utilization of neural network and Maximum Likelihood Algorithms in classification of satellite images to extract land cover bahmanshir and arvand coastal zone

The main purpose of satellite image processing is preparing thematic and efficient maps, so choosing appropriate classification algorithm has important role in this case. In parametric methods such as maximum likelihood main problem is their dependence on the statistical distribution of input data. Artificial neural network is nonparametric classification method which is not dependent on any particular distribution and extract desired functions from within data. This study aimed to compare the efficiency of neural network and maximum likelihood to classify land cover Using Landsat Satellite Images. Determine classes and samples to classify land cover Using field operations, topographic maps, aerial photographs and maps were made and using the above information four classes vegetation cover, building, water and outdoor were selected. After applying two algorithms, the neural network and maximum likelihood on the Landsat 8 satellite image with OLI sensors, land cover map of the arvand coastal area was prepared. Multi-layer perceptron network neural network structure consists of three input neurons, seven intermediate neurons, and four output neurons. For network training, a back propagation algorithm has been used. with Kappa coefficient, the accuracy of the classification methods was evaluated. Based on the results, Artificial neural network method with kappa coefficient of 0.92 in comparison to maximum probability algorithm with kappa coefficient of 0.79 has a better performance in providing land cover map of the arvand coastal area which is due to Neural network is nonparametric and nonlinear.