Centre de documentation
scientifique

Dépouillements

A framework for supervised image classification with incomplete training samples / Q. Guo in Photogrammetric Engineering & Remote Sensing, PERS, vol 78 n° 6 (June 2012)  

Public [article]  inPhotogrammetric Engineering & Remote Sensing, PERS > vol 78 n° 6 (June 2012) . - pp 595 - 604 Titre : A framework for supervised image classification with incomplete training samples Type de document : Article/Communication Auteurs : Q. Guo, Auteur ; W. Li, Auteur ; J. Chen, Auteur Année de publication : 2012 Article en page(s) : pp 595 - 604 Note générale : Bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] carte d'occupation du sol [Termes IGN] classification dirigée [Termes IGN] classification par maximum de vraisemblance [Termes IGN] échantillonnage d'image [Termes IGN] extraction de couche Résumé : (Auteur) For traditional supervised classification methods, all land-cover types need to be exhaustively labeled to train the classifier. However, there are situations where the training sample classes are incomplete due to a lack of understanding of ground cover types in the image. In this study we propose a one-by-one (OBO) classification framework to address this incomplete training sample problem. The OBO approach is based on a one-class classifier (positive and unlabeled learning algorithm), and it extracts the land-cover type from the image one at a time. The performance of the proposed method was compared with a traditional supervised classifier using a high spatial resolution image. The average accuracy of the new method is 76.34 percent across different training sample sizes, whereas the accuracy of the classical approach is 66.46 percent, with an increase of 9.88 percent. The results demonstrate that the proposed new framework provides significantly higher classification accuracy than the classical approach at the 95 percent confidence level, and shows promise in dealing with the incomplete training sample problem for supervised image classification. Numéro de notice : A2012-249 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Article DOI : 10.14358/PERS.78.6.595 En ligne : https://doi.org/10.14358/PERS.78.6.595 Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=31695 [article]

Spatial resolution imagery requirements for identifying structure damage in a hurricane disaster: A cognitive approach / S. Battersby in Photogrammetric Engineering & Remote Sensing, PERS, vol 78 n° 6 (June 2012)  

Public [article]  inPhotogrammetric Engineering & Remote Sensing, PERS > vol 78 n° 6 (June 2012) . - pp 625 - 635 Titre : Spatial resolution imagery requirements for identifying structure damage in a hurricane disaster: A cognitive approach Type de document : Article/Communication Auteurs : S. Battersby, Auteur ; M. Hodgson, Auteur ; Jing Wang, Auteur Année de publication : 2012 Article en page(s) : pp 625 - 635 Note générale : Bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] analyse visuelle [Termes IGN] classification à base de connaissances [Termes IGN] cognition [Termes IGN] dommage matériel [Termes IGN] pouvoir de résolution géométrique [Termes IGN] risque naturel [Termes IGN] seuillage d'image [Termes IGN] tempête [Termes IGN] zone sinistrée Résumé : (Auteur) In disaster response, timely collection and exploitation of remotely sensed imagery is of increasing importance. Image exploitation approaches during the immediate (first few days) aftermath of a disaster are predominantly through visual analysis rather than automated classification methods. While the temporal needs for obtaining the imagery are fairly clear (within a one- to three-day window), there have only been educated guesses about the spatial resolution requirements necessary for the imagery for visual analysis. In this paper, we report results from an empirical study to identify the coarsest spatial resolution that is adequate for tasks required immediately following a major disaster. The study was conducted using cognitive science experimental methods and evaluated the performance of individuals with varying image interpretation skills in the task of mapping hurricane-related residential structural damage. Through this study, we found 1.5 m as a threshold for the coarsest spatial resolution imagery that can successfully be used for this task. The results of the study are discussed in terms of the likelihood of collection of this type of imagery within the temporal window required for emergency management operations. Numéro de notice : A2012-250 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Article DOI : 10.14358/PERS.78.6.625 En ligne : https://doi.org/10.14358/PERS.78.6.625 Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=31696 [article]