Evolving Neural Network Using Genetic Simulated Annealing Algorithms for Multi-spectral Image Classification

摘要

In this paper, an evolving neural network classifier using genetic simulated annealing algorithms (GSA) and its application to multi-spectral image classification is investigated. By means of GSA, the classifier presented is available to automatically evolve the appropriate architecture of neural network and find a near-optimal set of connection weights globally. Then, with Back-Propagation (BP) algorithm, the conformable connection weights for multi-spectral image classification can be found. The GSA-BP classifier, which is derived from hybrid algorithm mentioned above, is demonstrated on SPOT multi-spectral image data effectively. The simulation results demonstrated that GSA-BP classifier possesses better performance on multi-spectral image classification. Its overall accuracy is improved by 4%～6% than conventional classifiers.