Capability of Land Cover Mapping in Local-Scale Using Supervised Algorithms (Case study: Guilan Province)

There was a possibility to study earth coverage on a large scale using remote sensing data. The support vector machines (SVM), artificial neural network ‏)ANN( and maximum likelihood )ML‏( algorithms were used to Land cover classification on OLI sensors data and 4 kernels in Guilan province.

Classifications were based on training samples of 10 different covers in the entire Guilan province. To improve the classification accuracy on OLI image data, the MODIS atmospheric products used in 6SV atmospheric correction model. The OLI atmospheric corrected image segmented to 219000 polygons based on homogeneity. In this study 2% of polygons were used to test and training samples by the random statistical method. Polygons labeled to classes by field survey.

Applying ANN, SVM and ML algorithms on the OLI images after atmospheric corrected by 6SV model, the overall accuracy of classification improved 0.11%, 0.8%, and 1.9% respectively. The results indicated that the land cover map by RBF-SVM had overall accuracy and kappa coefficient with 75.6% and 0.72 respectively. In this algorithm accuracy of agriculture, range‏ shrub land and water body classes were ‏93.16%, 72.55% and‏ 96.57% respectively. The results of this study indicated that SVM algorithm improved overall accuracy 1.67% compared to the ML algorithm.

This research indicated that in land cover classification and mapping of Guilan province, the nonparametric SVM algorithm had more accurate than the ML parametric algorithm. According to the results of this research, it is suggested that atmospheric correction models should be used especially on the large and local images.