Scholarly record

A MACHINE LEARNING APPROACH FOR URBAN AREA MAPPING

Ana-Ioana Breaban, Valeria-Ersilia Oniga, Florian Stătescu

Abstract

The accelerated evolution of the anthropogenic land use together with the involved dynamics of the land use changes has deepen its environmental impact and increased the population demands. The usage of satellite images became in the last decades a complementary method for obtaining structural information on the urban land use due to major developments in domains such as remote sensing, geographic information systems and photogrammetry. Accurate information on urban built-up area is essential for multiple key areas like preparation of master plan and detailed development plans, improvement of urban infrastructure, provision of basic amenities, solid waste management system, renewable energy infrastructure, planning for smart cities, etc. Among various machine learning methods, random forest model is primary implemented due to its efficiency in handling diverse types of data. Therefore, this research investigates the performance of random forest algorithm in satellite image classification of urban areas by using spectral indices. The spectral confusion is characteristic for the heterogeneous urban land cover classes making less satisfactory to discriminate between classes sharing similar spectral characteristics. 12 indices were taking into analysis based on their applicability, targeting three specific land types: built-up areas, vegetation, and water. The built-up indices increase the contrast between bare land and built-up areas but it decreases the contrast between water bodies and other land covers. This drawback is overcome by combining it with specific indices for vegetation and water.

Publication Impact Profile

• Captures
• Mendeley - Readers: 12

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2021/2.1/s11.78

Publication details

Back to publication list