Research & Publications

Science in Service
of Country

AI First Nations partners with leading universities and research institutions to develop peer-reviewed environmental intelligence tools — grounded in Traditional Knowledge and validated by rigorous academic methodology.

97%Species Classification
98.96%Binary Detection
477Training Images
3Species Classified
1,000+Students · AATT
Featured Publication

AI First Nations – Data Science Project Final Report

RMIT University · COSC3043 · Semester 1, 2026

AI-Powered Mangrove Detection & Species Classification

Kirtan Shah · Thomas Blanks · Jake Dyl
Academic Supervisor: Dr Damiano Spina · Industry Supervisor: John Fejo

Peer Reviewed · June 2026

Acknowledgement of Country

"We acknowledge the Woi wurrung, Boon wurrung, and Wathaurong peoples of the Kulin Nation upon which training, testing and deployment of our machine learning model was conducted. We also acknowledge the Traditional Owners of the lands where data was collected. We pay our respects to Elders, past, present and emerging."

Abstract

A machine learning pipeline for detecting and classifying mangrove species from drone-captured aerial imagery. The system employs a two-stage pipeline: a binary classifier determines whether an image contains mangrove canopy, which if detected is passed to a multi-class species classifier. The EfficientNet-B0 advanced model achieved 98.96% binary detection accuracy and 97% species classification accuracy across yellow, orange, and red mangroves. All models were integrated into a locally-hosted Flask web application for field deployment by Indigenous rangers and land managers.

Model Performance Summary

Model Accuracy F1-Score
Random Forest Binary Classifier (Baseline) 93.75% 89.74%
EfficientNet-B0 Binary Classifier (Advanced) 98.96% 98.89%
EfficientNet-B0 Species Classifier (3-Class) 97% 97%

Research Team

Kirtan Shah
Team Lead · Supervisor Liaison

Weekly meetings, supervisor communication, tree-counting prototype development.

Jake Dyl
Model Development

Developed both neural network models — EfficientNet binary and species classifiers.

Thomas Blanks
Flask App · Project Management

Built the Flask web application and integrated ML models into the field interface.

Research Pipeline

What Comes Next

Phase 2 · Sem 2 2026

Autonomous Sub-Canopy Navigation

RMIT postgraduate project developing autonomous drone navigation beneath rainforest canopy — monitoring creek lines, root systems, and biodiversity indicators invisible from above.

Ongoing · KJR

Independent Model Validation

KJR (Dr Kelvin Ross & Aaron Bell) conducting independent evaluation and validation of the AI classification system for commercial deployment and government tender readiness.

Horizon · International

Cross-Location & Global Expansion

Cross-location validation across additional TO Country sites. International market development for mangrove monitoring across Southeast Asia and the Pacific.

Research Partners

Our Academic & Research Network

🎓

RMIT University

Melbourne, VIC · COSC3043

Primary research partner for AI mangrove classification and future sub-canopy autonomous systems. Academic supervisor: Dr Damiano Spina.

🔬

KJR

kjr.com.au · Independent Assurance

Independent AI system evaluation and technology assurance. Dr Kelvin Ross and Aaron Bell lead model validation and commercial readiness assessment.

🌊

Salty Monkeys

saltymonkeys.com.au · Torres Strait

Marine debris and sea country research partner. Collaborative drone AI applications for ghost net identification and coastal monitoring across the Torres Strait.

🤝

Interested in Partnering?

Universities · Government · Industry

We welcome research collaborations with universities, government agencies, and environmental organisations working on Indigenous land management and AI.