Comparison of ANN and SVM methods in extraction Land Use/ Land Cover maps from Landsat 8 satellite image (Case Study: Sufi Chay Basin)

Abstract

Land use and land cover maps are necessary for planning and natural resources management. In the way, remote sensing data have special place because of providing update data, repetitive covers and low cost images. Therefore Optimum Land Image/ Thermal Infrared Sensor were used to map land-use and land-cover in 1 and 2 level. Because of, this images are new thus radiometric correct was used ERDAS software model maker. Also Normalize Difference Vegetation Index (NDVI), Bare Soil Index (BI) and Principal Component Analyze (PCA) were used as inputs to improve classification accuracy. On the other hand kernels functional and polynomial ranks of Support Vector Machine method evaluated in side others bands and the best result of SVM method compared with Artificial Neural Network (ANN). The results indicated that SVM method has accuracy: 92% with Kappa Coefficient: 0.91 and ANN method has accuracy: 89% with kappa coefficient: 0.87 also SVM method has a good performance in the regions that, classes show similar spectral behavior.

Keywords

• land use
• landsat 8
• vegetation and bare soil indexes
• ann
• svm