THE VALUE OF DEEP LEARNING FOR LANDSCAPE REPRESENTATION COMPARISON BETWEEN SEGMENTATION IMAGES MAPS AND GIS
- Department of Civil and Environmental Engineering, University of Perugia, 06125 Perugia, Italy
Keywords: Image segmentation, Big data, Perception, Automation, Deep learning
Abstract. Landscape refers to the qualities of a place, the result of a structural, territorial and environmental component, and the attribution of meanings, which is certainly the fundamental issue of the interpretative process. Percepire etymologically derives from "per", which means "by means of, through", and "capere", which translates as "to take", "to collect" (information, sensory data), "to learn". Since images are derived from the territory, it is of first interest to propose a comparison between representations derived from automated processes on photographs and the synthetic data interpreting the territory inherent in the plans developed with GIS in order to obtain a more precise perceptual analysis. The emergence of new tools for the processing and reproduction of data offers new opportunities for the knowledge and representation of the landscape, in architectural and urban contexts, and the integrative support that these processes can bring to the representation of the qualities of a place have to be reinterpreted in a Spatial Information Dataset in order to make synthetic and intelligible information. Identifying specific themes by questioning these data through criteria and placing at the centre the capacity of the digital environment in its mathematisation to compare data, transforming them into information, in an automated process is aimed at the exploitation of Big Data and the full replicability of the procedure. In this way, it is possible to enter into the analysis of the quality of space, of that notion of landscape concieved as "that part of the territory perceived by the population that lives it".