Geoportals in Solving the Problem of Natural Hazards Monitoring

The article is devoted to the development of models and algorithms for analyzing the distribution of spectral channels based on pixel-by-pixel classification of metageosystems for the detection of natural and natural-technogenic processes using the example of fires, as well as the design and development of a geoportal system for monitoring and visualizing the situation with fires. The strengths and weaknesses of the machine learning models used, which were trained and fine-tuned based on multispectral images to solve the problem of fire localization, were revealed: support vector machines, K-neighbors Classifier, Random Forest Classifier, Gaussian NB, Logistic Regression. The advantages and disadvantages of these models are revealed, the optimal parameters of the models for solving this problem are experimentally selected. The geoportal system operates on the basis of the PostgreSQL database with the PostGIS extension, the Flask framework for building the system architecture, SQLLite storage for storing space images, as well as FastAPI and the object detector itself. Communication between services occurs through REST requests. Despite the fact that the project was developed to solve the problem of monitoring fires, the results obtained can be used in solving other problems related to the analysis and management of natural and natural-technogenic processes.