Convolutional neural networks in automatic control systems: The state-of-the-art

Abstract

Convolutional neural networks are type of deep neural networks used for classification, identification, prediction and object detection. They are sutable for dealing with input data of various dimensions, such as signals, images and videos. Their importance is confirmed by the fact that they are used more than any other type of deep networks. This is the reason for constant development of new algorithms that improve existing models or creation od new models that accelerate or ameliorate learning process. They are utilized in a wide range of scientific and industrial fields due to their possibility of achieving high accuracy and simplicity of implementation. In this paper structure of convolutional networks is presented and, in particular, novelties in the study of convolutional layer are discussed, where different types of convolution are interpreted. Additionaly, special attention has been paid to the use of these networks in control systems in recent years, as a result of the occurrence of Industry 4.0. During scientific work analysis, convolutional networks application are divided according to the dimensionality of input data, that is, according to the dimensionality of networks and the tasks that they can solve.