Unsupervised neural networks and data fusion of optical and SAR images for urban change detection

Since the early decades after the Second World War, characterized by an intense urbanization process, the control of the urban development has been lost in different European regions. Without a proper management, large urbanist areas have faced the stage of deindustrialization and the service sector’s development. Furthermore, the last two decades are marked by economic stagnation, demographic aging, foreign population’s rising, physical capital’s decline. The world’s population reached 7.7 billion in mid-2019, having added one billion people since 2007 and two billion since 1994. The global population will reach around 8.5 billion in 2030, 9.7 billion in 2050 and 10.9 billion in 2100. A system of regulatory constraints, implicit and explicit economic incentives and territorial organization has regulated the urban spatial development in the last two decades, making it impossible to keep the balance between new and old parts. Urban expansion has assumed the shape of suburbs: residential settlements disconnected from the focal points of the city, with low quality in architectures and urban design and without the possibility of performing elementary urban processes. This poor planning causes the rises of new districts that serve only as dormitories. Therefore, flow of people leave the suburbs in the day and come back in the night. To address the continuous suburb’s unchecked development, it is necessary to provide urban planning projects and to help this process. The identification of new constructions, especially in the early stage of the building process, can help local administration in their survey activities and urban planning. Knowledge about urban dynamics are usually derived from field surveys, aerial pictures or national censuses, which are costly and time-consuming techniques, consequently leading to delays in maps updates and to lacking of detailed information. However, today satellite remote sensing can be used to provide an objective and consistent view of urban areas and urban dynamics. However, because of their large spatial and spectral variability, mapping human settlements is still one of the most challenging remote sensing task. The purpose of this work is to give a new approach based on the fusion of different information contained in a single satellite product or in multiple satellites products to map urban changes and to observe suburb’s development. The algorithms developed can process data in a fast, automatic and accurate way with neural network techniques and unsupervised approach.