Development of a Web Service in the Form of an Interactive Map and a Classifier that Determines the Types of Buildings by Type Using Machine Learning Methods

Recent advancements in deep learning and neural networks have significantly increased their relevance for geospatial analysis tasks. A critical challenge in this domain involves evaluating urban infrastructure conditions, particularly through the functional classification of buildings (e.g., residential, commercial, governmental, and industrial). This study aims to develop an integrated software solution featuring: a data acquisition and preprocessing module for dataset compilation, a classifier module incorporating contemporary detection architectures, and an interactive web-based mapping interface for result visualization. The classifier module’s development involved rigorous evaluation of cutting-edge models, including YOLO variants and RF-DETR, with comprehensive metric analysis encompassing mean Average Precision (mAP), Recall, and other performance indicators. The implemented RF-DETR model achieved a mAP-50 score of 0.61, while the accompanying web service successfully demonstrates the classifiers’ operational capabilities. The complete work spans 114 pages and includes 68 figures, 6 tables, 126 references, and 5 appendices.