Many supervised classification algorithms have been proposed, however, they\nare rarely evaluated for specific application. This research examines the performance\nof machine learning classifiers support vector machine (SVM),\nneural network (NN), Random Forest (RF) against maximum classifier\n(MLC) (traditional supervised classifier) in forest resources and land cover\ncategorization, based on combination of Advanced Land Observing Satellite\n(ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR) and\nLandsat Thematic Mapper (TM) data, in Northern Tanzania. Various data\ncategories based on Landsat TM surface reflectance, ALOS PALSAR backscattering\nand their derivatives were generated for various classification scenarios.\nThen a separate and joint processing of Landsat and ALOS PALSAR\ndata were executed using SVM, NN, RF and ML classifiers. The overall classification\naccuracy (OA), kappa coefficient (KC) and F1 score index values were\ncomputed. The result proves the robustness of SVM and RF in classification of\nforest resource and land cover using mere Landsat data and integration of\nLandsat and PALSAR (average OA = 92% and F1 = 0.7 to 1). A two sample\nt-statistics was utilized to evaluate the performance of the classifiers using different\ndata categories. SVM and RF indicate there is no significance difference\nat 5% significance level. SVM and RF show a significant difference when\ncompared to NN and ML. Generally, the study suggests that parametric classifiers\nindicate better performance compared to parametric classifier.
Loading....