The paper is devoted to the problem of damage detection based on satellite image analysis, shows the main features of research on this problem and existing solutions, their advantages and disadvantages. The need to improve the methods of detecting damage on satellite images, to develop a distributed software architecture for analysing a large number of satellite images is identified. Particular attention is paid to the detection of damages as a result of hostilities, which is currently highly significant for Ukraine. As a result of the work, a neural network was created to determine the damage caused by hostilities on the territory of Ukraine, a machine learning method was developed that allowed training of the neural network model to determine the damage caused by hostilities on Ukrainian territory based on the analysis of satellite images, an approach to visualizing the damage detected by analysing satellite images was developed, and the effectiveness and scalability of the developed methods and software was studied. The effectiveness of horizontal scaling in comparison with vertical scaling is shown – a large number of inexpensive and low-performance computing nodes showed a better result of processing efficiency than a single node that used an expensive and powerful GPU.