Detecting mangrove seedlings from UAV imagery using deep learning for restoration monitoring

We’re pleased to share a newly published peer-reviewed study led by our colleague Yuri Shendryk, co-authored with Maria P. Vilas, exploring how ultra-high-resolution UAV imagery and deep learning can be used to detect mangrove seedlings for restoration monitoring.

Published in Scientific Reports (2026), the research addresses a critical gap in mangrove restoration - accurately detecting and tracking early seedling (~15cm in size) establishment across large, difficult-to-access coastal environments.

The study evaluates two deep learning approaches - density estimation and object detection - applied to sub-centimeter UAV imagery collected across 22 restoration sites in Abu Dhabi. Results show that a density-based model (MaxViT-UNet) outperformed state-of-the-art object detection method (DETR), achieving an F1-score of 0.70 in identifying mangrove seedlings under challenging field conditions.

Importantly, the research also highlights real-world constraints such as tidal variation, labeling uncertainty, and image resolution - reinforcing why restoration monitoring must be both technically rigorous and operationally realistic.

This work demonstrates how UAV-enabled AI can support scalable, evidence-based monitoring of mangrove restoration, helping practitioners move beyond manual surveys towards repeatable, data-driven insights on early establishment and survival.

Read the full paper here.

Attribution

Shendryk, Y. & Vilas, M. P. (2026). Detecting mangrove seedlings from UAV imagery using deep learning for restoration monitoring. Scientific Reports.