STUDY ON THE APPLICABILITY OF DEEP LEARNING-BASED REGION EXTRACTION METHOD IN RIVER LANDSCAPE

In previous studies, we constructed models to predict psychological evaluations based on physical indicators in urban river landscapes. However, the conventional manual aggregation of physical indicators costs significant effort. Therefore, we applied semantic segmentation to automatically extract the area ratios of landscape elements using 48 sites in the vicinity of the Ota River, Hiroshima. We used the obtained elements area ratio data to predict psychological evaluations and mapped the predicted values for 255 locations. As a result, the deep learning-based extraction results of green area ratios and building area ratios were significantly close to the results of manual work.