One of the features of the development of modern infrastructure of the state is the rapid growth of cities. The growth of urban areas requires the improvement of management systems. To implement effective management of a modern city, it is necessary to obtain timely data, which is ensured by conducting appropriate monitoring. One of the main requirements for such monitoring is the accuracy of the results obtained. At the same time, the most important thing in monitoring is to identify changes in the urban environment and analyze the causes of their occurrence.
An effective approach to improving the accuracy of the results of monitoring the urban environment is an approach based on the development of knowledge-oriented decision support systems. It is proposed to consider deep neural networks and hybrid (fuzzy neural network) models as the basic mathematical apparatus for formalizing knowledge of this class. Traditionally, the technology for creating knowledge-oriented decision support systems is based on classical technologies for the development of intelligent systems. In this case, the prototyping method is used as the main approach, which is based on the implementation of a certain sequence of stages in the development of decision support systems with an intermediate representation, refinement and formation of the corresponding prototype. At the same time, the stages of classical technologies for creating intelligent systems are characterized by a rather arbitrary, not formalized definition of the boundaries of the implementation of these stages and the transitions between them. All this significantly complicates ensuring the manufacturability of creating decision support systems for monitoring the urban environment from an industrial point of view. That is, the issues of an integrated approach to the development of decision support systems at an industrial level for recognizing buildings on digital space and aerial photographs using deep neural networks and hybrid (fuzzy neural network) models are insufficiently studied.
Thus, in the subject area, there is a contradiction, which consists, on the one hand, in the need to develop knowledge of oriented decision support systems for recognizing buildings in digital space and aerial photographs using deep neural networks and fuzzy logic, on the other hand, in the possibilities existing technologies for the development of such a class of systems.
To solve this contradiction in the dissertation work, an urgent scientific problem was formulated and solved - the construction of information technology for the development of a decision support system for recognizing buildings on space and aerial photographs to increase the efficiency of automated monitoring of the urban environment.
In the course of the dissertation work, the model of the convolutional neural network for the segmentation of objects on digital images was improved, in which, unlike the known ones, a pretrained convolutional neural network with a deeper architecture is used as a neural network narrowing block for the feature extraction subnetwork, and as a classifier it is used modified Wang-Mendel neural network, which implements operations on interval fuzzy sets of the second type. This improvement allows to provide more accurate segmentation of objects in digital images. The method of formalizing knowledge about the semantic segmentation of buildings on space and aerial photographs was also further developed, in which, in contrast to the known ones, the developed formalization apparatus is built on the basis of using an improved convolutional neural network model for object segmentation and a modified teaching transfer method using several bottlenecks (intermediate connections between the narrowing and widening blocks of the advanced convolutional neural network model). This improves the quality and shortens the training time for the fuzzy convolutional neural network model. For the first time, an information technology has been built for the development of a decision support system for recognizing buildings on space and aerial photographs during automated monitoring of the urban environment, which, based on functional modeling, formally represents the process of developing a decision support system using a fuzzy convolutional neural network model. The presented technology makes it possible to ensure unification and standardization of the process of developing a decision support system of the appropriate class.
Developed and improved technologies, models and methods have qualitatively new properties and allow solving the scientific problem of building information technology for developing a decision support system for recognizing buildings on space and aerial photographs to increase the efficiency of automated monitoring of the urban environment. This allows you to get a gain in accuracy and completeness when solving the segmentation problem on average up to 3%.
