A Two-Stream Deep Learning Framework for Robust Coral Reef Health Classification: Insights and Interpretability

Coral reefs, vital ecosystems for marine life and coastal communities, are facing grave threats from climate change and human activities, including pollution. To effectively conserve these ecosystems, monitoring and categorizing coral health into three states: Need to see healthy, Dead Coral and Bleached Coral. In this study, we introduce a two-stream deep learning framework designed to robustly classify coral health, emphasizing interpretability as integral to its applicability. The framework uniquely incorporates two parallel processing streams: one dedicated to analyzing the structural aspects of coral imagery and another focused on capturing color variations, both critical for accurate health assessment. We created a specialized dataset of 20,000 images by combining five public datasets, applying a class fusion process to ensure consistent class definitions across diverse data sources. Our methodology addresses challenges such as underwater variation and class imbalance through tailored data augmentation, multi-scale feature analysis, temporal modeling, and weighted loss functions. These innovations allow the model not only to achieve high accuracy but also to provide insights into the decision-making process, enhancing the model’s utility for ecological researchers and conservation efforts. The experiment successfully identifies the corals’ health states with an accuracy of 98.56% and mAP of 0.98, demonstrating the framework’s reliability for coral health state identification. To make ecological researchers and conservationists understand the model’s decision-making process, we use explainable AI (XAI) tools such as the Grad-CAM and LIME. More evaluation metrics, including precision, recall, F1-score, confusion matrices, and precision-recall curves, are also conducted to validate the performance of the model. An innovative approach combining technology and conservation to monitor coral reefs using AI-driven, this scalable and accurate AI-driven approach provides a potential solution for monitoring coral reefs. The results highlight the promising role of advanced AI in global conservation by lessening the ecological decline of coral reef systems.