Dimas Ayyub Alghafiqi, Aulia Rahman, Maya Fitria, Yudha Nurdin, Rahmad Dawood
Mangrove ecosystems play a vital role in coastal protection, biodiversity conservation, and carbon sequestration but remain under severe threat from deforestation and climate change. This study presents a deep learning–based approach for automated mangrove detection using UAV imagery and a modified CenterNet architecture with an EfficientNet-B0 backbone. A UAV-derived dataset containing 411 high-resolution images and over 20,000 annotated instances across three mangrove genera—Rhizophora, Avicennia, and Nypa—was employed. Data augmentation techniques, including horizontal flipping and 90° rotation, were applied to improve model generalization. The model was trained for 100 epochs with a batch size of 16, a learning rate of 1×10-4, and the Adam optimizer, utilizing a combination of heatmap, offset, and size losses. Training and evaluation were conducted on a Google Colab environment with an NVIDIA T4 GPU. Experimental results achieved an overall mean Average Precision (mAP) of 0.4530, precision of 0.7803, and recall of 0.8319. Class-wise performance showed the highest Average Precision (AP) for Rhizophora (0.7467) and the lowest for Avecennia (0.1633). These findings highlight the potential of the CenterNet–EfficientNet-B0 framework for efficient mangrove mapping and monitoring using UAV imagery. ©2025 IEEE.
Department of Electrical and Computer Engineering, Universitas Syiah Kuala, Aceh, Indonesia; Department of Electrical and Computer Engineering, Universitas Syiah Kuala, Interactive Intelligent Systems Research Group, Universitas Syiah Kuala, Aceh, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock