AutoCoralMatch: Automated Coral Health Assessment                             
 
   
     

AutoCoralMatch: An Open-Source Foundation Model-Driven Framework for Patch-Level Coral Health Assessment and Automated Bleaching Detection

     

Lyes Saad Saoud and Irfan Hussain

     

Khalifa University Center for Autonomous Robotic Systems
Khalifa University, Abu Dhabi, United Arab Emirates

Preprint 2025

     
                  Paper                           Code                           Datasets              
   
 
 
   
      AutoCoralMatch system illustration      

Figure 1: Overview of the AutoCoralMatch system. A modular, offline desktop application for coral health assessment using CoralWatch cards. Left:} The user interface includes upload, dehazing, card detection, segmentation, rectification, coral-only extraction, and report generation (CSV/PDF). A built-in annotation tool enables patch-level labeling. Middle: Intermediate outputs from each pipeline stage, including isolated CoralWatch cards, the margin region, and coral-only imagery. Right: Key computational steps: (1) RAUNE-Net dehazing, (2) Grounding DINO detection, (3) SAM2 segmentation, (4) perspective rectification, (5) \textbf{YOLOv8} patch detection, (6) \textbf{HSV extraction}, (7) \textbf{HSV-based} health estimation, and (8) structured output. All components are GPU-accelerated, locally executable, and modularly designed.

   
 

System Demonstrations

Figure 2: Demonstration of the AutoCoralMatch Desktop Application (Main Processing Pipeline).

Figure 3: Demonstration of the CoralWatch Chart Annotation Tool (Data Preparation & Training).

 
   

Abstract

   
     

Coral reefs are highly sensitive indicators of climate change, making accurate and scalable monitoring tools critical for global conservation. Traditional CoralWatch-based assessments rely on manual diver observations and post-hoc color matching, which are labor-intensive, subjective, and difficult to scale.

     

We present AutoCoralMatch, an open-source, modular software framework for automated coral health assessment from underwater imagery. The system integrates underwater image dehazing (RAUNE-Net), CoralWatch card detection using foundation vision models (Grounding DINO and SAM2), perspective rectification, patch-level color extraction, and health classification based on \textbf{perceptual color metrics in HSV space}. The software includes a user-friendly desktop interface supporting batch processing, real-time feedback, report generation, and structured metadata logging.

     

To validate system performance, we conducted controlled marine trials with over 20 coral morphologies under varying lighting and turbidity. Results demonstrate robust card detection, geometric alignment, and consistent patch scoring across conditions. The platform reduces human bias, eliminates the need for diver-based annotation, and enables repeatable, low-cost, and AI-assisted reef monitoring workflows.

     

AutoCoralMatch addresses key needs in environmental modeling and software: transparent algorithmic design, generalizability across reef environments, and reproducibility through publicly available code, pretrained models, and datasets. It supports long-term coral health modeling and can be integrated into decision-support systems for marine conservation.

   
 
 
   

BibTeX

    

@article{AI-CoralWatch,
  author = {Saad Saoud, Lyes et al.},
  title = {AutoCoralMatch: An Open-Source Foundation Model-Driven Framework for Patch-Level Coral Health Assessment and Automated Bleaching Detection},
  year = {2025},
  publisher = {Preprint},
  doi = {......},
  url = {https://arxiv.org/...}}