Comparison of Support Vector Machine, Neural Network, and Maximum Likelihood Methods for the Separation of Lithological Units

Since the spectrum behavior of rocks is similar, preparing of geological maps by using of multi-spectral images could be difficult. In this study, the Support Vector Machine (SVM) is used as one of the image classification methods, which has flexibility for individual conditions. The kernels of SVM are performed by Maximum Likelihood (MLC) and Neural Network Classification (NNC) methods in order to produce geological maps and training samples of hand and laboratory samples, as well. The results show that the SVM method has highest precision (83.42%) in all three kernels in comparison with other two methods. Furthermore, this method can reach the 100% precision if uses only 50% of training samples, while in other methods (i.e. MLC and NNC) cannot do so. Comparing of results that gained from Jefferis-Matusita method with those from SVM indicate that this method is much more operative than other ones for low discriminative data, and then its efficiency is high for producing geological maps.