A Diffusion Model and Knowledge Distillation Framework for Robust Coral Detection in Complex Underwater Environments

Abstract

Coral reefs play a crucial role in marine ecosystems, but their sustainability is increasingly threatened by climate change and human activities. To aid in the protection and monitoring of these ecosystems, developing advanced artificial intelligence (AI)-based automated detection technologies is essential. This paper introduces the MambaCoral-Diffusion Detection framework (MambaCoral-Diffusion Detection, MCDD), an AI-driven approach for robust coral detection, designed to enhance performance in complex underwater environments—a critical challenge in marine engineering. Key AI contributions include integrating a diffusion model to generate diverse training data to address data imbalance and employing an adaptive global-local attention mechanism within the Mamba architecture’s Spatial Sensing Dual Detection (SS2D) module to enhance feature extraction accuracy. Furthermore, a Channel-Aware Intersection over Union (IoU) knowledge distillation technique is presented, enabling efficient knowledge transfer from complex teacher models to lightweight student models, optimizing the framework for practical deployment with reduced computational demands. From an engineering perspective, MCDD significantly advances automated coral detection in challenging underwater conditions, providing a reliable solution for monitoring marine ecosystems. Experimental results show that MCDD achieves high performance on the Soft Coral dataset, reaching 31.5 frames per second (FPS), the mean average precision at 50\% IoU threshold of 0.843, and the mean average precision averaged across IoU thresholds from 50% to 95% of 0.566, with only 6.5 million parameters and 13.6 billion floating point operations per second (GFLOPs). The code and dataset are available at https://github.com/RDXiaoLu/MambaCoral-DiffDet.git