Mapping Plant Functional Traits with Convolutional Neural Networks Trained on Spectral Images of Discrete Anisotropic Radiative Transfer Model

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F86652079%3A_____%2F25%3A00645151" target="_blank" >RIV/86652079:_____/25:00645151 - isvavai.cz</a>

Výsledek na webu

<a href="https://ieeexplore.ieee.org/xpl/conhome/11242230/proceeding" target="_blank" >https://ieeexplore.ieee.org/xpl/conhome/11242230/proceeding</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/IGARSS55030.2025.11242993" target="_blank" >10.1109/IGARSS55030.2025.11242993</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Mapping Plant Functional Traits with Convolutional Neural Networks Trained on Spectral Images of Discrete Anisotropic Radiative Transfer Model

Popis výsledku v původním jazyce

In this study, we demonstrated how spectral images of vegetation canopies generated by physically-based radiative transfer modeling can be used for machine learning of convolutional neural networks (CNN) capable of mapping plant biochemical functional traits. Results of ResNet50 CNN, applied on multispectral reflectance images of maize and sugar beet crop canopies simulated with the Discrete Anisotropic Radiative Transfer (DART) model, indicated acceptable accuracies of leaf chlorophyll a+b content estimates for early growth stages of maize and sugar beet crops (R2 = 0.83, RMSE = 6.9 and 6.6 μg.cm-2). Estimations for the later and across different growth stages of both crops were, however, inaccurate. U-Net CNN, trained on DART reflectance images of a tall eucalyptus forest and applied on a drone-based hyperspectral image, retrieved content of leaf chlorophylls comparably to a random forest, but estimates of carotenoids and anthocyanin were underestimated.

Název v anglickém jazyce

Mapping Plant Functional Traits with Convolutional Neural Networks Trained on Spectral Images of Discrete Anisotropic Radiative Transfer Model

Popis výsledku anglicky

In this study, we demonstrated how spectral images of vegetation canopies generated by physically-based radiative transfer modeling can be used for machine learning of convolutional neural networks (CNN) capable of mapping plant biochemical functional traits. Results of ResNet50 CNN, applied on multispectral reflectance images of maize and sugar beet crop canopies simulated with the Discrete Anisotropic Radiative Transfer (DART) model, indicated acceptable accuracies of leaf chlorophyll a+b content estimates for early growth stages of maize and sugar beet crops (R2 = 0.83, RMSE = 6.9 and 6.6 μg.cm-2). Estimations for the later and across different growth stages of both crops were, however, inaccurate. U-Net CNN, trained on DART reflectance images of a tall eucalyptus forest and applied on a drone-based hyperspectral image, retrieved content of leaf chlorophylls comparably to a random forest, but estimates of carotenoids and anthocyanin were underestimated.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20705 - Remote sensing

Projekt

—

Návaznosti

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

IEEE International Symposium on Geoscience and Remote Sensing (IGARSS 2025)

ISBN

979-8-3315-0810-4

ISSN

—

e-ISSN

—

Počet stran výsledku

5

Strana od-do

355-359

Název nakladatele

IEEE

Místo vydání

Brisbane

Místo konání akce

Brisbane

Datum konání akce

3. 8. 2025

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—