Details
| Originalsprache | Englisch |
|---|---|
| Seiten (von - bis) | 307-315 |
| Seitenumfang | 9 |
| Fachzeitschrift | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Jahrgang | 5 |
| Ausgabenummer | 2 |
| Publikationsstatus | Veröffentlicht - 17 Mai 2022 |
| Veranstaltung | 2022 24th ISPRS Congress on Imaging Today, Foreseeing Tomorrow, Commission II - Nice, Frankreich Dauer: 6 Juni 2022 → 11 Juni 2022 |
Abstract
We present an approach for detecting early signs for upcoming forest damages by training a Convolutional Neural Network (CNN) for the pixel-wise prediction of the remaining life-time (RLT) of trees in forests based on Sentinel-2 imagery. We focus on a scenario in which reference data are only available for a related task, namely for a bi-temporal pixel-wise classification of forest degradation. This reference is used to train a CNN for the pixel-wise prediction of forest degradation. In this context, we propose a new sub-sampling-based approach for compensating the effects of a heavy class imbalance in the training data. Using the resulting classification model, we predict semi-labels for images of a Sentinel-2 time series, from which training data for a CNN designed to regress the RLT can be derived after some label cleansing. However, due to data gaps in the time series, e.g. caused by clouds, only intervals can be derived for the target variable to be regressed, and for some training pixels one of the interval limits may even be unknown. Consequently, we propose a new loss function for training a CNN for regressing the RLT that only requires the known interval limits. The method is evaluated on a data set in Germany, covering a time-span of 5 years. We show that the proposed sub-sampling strategy for dealing with strong label imbalance when training the classifier significantly reduces the training time compared to other approaches. We further show that our model predicts the RLT with a maximum error of two months for 80% of the forest pixels that die within one year from the acquisition date of the Sentinel-2 image.
ASJC Scopus Sachgebiete
- Physik und Astronomie (insg.)
- Instrumentierung
- Umweltwissenschaften (insg.)
- Umweltwissenschaften (sonstige)
- Erdkunde und Planetologie (insg.)
- Erdkunde und Planetologie (sonstige)
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
in: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Jahrgang 5, Nr. 2, 17.05.2022, S. 307-315.
Publikation: Beitrag in Fachzeitschrift › Konferenzaufsatz in Fachzeitschrift › Forschung › Peer-Review
}
TY - JOUR
T1 - DEEP LEARNING FOR THE DETECTION OF EARLY SIGNS FOR FOREST DAMAGE BASED ON SATELLITE IMAGERY
AU - Wittich, Dennis
AU - Rottensteiner, Franz
AU - Voelsen, Mirjana
AU - Heipke, Christian
AU - Müller, Sonke
N1 - Funding information: This work was partially funded by the Federal Ministry for Economic Affairs and Climate Action, Germany (Bundesmin-isterium für Wirtschaft und Klimaschutz, Funding codes 50EE2017A and 50EE2017B).
PY - 2022/5/17
Y1 - 2022/5/17
N2 - We present an approach for detecting early signs for upcoming forest damages by training a Convolutional Neural Network (CNN) for the pixel-wise prediction of the remaining life-time (RLT) of trees in forests based on Sentinel-2 imagery. We focus on a scenario in which reference data are only available for a related task, namely for a bi-temporal pixel-wise classification of forest degradation. This reference is used to train a CNN for the pixel-wise prediction of forest degradation. In this context, we propose a new sub-sampling-based approach for compensating the effects of a heavy class imbalance in the training data. Using the resulting classification model, we predict semi-labels for images of a Sentinel-2 time series, from which training data for a CNN designed to regress the RLT can be derived after some label cleansing. However, due to data gaps in the time series, e.g. caused by clouds, only intervals can be derived for the target variable to be regressed, and for some training pixels one of the interval limits may even be unknown. Consequently, we propose a new loss function for training a CNN for regressing the RLT that only requires the known interval limits. The method is evaluated on a data set in Germany, covering a time-span of 5 years. We show that the proposed sub-sampling strategy for dealing with strong label imbalance when training the classifier significantly reduces the training time compared to other approaches. We further show that our model predicts the RLT with a maximum error of two months for 80% of the forest pixels that die within one year from the acquisition date of the Sentinel-2 image.
AB - We present an approach for detecting early signs for upcoming forest damages by training a Convolutional Neural Network (CNN) for the pixel-wise prediction of the remaining life-time (RLT) of trees in forests based on Sentinel-2 imagery. We focus on a scenario in which reference data are only available for a related task, namely for a bi-temporal pixel-wise classification of forest degradation. This reference is used to train a CNN for the pixel-wise prediction of forest degradation. In this context, we propose a new sub-sampling-based approach for compensating the effects of a heavy class imbalance in the training data. Using the resulting classification model, we predict semi-labels for images of a Sentinel-2 time series, from which training data for a CNN designed to regress the RLT can be derived after some label cleansing. However, due to data gaps in the time series, e.g. caused by clouds, only intervals can be derived for the target variable to be regressed, and for some training pixels one of the interval limits may even be unknown. Consequently, we propose a new loss function for training a CNN for regressing the RLT that only requires the known interval limits. The method is evaluated on a data set in Germany, covering a time-span of 5 years. We show that the proposed sub-sampling strategy for dealing with strong label imbalance when training the classifier significantly reduces the training time compared to other approaches. We further show that our model predicts the RLT with a maximum error of two months for 80% of the forest pixels that die within one year from the acquisition date of the Sentinel-2 image.
KW - Deep Learning
KW - Forest Monitoring
KW - Label Imbalance
KW - Regression
KW - Sentinel-2
UR - http://www.scopus.com/inward/record.url?scp=85132279336&partnerID=8YFLogxK
U2 - 10.5194/isprs-annals-V-2-2022-307-2022
DO - 10.5194/isprs-annals-V-2-2022-307-2022
M3 - Conference article
AN - SCOPUS:85132279336
VL - 5
SP - 307
EP - 315
JO - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
JF - ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
SN - 2194-9042
IS - 2
T2 - 2022 24th ISPRS Congress on Imaging Today, Foreseeing Tomorrow, Commission II
Y2 - 6 June 2022 through 11 June 2022
ER -