Industrial Automatic Control Systems and Controllers Annotation << Back The Application of Convolutional Neural Networks (CNN) to Search for Patterns of Various Nature in Data Series E.A. Isaev, V.A. Samodurov, D.V. Pervukhin, E.K. Filyugina The substantial progress in the development of modern machine vision technologies allows us to consider their applicability to traditional scientific and practical tasks, related to the analysis of various types of data and, specially, a time series analysis. In a number of industries data analysis is performed visually, since this method allows identifying patterns and simplify an information perception for the operator-analyst (a human). Thus, it can be assumed, that the implementation of a data analysis using machine vision algorithms can be at least as accurate as human. Even without taking into account the nature of the data. This approach can be widely used in practical spheres. Examples of industries where impartial and unemotional analytics are relevant and large volumes of real-time data are presented, are realtime data processing systems – trading systems, IoT systems, process control systems, etc. In this work we develop an approach to use a machine vision data processing method for classification of an astronomical data of the PRAO radio telescope. Currently, for processing time series received from the radio telescope, this observatory uses a proprietary system based on an algorithmic mathematical approach. This system allows to achieve an accuracy of automatic data classification of about 68 %. The developed and tuned CNN neural network makes it possible to obtain an accuracy of at least 94 % in production mode, while the classification time decreased from a few days to 0,5 seconds. This results surely shows improvements in productivity and quality of the classification process. Universal by design nature of this approach allows to consider it implementation to other various classes of similar problems. Keywords: CNN; machine vision systems; supervised learning; big data analysis. DOI: 10.25791/asu.12.2019.1069 Contacts: - Pp. 24-32.