Selection of the optimal neural network of a Smart building system for object recognition

Abstract

Recently, artificial intelligence has captured a number of solutions that help us in everyday life. It will not surprise anyone now. It is in every modern smartphone and helps us to search, sort and process information faster. Today, smart home systems have taken a significant place among the developments in the field of information technology. The huge amount of data generated in them and the variety of formats of this data does not allow to create a universal mechanism for their productive processing. Therefore, the integration of neural networks, which is best suited for individual tasks, will provide us with high efficiency of smart home systems with minimal errors in decisions. Year by year, the interest in solving more complex tasks of object recognition is growing, due to automation needs for shaped communication processes in intelligent systems. Therefore, improving the implementation of the recognition of computer image systems is relevant. One of the promising directions for solving this problem is based on the use of artificial neural networks and neurocomputers as the most progressive in relation to the problems of classification of pattern recognition tasks. In our time, a large number of neural network architectures are proposed for application in the recognition of objects. The analysis of the proposed solutions shows that there is still no such model that would be the best among all the resulting performance parameters. Prospects for the improvement of architecture are seen in convolutional neural networks. The advantages of roller networks over multilayers are to use a common weight in the roller coasters, which means that for each pixel of the layer is used the same filter.