Example of object based image analysis applied to coarse resolution images. Application to landscape classification in France
Object based image analysis (OBIA) is a powerful technique for classification of remote sensing images. Although it is generally used with very high resolution images, its application is not linked to a particular spatial resolution, and OBIA can be applied to classify coarse resolution images, such as MODIS, Meris, and the future Sentinel-3 sensor, OLCI. In this study, a segmentation of the French landscapes is made from MODIS images, including vegetation and texture indices, by applying OBIA. Different segmentations have been generated using different segmentation parameters and input variables. Since no ground data is available for training and validating the classification, unsupervised evaluation methods are used to select the best input variables and the best segmentation parameters. The best segmentation is shown to be the one including texture indices, and leads to 84 radiometrically homogeneous regions. From the results of the segmentation, a non supervised classification is performed and 36 different classes are identified.
| Main Authors: | , , |
|---|---|
| Format: | conference_item biblioteca |
| Language: | eng |
| Published: |
s.n.
|
| Subjects: | U30 - Méthodes de recherche, B10 - Géographie, |
| Online Access: | http://agritrop.cirad.fr/566815/ http://agritrop.cirad.fr/566815/1/document_566815.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Object based image analysis (OBIA) is a powerful technique for classification of remote sensing images. Although it is generally used with very high resolution images, its application is not linked to a particular spatial resolution, and OBIA can be applied to classify coarse resolution images, such as MODIS, Meris, and the future Sentinel-3 sensor, OLCI. In this study, a segmentation of the French landscapes is made from MODIS images, including vegetation and texture indices, by applying OBIA. Different segmentations have been generated using different segmentation parameters and input variables. Since no ground data is available for training and validating the classification, unsupervised evaluation methods are used to select the best input variables and the best segmentation parameters. The best segmentation is shown to be the one including texture indices, and leads to 84 radiometrically homogeneous regions. From the results of the segmentation, a non supervised classification is performed and 36 different classes are identified. |
|---|