Centre dedocumentation
scientifique

Forest object-oriented classification with customized and automatic attribute selection / Olivier de Joinville (2010)    

Public Titre : Forest object-oriented classification with customized and automatic attribute selection Type de document : Article/Communication Auteurs : Olivier de Joinville , Auteur Editeur : International Society for Photogrammetry and Remote Sensing ISPRS Année de publication : 2010 Collection : International Archives of Photogrammetry and Remote Sensing, ISSN 0252-8231 num. 38-8 Conférence : ISPRS 2010, Technical Commission 8 Symposium, Networking the World with Remote Sensing 02/08/2010 12/08/2010 Kyoto Japon ISPRS OA Archives Commission 8 Importance : pp 669 - 674 Format : 21 x 30 cm Note générale : bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] classification barycentrique [Termes IGN] classification orientée objet [Termes IGN] composition colorée [Termes IGN] forêt domaniale [Termes IGN] image SPOT [Termes IGN] matrice de confusion [Termes IGN] segmentation d'image [Termes IGN] Somme (80) Résumé : (auteur) This paper presents a semi-automatic method to optimize object-oriented classification without photointerpretation. The thematic studied is the forest (Crecy forest in the north of France). A SPOT 2 image at 20 m spatial resolution was analysed in a near infrared colour composite (green, red and infrared). New classification methods no longer work with pixels, but with regions derived from the previously segmented image [TRIAS 2006], [BENCHERIF 2009].The first step consists in image segmentation based on several criteria, a scale parameter and an homogeneity factor made up of two complementary factors: shape and radiometry. Two segmentations have been computed: one at very large scale (no more than 20 regions) in order to establish a manually made classification with only 2 classes: forest and no forest (this latter will not be classified). Another one at a smaller scale which will be used to select the test samples (also called training area) on the forest area. Once both segmentations and manual classification are completed and validated (essentially visually), the objective of this study is to determine semi automatically the most adapted attributes for each training area (5 training areas have been selected). Therefore, for all selected training areas, attributes are automatically selected, consecutively based on three criteria: radiometry, shape and texture. For each of these criteria, a maximum number of attributes is fixed among all potentially interesting attributes and the optimum attribute combination is automatically selected with respect to a statistical parameter derived from a distance matrix. The distance matrix optimizes the separation between the training areas. Then, 3 classifications were set up, each of them with the optimum automatically selected attribute combination derived from the previous step. For each of these classifications, a confusion matrix will be computed. For each training area its confusion rate with other training areas was computed and the lowest confusion rate was selected as the criterion. For instance, if there is a training area which has 35 % of confusion pixels with other classes for a radiometric combination, 25% for a textural combination and 5 % for a morphologic one (shape criterion), this training area will be affected with a morphologic attribute combination. The result is thus a new classification with the new customized attributes for each training area. In the assessment of this classification, the confusion rate for each class decreases significantly. Then, reliability maps are built to determine the risk of confusion between the classes. Test results are so far encouraging. Due to this new method, the confusion rates decrease significantly with respect to a standard nearest neighbour approach. Numéro de notice : C2010-032 Affiliation des auteurs : ENSG (1941-2011) Thématique : IMAGERIE Nature : Communication nature-HAL : ComAvecCL&ActesPubliésIntl DOI : sans En ligne : http://www.isprs.org/proceedings/XXXVIII/part8/pdf/W07P02_20100218000017.pdf Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=90642

Documents numériques

en open access Forest object-oriented classification ... - pdf éditeur Adobe Acrobat PDF

Updating and improving the accuracy of a large 3D database through the careful use of GCPs and Icesat data : Example of REFERENCE3D / Emilie Le Hir (2010)  

Public Titre : Updating and improving the accuracy of a large 3D database through the careful use of GCPs and Icesat data : Example of REFERENCE3D Type de document : Article/Communication Auteurs : Emilie Le Hir, Auteur ; Laurent Cunin , Auteur ; Marc Bernard, Auteur Editeur : International Society for Photogrammetry and Remote Sensing ISPRS Année de publication : 2010 Collection : International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, ISSN 1682-1750 num. 38-part8 Conférence : ISPRS 2010, Technical Commission 8 Symposium, Networking the World with Remote Sensing 02/08/2010 12/08/2010 Kyoto Japon ISPRS OA Archives Commission 8 Importance : pp 800 - 804 Format : 21 x 30 cm Note générale : bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Bases de données localisées [Termes IGN] données ICEsat [Termes IGN] image SPOT 5 [Termes IGN] image SPOT-HRG [Termes IGN] modèle numérique de surface [Termes IGN] point d'appui [Termes IGN] Référence-3D Résumé : (auteur) Onboard SPOT 5, the HRS instrument systematically collects stereopairs around the Globe since 2002. Each stereopair can encompass an area up to 600 km x 120 km within a single pass (i.e. 72 000 km² stereoscopic strips). From this time, SPOT 5 stereoscopic imagery becomes one of main satellite data sources for accurate DEM extraction. Spot Image and French National Cartographic Institute (IGN) decided in 2002 to design and build a worldwide accurate database called Reference3D™ using HRS data. This database consists of three information layers: Digital Elevation Model at 1-arc-second resolution (DTED level 2), Orthoimage at 5m resolution and Quality Masks. Huge efforts have been made to standardize the process in order to offer affordable prices. Numéro de notice : C2010-035 Affiliation des auteurs : IGN+Ext (1940-2011) Thématique : IMAGERIE Nature : Communication nature-HAL : ComAvecCL&ActesPubliésIntl DOI : sans En ligne : https://www.isprs.org/proceedings/XXXVIII/part8/pdf/W08L42_20100304211526.pdf Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=91729