Building recognition system based on deep learning

Abstract

Deep learning architectures based on convolutional neural networks (CNN) are very successful in image recognition tasks. These architectures use a cascade of convolution layers and activation functions. The setup of the number of layers and the number of neurons in each layer, the choice of activation functions and training optimization algorithm are very important. I present GPU implementation of CNN with feature extractors designed for building recognition, learned in a supervised way and achieve very good results.