Search

In remote sensing, there is a large number of algorithms to classify a satellite image. Among these classification algorithms, the Random Forest appears to be particularly powerful. The objectives of this study are to evaluate (1) the importance of image selection for the level of accuracy of the training model and (2) the nature of the relationship between the level of accuracy of the model and the overall accuracy of the thematic map resulting from the classification of the satellite image with this algorithm. Based on a Landsat 8 OLI image taken over a tropical mountain area : the West Cameroon region, 35 models were built and tested. Analysis shows that the level of overall accuracy of the Random Forest results are closely dependent on the accuracy of the training model used to classify the satellite image, and on the choice of the images used to train this model. Moreover, the selection of these images is itself dependent on the quality of the Regions of interest (ROI) that will be used to build the model. It is therefore important to place particular emphasis on the quality of the input data in order to guarantee satisfactory results with this algorithm. Otherwise, the performance of this algorithm could be inferior to that of other classification algorithms. [ABSTRACT FROM AUTHOR]