INDIVIDUAL TREE IDENTIFICATION IN URBAN AREAS FROM AERIAL IMAGES USING MASK R-CNN

Abstract

Trees in urban centers can bring many benefits to population health. Each city must be responsible for the planning and management of urban forestation, but it is difficult to check if all places are properly wooded. With technological advances, there is the possibility of developing tools to help these tasks in a computational way. In this paper, we preset a low cost tree identification method that uses a Mask R-CNN deep neural network. The experiments performed presented a correct rate of 91.39% in the identification of the trees from aerial photographs obtained by drone.