Use of UAV in inventory of an old orchard – Case study Světlá

Abstract

The conventional inventory is time-consuming and expensive, however the remote sensing method is a useful tool to make orchard inventory faster and cheaper. The aim of this work is to carry out an old orchard inventory using a low-cost system boarded on a drone equipped only with a camera RGB (Red, Green and Blue) with an added NIR (Near-infrared) filter thus providing an automated approach for orchard managers. First, the position measurement of some individual trees was done and then, an Unmanned Aerial Vehicle (UAV) equipped with RGB (visible part of the spectrum) and NIR image camera was used to create the orthophoto images with 5 cm of spatial resolution. The proposed method includes Digital Surface Model (DSM) creation, individual tree location, tree species classification, and field verification of results. In this study, eight different species were identified and RGB, NIR spectral bands and Normalized Difference Vegetation Index (NDVI) were used as input features for classification. Several classification methods were compared in this study such as the Classification and Regression Trees (CART) method, which obtained a 79% of accuracy being Nut and Rose the most predicted species; moreover the producer's accuracy for eight species is ranging from 0.78-0.91. The results of Boosted Trees method, Random Forestand Supervised Classification technique showed an accuracy of 72%, 66% and 65%, respectively. In this research, the possibility to use drone technology to create an old orchard inventory was analyzed as well as the most accurate methodology to carry out tree species classification.