ML models and neural networks for analyzing 3D data spatial planning tasks: Example of Khasansky urban district of the Russian Federation. (November 2022)

• Record Type: Journal Article
• Authors: Akylbekov, Olzhas; Said, Nidal Al; Martínez-García, Rebeca; Gura, Dmitry
• Is Part Of: Advances in engineering software. Volume 173(2022)
• Publication Date: 2022-11
• Subjects: Spatial planning -- Landscape structure -- GIS -- Machine learning -- Neural networks -- SDGs; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2022.103251
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: The map received allowed setting the designation purpose of the land plots and their rating in terms of suitability for construction works. Testing the neural network for terrain recognition showed the model's training error of up to 1.3%. The obtained results can be used for terrain recognition and expert evaluation of land plots. Abstract: The article addresses the issue of the Earth's surface surveillance. It was determined that artificial intelligence tools and machine learning methods are required to enhance the effectiveness of image recognition. Cores, Era 5, and ArcGIS cloud computing services (APIs) were installed to evaluate spatial data quality. The article presents the results of analyzing climate data received through applications developed by the authors. Raster terrain images of Slavyanka city located in Khasansky District of the Russian Federation were obtained, and corresponding graphs of temperature regimes were built. The root-mean-square deviation (RMSD) value calculated using machine learning techniques served as the basis for developing a neural network. It was later used to define the land features of the Khasansky city district to evaluate the possibility of construction works in this area. The map received allowed setting the designation purpose of the land plots and their rating in terms of suitability for construction works and defining areas not serviceable for any civil engineering activities. Complex climatic factors and obtained raster images of the terrain were considered. Testing the neural network for terrain recognition showed the model's training error of up to 1.3%. In terms of statistics, RSSD (2.5 ± 0.95), the Pearson's criterion, the Student's criterion, and the statistical significance of the results were estimated. The obtained results can be used for terrain recognition and expert evaluation of land plots.