Integration of Neural Network Models in Spatial Data Analysis Systems

The article presents a practical solution in the field of integration of neural network models in spatial data analysis systems. The article shows that the deep neural network model can to be differentiated by topology that defines availability in her structure layers certain type and connections between them and architecture neural networks defines format representation input data and view weekend signal models, parameters functioning and strategies learning. Component visual design neural network models allows visualize neural network models in form graph - diagrams with opportunity interactive online editing them topology and architecture. For efficient analysis spatial data implemented broadcast graphic representation models in performance on the language Python based on the application frame deep machine learning Keras, and also on the language MATLAB and package applied modules deep Learning toolbox. For solutions tasks broadcasts graphic representation neural network models in program the code her structure described oriented acyclic graph, and key algorithm transformations graphic representation models in instructions on the language programming general destination is topological sorting.

I. Introduction

The process of information support for making management decisions in the field of managing territorial systems (metageosystems) and responding to environmental, socio-economic threats should be based on the use of large arrays of spatial data containing information about the spatio-temporal structure of natural potential, population and economy. An effective platform solution for solving the designated problem is digital spatial data infrastructures. This class of information telecommunication systems combines the components of storage, analysis, visualization and distribution of spatial data and interacts with external objects represented by key actors, consumers and providers of spatial data.