What question did this study set out to answer?

The primary aim is to improve the accuracy of tree species mapping through advanced AI techniques and UAV data integration.

July 10, 2026ISPRS annals of the photogrammetry, remote sensing and spatial information sciences0 citationsOpen Access

Synergizing Foundation Model Transfer and Phenological Information for Fine-Grained Forest Segmentation

YGYoucef Ben GhorbelDeutsches Zentrum für Luft- und Raumfahrt e. V. (DLR)GMGuneet MutrejaDeutsches Zentrum für Luft- und Raumfahrt e. V. (DLR)MWMareike WeishauptDeutsches Zentrum für Luft- und Raumfahrt e. V. (DLR)

Key Points

The primary aim is to improve the accuracy of tree species mapping through advanced AI techniques and UAV data integration.
Investigated a two-phase framework combining a foundation model and UAV imagery.
Utilized a Vision Transformer-based model (FoMo-Net) to initialize a DeepLabv3+ architecture.
Employed multi-temporal UAV RGB data through change composites and pseudo-labeling for analysis.
Achieved an Overall Accuracy (OA) of 78.21% in tree species mapping.
Foundation model initialization significantly enhanced feature generalization.
Utilizing multi-temporal cues proved crucial for distinguishing closely related species.

Abstract

Abstract. Accurate mapping of tree species is fundamental for sustainable forest management, biodiversity monitoring, and ecological research. Recent advances in Uncrewed Aerial Vehicle (UAV) photogrammetry provide rich spatial and spectral information at unprecedented detail (0.02 m Ground Sampling Distance). Meanwhile, the development of large-scale foundational models, pre-trained on expansive, multi-modal remote sensing datasets, offers highly transferable representations critical for advancing geoscience applications. This paper investigates the underexplored integration of foundation model pre-training with multi-temporal high-resolution UAV imagery. We introduce a novel two-phase framework: first, leveraging a Vision Transformer (ViT)-based foundation model (FoMo-Net) pre-trained on the multi-scale FoMo-Bench benchmark to initialize a DeepLabv3+ architecture; and second, fusing multi-temporal UAV data (May and September RGB) through change composites and pseudo-labeling to exploit species-specific phenology. The proposed approach, tested on the multi-class Québec Trees Dataset, yields an Overall Accuracy (OA) of 78.21%, demonstrating that foundation model initialization significantly boosts feature generalization, while multi-temporal cues are essential for disambiguating closely related species.

问 AI

Bookmark

View Full Paper