Woody plants area estimation using ordinary satellite images and deep learning

Abstract

Modern solutions based on machine learning and map data are discussed in the paper. A convolutional neural network is proposed to use for urban spaces tree canopy evaluation. The developed method allowed to formulate a criterion for assessing the “green” infrastructure of a territory. The index, called Abin, is proposed to be used to estimate the area of woody plants. The process of preparing training datasets and carrying out search studies on the model of the neural network used are described. The results of training and testing of trained networks in arbitrary areas of the city are given. The analysis of the final results is carried out for the training errors revealed in the test cases, the prospects for the application of the trained models, the proposed method for estimating the “green” environmental conditions by the Abin index.