06-12-2017, 03:39 PM
Soil cover classification is one of the remote sensing applications, in order to identify characteristics such as land use using typically multispectral satellite data. Numerous algorithms have been developed for classification purposes and different classifiers have their own characteristics. The different data and the study area, especially the complexity of the landscape, have a different impact on the different classifiers. Therefore, the objective of this study is to compare the approaches of Neural Network and Maximum Likelihood and to find an adequate classifier in the classification of the land cover using satellite images of medium spatial resolution in the tropical equatorial region. These two classifiers were tested using data from the Landsat Thematic Mapper on the island of Penang, Malaysia, using the same training sample data sets. Five classes of land cover were classified: forest, pasture, urban, water and clouds. In addition, the study was also carried out to obtain the performances of both classifiers for the purpose of terrestrial coverage mapping. The general classification accuracy and the Kappa coefficient were calculated.