Résumé:
Since the late eighties, the development of land-use maps, based on techniques for classifying satellite imagery, has been a very important step forward. Several methods that take into account the spectral, textural and contextual dimensions have been developed. This work attempts through the study of the object-oriented classification, experimented on an extract of image Landsat8_OLI acquired on 05/05/2014 of the region of the watershed of the Oued Kramis, western region of Algeria, d To evaluate the contribution of this approach to the classification of land use. The segmentation tests carried out during the experiment on this extract were very beneficial. This experiment also confirmed that the manipulation of the Scal level and Merge level parameters of the image during the segmentation process requires several combinations between these two variables as well as other contextual attributes arising from the ground truth, other multi-sources, to refine the results of the segmentation. The num
ber of classes interpreted for the object-oriented classification on our working area was equal to 10 classes. The results obtained by the chi-square statistical method show that the use of the Scal level (SL) of 60% and 70% with a level of Merge level (ML) of 80% or 90% and that of 60 % With a Merge Level (ML) of 60% respectively gives high compatibility for the 10 land-use classes of the area of interest.
