Intelligent method for building two-dimensional cards of the locality, based on the use of the carding of the classifi cator ensemble, trained with the application of the deep education method

Kuznetsov Vladimir Vyacheslavovich

Building an efficient multi-class classifier for creating two-dimensional maps of the terrain in situations with difficult weather conditions and shooting at night is a difficult task. Simple classifiers based on the use of support vector machines and combined into an ensemble of classifiers using the AdaBoost algorithm due to their susceptibility to the noise component are poorly suited in these situations. For these purposes, you can use a neural network that will do the classification into many classes and will be multi-layered. The technique of learning of such neural networks is called deep learning. A comparative study of simple classifiers based on SVM, combined into an ensemble of classifiers using the AdaBoost algorithm and classifiers trained using deep learning techniques, the Boltzmann stochastic machine, the conjugate gradient method and combined into an ensemble of classifiers using the AdaBoost algorithm was conducted. The result of the experiments is a significant reduction in the percentage of error and recognition time in situations with night shooting and with complex weather conditions for classifiers based on the use of deep learning.

Keywords:Deep Learning, AdaBoost, Restricted Boltzmann machine, support vector machine, conjugate-gradient method