SEMIAUTOMATIC DETECTION OF TREES IN IRRIGATED MANGO ORCHARD FROM DRONE IMAGES

Abstract

The monitoring of plant populations in agricultural areas is essential to follow the productivity, assisting in planning and decision making. Thus, our objective was to propose a protocol for remote detection of mango trees in the Low-Middle of the Sao Francisco River Valley, by using free software and plugins applied on aerial drone images. The study was conducted in three mango orchards. We used digital models extracted from orthomosaics created under three level of quality; then they were evaluated on the package QGIS with the plugins ‘Tree Density Calculator’ and ‘SAGA GIS’. The results were evaluated with the indices Precision, Recall and F1–Score. The precision index was higher for low-quality processing; while the recall index showed higher values under medium and high quality, indicating that the higher the quality of the processing, the greater is the chance of acquiring an efficient tree counting. The highest F1– Score values were observed for the Tree Density Calculator plugin with low processing resolution. We recommend using this protocol for the remote identification and counting of mango trees, in a semiautomatic methodology by using aerial images obtained using drones and free software and plugins.