Centre de documentation
scientifique

Improving remote sensing classification: A deep-learning-assisted model / Tsimur Davydzenka in Computers & geosciences, vol 164 (July 2022)  

Public [article]  inComputers & geosciences > vol 164 (July 2022) . - n° 105123 Titre : Improving remote sensing classification: A deep-learning-assisted model Type de document : Article/Communication Auteurs : Tsimur Davydzenka, Auteur ; Pejman Tahmasebi, Auteur ; Mark Carroll, Auteur Année de publication : 2022 Article en page(s) : n° 105123 Note générale : bibliographie Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] apprentissage profond [Termes IGN] classification par réseau neuronal convolutif [Termes IGN] données d'entrainement (apprentissage automatique) [Termes IGN] image à haute résolution [Termes IGN] modèle stochastique [Termes IGN] précision de la classification Résumé : (auteur) In many industries and applications, obtaining and classifying remote sensing imagery plays a crucial role. The accuracy of classification, in particular the machine learning methods, mainly depends on a multitude of factors, among which one of the most important ones is the amount of training data. Obtaining sufficient amounts of training data, however, can be very difficult or costly, and one must find alternative ways to improve the accuracy of predictions. To this end, a possible solution that we provide in this study is to use a stochastic method for producing variations of the training images that will retain the important class-wide features and thereby enrich the machine learning's “understanding” of the variabilities. As such, we applied a stochastic algorithm to produce additional realizations of the limited input imagery and thereby significantly increase the final overall accuracy in a deep learning method. We found that by enlarging the initial training set by additional realizations, we are able to consistently improve classification accuracy, compared with generic image augmentation approaches. The results of this study show that there is a great opportunity to increase the accuracy of predictions when enough data are not available. Numéro de notice : A2022-388 Affiliation des auteurs : non IGN Thématique : IMAGERIE Nature : Article DOI : 10.1016/j.cageo.2022.105123 Date de publication en ligne : 29/04/2022 En ligne : https://doi.org/10.1016/j.cageo.2022.105123 Format de la ressource électronique : URL article Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=100672 [article]