The use of convolution artificial neural networks for drones autonomous trajectory planning

Abstract

In article the problem of recognition of dynamic images is considered. The main tool of the solution of this problem offers use of artificial neural networks of deep learning or convolution networks. The review of characteristics of networks of this kind and their main architecture is given. The quantity of layers and dimensions of local networks at their use is proved in problems of autonomous navigation of drones on the basis of systems of technical sight. Methods of training of such systems of navigation as the problem of formation of the training data is not less difficult, than creation of networks of deep training and systems of autonomous navigation on their base are given. The review of program instruments of creation of networks of deep training is given. The computing resources consumed by them are estimated.