodiversity management. The problem definition centers on the inefficiency of existing coral monitoring practices, which rely largely on manual field surveys and subjective expert analysis of reef images. Such methods are not only labor-intensive and costly but also inconsistent due to human error and the challenging nature of underwater imaging, where issues like low illumination, turbidity, and occlusions complicate the process. The traditional systems for coral identification have relied on manual annotation and classical computer vision techniques using handcrafted features such as shape, texture, and color descriptors. This research presents an intelligent system for underwater coral reef species identification using deep feature extraction and machine learning techniques. Coral reef images are processed using a Vision Transformer (ViT) model to extract robust feature representations, which are then used to train multiple classifiers, including Natural Gradient Boosting (NGB), Histogram Gradient Boosting (HGB), Extreme Gradient Boosting (XGB), and the proposed scope rules classifier (SRC). Performance evaluation using accuracy, precision, recall, F-score, confusion matrices, and ROC curves shows that the proposed SRC model significantly outperforms baseline models. Additionally, an Explainable Artificial Intelligence (XAI) module and Telegram bot interface provide interpretable coral analysis and real-time prediction capabilities for practical marine monitoring applications.