An efficient hybrid approach to the remote sensing land cover classification
Résumé
The support vector machine (SVM) is considered as one of the powerful algorithm in machine learning, so, it is has been widely used in remote sensing land cover classification field. However, this technique suffers from its sensitivity to the parameters setting and training set. For that, we propose in this research an interesting selection of parameters setting and a hybridization of K- means algorithm to SVM in order to filter an optimal training set. A proposed approach is experimented on Spot 5 images including variable objects like roads, vegetation, buildings..etc. The obtained results carried out that SVM hybridized to K-means approach significantly outperforms the SVM approach used alone, with higher classification accuracy, much fewer input training data and a low computational time.