Land-use Mapping of Valencia City Area from Aerial Images and LiDAR Data

摘要

Land-use classification of urban environments is usually limited by the number and complexity of the considered classes and the capability of the selected methodology for the efficient discrimination of these classes. Thus, this paper analyses and assesses the performance of a contextual object-based classification methodology in urban environments considering a comprehensive land-use legend, including several complex urban land-uses –historical buildings, urban buildings, open urban buildings, semi-detached houses, detached houses, industrial/warehouse buildings, religious buildings, commercial buildings, public buildings, gardens and parks–, and agricultural classes –arable lands, citrus orchards, irrigated crops, carob-trees orchards, rice crops, forest, greenhouses–. Object-based approach was achieved by using cadastral plot limits for object definition. An exhaustive set of object-based descriptive features were computed informing about the spectral, texture, structural, geometrical, three-dimensional and contextual properties. Classification was performed by means of decision trees algorithm combined with boosting multi-classifier. The overall accuracy reached classifying the urban area of Valencia reached 84.8%, which is a significantly high value when considering a large number of complex urban classes.