The effect of the activation functions on the classification accuracy of satellite image by artificial neural network

Abstract The effect of using two different activation functions on the classification accuracy using artificial neural network is presented. Where two activation functions are implemented to classify a satellite image which are the logistic and hyperbolic activation function. The effect of utilizing different number of hidden layer for fixed iteration number on classification accuracy and the required computation time are analyzed. The results showed that the accuracy of the results of the logistic activation function was not affected with the number of iterations compared to the hyperbolic activation function, whereas the hyperbolic activation function showed more stability than the logistic activation function with the number of hidden layers changing.