Дослідження методів спектрального аналізу зображень для детектування лісових пожеж з використанням технологій обчислювального інтелекту

Abstract

The article studies the methods of spectral analysis of images for detecting forest fires using computer intelligence technologies. For spectral analysis, satellite images of wildfires from the open source NASA Earth Observatory were used, which is the main source of satellite images and other scientific information related to climate and the environment provided by the National Aeronautics and Space Administration for a wide range of research.  The Volume Segmenter tool of the MATLAB package was used to create a three-dimensional model of the fire, with the help of which a three-dimensional zone of fire was selected. The color image of the fire was segmented, threshold values ​​for color channels were set based on different color spaces, and a binary segmentation mask was created for the color image, using the MATLAB Color Thresholder tool. The conjugate gradient method was used to create a neural network for analyzing the image of the spread of fire. This method was chosen because it does not require a lot of system memory, and also provides the ability to structure any data, constantly improving its properties. To train the model, several passes were carried out so that the neural network showed the most accurate results. The results of the study will be used to create a web service in real time, which will record hazardous natural phenomena, generate notifications and make recommendations for the prompt elimination of their consequences.