ROOFTOP SURFACES AND TYPES IDENTIFICATION USING VHR SATELLITE IMAGES FOR ELECTRIFICATION POTENTIEL IN AFRICA

Automatic extraction and recognition of roof structures, surfaces and types from remotely sensed data is one of the most notable challenges for installing urban photovoltaics panels which is of great importance for policymakers planning and investing in distributed energy infrastructures and electrification. DL techniques applied on VHR satellite images, allows to overcome the limitations of roofs surveys in providing this mapping at large scales. This paper proposes a DL based approach for mapping the location and identifying the type of roof surfaces starting from VHR images. The originality of this work is the automatization of roof types classification (metal, concrete, wood, etc.) independently from the country style (Africa, Europe, etc.). Indeed, to classify roof types using DL techniques you need a dataset covering all roof types. Data collection and labelling is usually done manually which is very time consuming. The proposed approach constructs a new dataset of roof types adapted for every region of interest using DL features extraction and clustering. It overcomes the appearance of new roof types especially in developing countries. The experimental results show that proposed approach can effectively and accurately detect and recognize roof types and has competitive performance. .