Automated Generation of Tree Crown Images Dataset for Neural Network Training using UAV-Acquired Imagery

Abstract: Tree mapping is an important type of information demanded in various fields of research and practice. However, it is an expen-sive and time-consuming process, which makes it difficult to monitor large areas. Therefore, automated methods are neces-sary to optimize tree mapping in forest areas. This article pro-poses a software tool based on a dataset of tree crown images acquired using unmanned aerial vehicles (UAV), necessary for training neural networks. The basis of the accumulated dataset consists of tree crowns in high spatial resolution (1–10 cm) RGB images. The current dataset contains coniferous tree crown images acquired from different heights. To prepare this dataset, a method for extracting the crowns of each tree was applied in a semi-automatic image annotation program. The proposed tool can be successfully used for training neural net-works and other machine learning approaches as well as com-puter vision techniques.

Keywords: image segmentation, crowns, tree, rgb, unmanned aerial vehicle (uav)