Identification of Agricultural Crop Types in Northern Israel using Multitemporal RapidEye Data

Autor(en): Beyer, Florian
Jarmer, Thomas 
Siegmann, Bastian
Stichwörter: ALGORITHMS; crops; DISCRIMINATION; Imaging Science & Photographic Technology; LAND-COVER CLASSIFICATION; LU/LC; MULTISPECTRAL IMAGES; multitemporal classification; phenology; RapidEye; Remote Sensing; SENSED DATA; SEPARABILITY; SUPPORT VECTOR MACHINES; TM IMAGERY; TRAINING DATA; VEGETATION
Erscheinungsdatum: 2015
Herausgeber: E SCHWEIZERBARTSCHE VERLAGSBUCHHANDLUNG
Journal: PHOTOGRAMMETRIE FERNERKUNDUNG GEOINFORMATION
Ausgabe: 1
Startseite: 21
Seitenende: 32
Zusammenfassung: 
Accurate land use / land cover classification (LU/LC) of agricultural crops still represents a major challenge for multispectral remote sensing. In order to obtain reliable classification accuracies on the basis of multispectral satellite data, merging crop classes in rather broad classes is often necessary. With regard to the rising availability and the improving spatial resolution of satellite data, multitemporal analyses become increasingly important for remote sensing investigations. For the separation of spectrally similar crops, multidate satellite images include different growth characteristics during the phenological period. The present study aims at investigating a way to perform highly accurate classifications with numerous agricultural classes using multitemporal Rapid Eye data. The Jeffries-Matusita separability (JM) was used for applying a pre-procedure in order to find the best multitemporal setting of all available images within one crop cycle, consisting of two cultivation periods P1 with 16 agricultural classes and P2 with 27 agricultural classes. Only one critical class pairing occurred for both PI and P2 taking into account the best multitemporal dataset. The maximum likelihood (ML) classifier and the support vector machine (SVM) were compared using the most suitable multitemporal images. Both algorithms achieved very high overall accuracies (OAA) of over 90%. SVM was slightly better with a classification accuracy of P1-OAA = 96.13% and P2-OAA = 94.01%. ML provided a result of OAA = 94.83% correctly classified pixels for P1 and OAA = 93.28% for P2. The processing time of ML, however, was significantly shorter compared to SVM, in fact by a factor of five.
ISSN: 14328364
DOI: 10.1127/pfg/2015/0249

Show full item record

Page view(s)

3
Last Week
0
Last month
0
checked on Feb 21, 2024

Google ScholarTM

Check

Altmetric