Évaluation des méthodes de classifications dirigées (spectrale et orientée objet) sur les images satellitaires à THRS

Abstract

This study aims to analyze the performance of spectral classifications and object oriented classification, for mapping of the urban area of Cocody and Attécoubé. To achieve our aim, a satellite image at very high spatial resolution (VHSR) was used. For this work, these two methods of classifications were successively applied to three test sites, namely a well-built area, an area with poor housing and an undeveloped area. The results of two treatments are compared to analyze performances. It appears from these treatments that object-oriented classification is better suited than the spectral approach for mapping sites built because it improves the mapping accuracy of 4,58%. At sites with poor housing, the two classifications show almost similar comprehensive details. Thus, the spectral classification provides an improvement of 0,73% compared to the object-oriented approach. On undeveloped land, the spectral classification is best shown with an overall improvement of 8.22% compared to the object-oriented approach. In short, the object-oriented classification seems appropriate for the extraction of units of well-structured sites. When the level of organization of the site decreases, the performance of the object oriented classification fall in favour of the spectral classification.urbanization, spectral classification, object-oriented classification, remote sensing, Côte d’Ivoire

Keywords

• Côte d’Ivoire
• urbanisation
• télédétection
• classification spectrale
• classification orientée objet