MACHINE LEARNING AND LANDSCAPE QUALITY. REPRESENTING VISUAL INFORMATION USING DEEP LEARNING-BASED IMAGE SEGMENTATION FROM STREET VIEW PHOTOS

The study is centered on the value of visual perception in the measurement of landscape quality. The research aims to define a digital methodological process and criterion for assessing the quality of a landscape, using along a road georeferenced image as open source big data. Artificial intelligence system, trained to recognize and quantify the elements present, processes these images associating area data, therefore converted them into values according to specific criteria. In each image, it evaluates positive or negative characteristics of the path, and the sum of all big data values generates an index (L-value). This approach is tested in different case studies, validating AI results with Collective Intelligence, using anonymous questionnaires. The proposed process transforms the perceptual data inherent in the photographs into information, from which it extrapolates a knowledge path synthesized in map, representation of perceived qualities of the landscape.