The work presents the methodology for the development and training of the Siamese neural network for comparing aerial photographs with terrain maps. The proposed approach is aimed at identifying stable and informative features in images, which allows to increase the accuracy and automation of the matching process. The presented method uses two identical networks that are trained in parallel, which ensures a reduction in the gap in characteristics between the compared images. The Siamese neural network efficiently handles images of varying quality and detail, making it ideal for comparing aerial photographs with terrain maps. The developed tool allows for quick analysis and comparison of aerial photographs with terrain maps with high efficiency and accuracy, which contributes to the expansion of the field of application in geoinformation research and engineering applications. Figs.: 3. Refs.: 11.