elib
DLR-Header
DLR-Logo -> http://www.dlr.de
DLR Portal Home | Impressum | Datenschutz | Kontakt | English
Schriftgröße: [-] Text [+]

Estimating Soil Parameters from DESIS Images using Deep Learning

Zhao, Xiangyu und Heiden, Uta und Karlshöfer, Paul und Xiong, Zhitong und Zhu, Xiaoxiang (2023) Estimating Soil Parameters from DESIS Images using Deep Learning. WHISPERS 2023, 2023-10-31 - 2023-11-02, Athens, Greece.

[img] PDF
1MB

Kurzfassung

There are several soil parameters that play a significant role in soil health and thus, crop production. Compared with traditional fieldwork by collecting soil samples, digital soil mapping by remote sensing data has many advantages, such as estimating soil properties efficiently in large areas. Multispectral and hyperspectral data have been used already widely for soil parameter retrieval using spectral soil models. Mostly, hyperspectral data have largely outperformed the model and prediction of soil parameters. However, the accuracy and uncertainty of both model results depend largely on the density of calibration points, which is especially problematic for large areas such as regions and countries. Therefore, new methods are needed that take into account the sparsity of calibration data during the training of the model. This work focuses on the SOC content for the whole Bavarian region in Germany (~70.000 km²). The soil data source is LFU (Bayerisches Landesamt fur Umwelt) and LUCAS 2018 (Land Use and Coverage Area Frame Survey). After data selection, we use 1171 soil samples. As for the hyperspectral images, we use DESIS data in Bavaria, whose spectral range is 400-1000 nm. We use 603 hyperspectral images in experiments. To get spectral reflectance for bare soils, we build temporal reflectance composites surrounding each soil sample from the original images. Specifically, We compute the NDVI value for each pixel and then filter pixels by NDVI threshold to filter out vegetated pixels. Temporal composites are then generated by the pixel-based averaging of the filtered images. These composites are fed into the deep learning model. The model would output the SOC value. Regarding the model framework, it consists of CNN layers followed by fully connected layers. To solve the sparsity of data availability, data augmentation, and transfer learning methodology are investigated in this work. During our experiments, we use cross-validation to evaluate the performance. Root Mean Square Error and R Square are used as the evaluation metrics.

elib-URL des Eintrags:https://elib.dlr.de/199237/
Dokumentart:Konferenzbeitrag (Vortrag)
Titel:Estimating Soil Parameters from DESIS Images using Deep Learning
Autoren:
AutorenInstitution oder E-Mail-AdresseAutoren-ORCID-iDORCID Put Code
Zhao, XiangyuTU MunichNICHT SPEZIFIZIERTNICHT SPEZIFIZIERT
Heiden, Utauta.heiden (at) dlr.dehttps://orcid.org/0000-0002-3865-1912NICHT SPEZIFIZIERT
Karlshöfer, Paulpaul.karlshoefer (at) dlr.deNICHT SPEZIFIZIERTNICHT SPEZIFIZIERT
Xiong, Zhitongzhitong.xiong (at) tum.deNICHT SPEZIFIZIERTNICHT SPEZIFIZIERT
Zhu, Xiaoxiangxiao.zhu (at) tum.deNICHT SPEZIFIZIERTNICHT SPEZIFIZIERT
Datum:2023
Referierte Publikation:Nein
Open Access:Ja
Gold Open Access:Nein
In SCOPUS:Nein
In ISI Web of Science:Nein
Status:veröffentlicht
Stichwörter:Hyperspectral, DESIS, soil parameter, machine learning, deep learning, spectral reflectance, bare, temporal composite
Veranstaltungstitel:WHISPERS 2023
Veranstaltungsort:Athens, Greece
Veranstaltungsart:internationale Konferenz
Veranstaltungsbeginn:31 Oktober 2023
Veranstaltungsende:2 November 2023
Veranstalter :IEEE/GRSS
HGF - Forschungsbereich:Luftfahrt, Raumfahrt und Verkehr
HGF - Programm:Raumfahrt
HGF - Programmthema:Erdbeobachtung
DLR - Schwerpunkt:Raumfahrt
DLR - Forschungsgebiet:R EO - Erdbeobachtung
DLR - Teilgebiet (Projekt, Vorhaben):R - Optische Fernerkundung
Standort: Oberpfaffenhofen
Institute & Einrichtungen:Institut für Methodik der Fernerkundung > Photogrammetrie und Bildanalyse
Hinterlegt von: Heiden, Dr.rer.nat. Uta
Hinterlegt am:16 Nov 2023 14:17
Letzte Änderung:24 Apr 2024 20:59

Nur für Mitarbeiter des Archivs: Kontrollseite des Eintrags

Blättern
Suchen
Hilfe & Kontakt
Informationen
electronic library verwendet EPrints 3.3.12
Gestaltung Webseite und Datenbank: Copyright © Deutsches Zentrum für Luft- und Raumfahrt (DLR). Alle Rechte vorbehalten.