Optimized BP neural network classifier based on genetic algorithm for land cover classification using remotely-sensed data

A new Error Back Propagation algorithm based Genetic Algorithm was proposed in the article, and the key steps and framework were described in detail. The algorithm gives attention to two optimization algorithms, genetic algorithm and back propagation algorithm, which have the advantage during searching the infinitesimal point in local space and global space respectively, avoiding the risk of premature convergence while BP network was training; compared with BP algorithm, the end total mean square error of the network is more stable even if people redo the whole course several times. The data from China-Brazil Earth Resources Satellite were used to validate the algorithm; meanwhile the setting parameters and change processing of parameter were depicted carefully. Maximum likelihood classifier, back propagation neural network classifier were involved for a comparison purpose. The experiment results show that the new algorithm cannot only run with better efficiency, but also achieve the best classification accuracy.