COMPUTER INTELLIGENCE TECHNOLOGIES USAGE FOR IMAGES ANALYSIS WITH THE PURPOSE OF DEVELOPING AND DESIGNING A DECISION SUPPORT SYSTEM FOR MONITORING AND PREVENTION OF FOREST FIRES IN UKRAINE

Abstract

The article presents the results of research and development of a neural network model for monitoring natural emergencies, namely forest fires in Ukraine. A study of forest fires on the territory of Ukraine was conducted. Image analysis was carried out. A neural network model was built. It was pointed out that the development and design of the DSS for monitoring and preventing forest fires in Ukraine based on AI technologies and image analysis is a scientifically new task. The results of the study can be used for the development and implementation of more effective and reliable forest fire monitoring and prevention systems in Ukraine. A detailed analysis of the literature on the topic was carried out. Weaknesses were identified and tasks for research were set. The study was carried out using images from the open sources of the NASA Earth Observatory. Python libraries: Keras, TensorFlow, PyTorch were used to process and analyze satellite images. With the help of modeling methods, the system architecture is designed and usage options are shown. It was found that an effective approach to reducing the risk of natural disasters is the development and implementation of a modern decision support system (DSS) for each region of Ukraine. Such a DSS accumulates information about the technical, social and economic characteristics of the region in order to build an effective strategy for the prevention of natural disasters. The neural network learning model was given as follows. The dataset contains raw images where they are labeled as fire, no fire, or fire initiation. The image needs to be further processed before it can be used to train the model. The article describes the learning algorithm of a neural network. A formula was presented with a detailed explanation of the coefficients. Several different neural network models were built to identify the most effective learning option. As a result of convolutional neural network training, it was found that it is best to use all features for training. A neural network accuracy of 92% was achieved. In the future, the algorithm can be improved to improve accuracy. The results of the neural network were used for the SPPR for the early detection and prevention of forest fires in Ukraine.