Parameter investigation of Artificial Neural Network and Support Vector Machine for image classification

The use of satellite data for LU/LC classification is an important research goal. Forests degradation and deforestation aggravate global warming, and the LU/LC information obtained from satellite data is now very essential for global environmental conservation and ecosystem. Especially the typical example of getting the accurate LU/LC information is the reducing emission from forest deforestation and degradation. The LU/LC information is valuable because it can be obtained on regular basis even across the vast areas like natural forests. Different researchers used different methods and tools for LU/LC classification such as Parallelepiped, Minimum Distance, Binary Encoding, Spectral Angle Mapper, Spectral Information Divergence as supervised classifiers. All these methods used for different purposes and with different parameters. This research is formulated to evaluate the performance of the Artificial Neural Network & Support Vector Machine for Land Use and Land Cover (LU/LC) classification using different parameters such as (Training Rate, Training Threshold Contribution, Training Momentum) in ANN and (Penalty Parameter, Pyramid Level) in SVM using ENVI 4.7. The Abbottabad test patch is used for LU/LC classification. Five classes Agriculture Land, Settlements, Water Bodies, Forest and Barren Land are used in this research. The ANN & SVM classifiers are applied to the Abbottabad test patch using different parameters and evaluate the best classifier according to their accuracy in confusion matrix. The ANN gives the overall accuracy of 99.35% and SVM gives the overall accuracy of 99.88%. From simulation it showed that SVM gives better result than ANN in terms of accuracy.