Abstract
In this paper a hybrid global optimization method GLPS is further investigated and applied for feed-forward neural networks supervised learning. The method is initially tested on several benchmark problems and subsequently employed for pattern recognition problem. The proposed technique is used for training Neural Networks (NN) that have to inspect and classify three types of cork tiles images. During the feature extraction phase, statistical textural characteristics are obtained from the tiles' images and then used for training several different NN architectures. Results from the testing phase are discussed and analysed, showing good generalization abilities of the trained NN. Finally, directions of future work are briefly stated.