COMPARISON OF NINE IMAGE CLASSIFICATION METHODS ON LANDSAT 7 IMAGERY

• Main Authors: Victor STRÎMBU, Vlad STRÎMBU, Wesley PALMER, Jean GOURD
• Format: Article
• Language: English
• Published: Editura Universitatii din Oradea 2014-12-01
• Series: Analele Universităţii din Oradea: Seria Geografie
• Subjects: ArcGIS; ERDAS Imagine; gis; landsat; land cover
• Online Access: http://geografie-uoradea.ro/Reviste/Anale/Art/2014-2/7.AUOG_661_Victor.pdf

Nine different widely used classification methods available in ArcGIS and ERDAS software packages weretested on Landsat 7 imagery with the objective to compare their performance and adequacy in classifying six major land cover elements: urban/commercial, residential, bare soil, vegetation, forest and water. A brief background for each classification method was provided, after which the results of each algorithm were visually compared and analyzed. Finally, the kappa coefficient was used as a quantitative metric to asses the agreement between methods. This study showed that different results are obtained when using different classification methods; in consequence the classification method must be carefully selected according to the objective and the available data.. The finality of this work is to provide the average GIS software user with the understanding on how the classification method impacts the classification result, and a starting point in deciding what GIS software tool would be more appropriate given a certain context and goal.