Classificação supervisionada de imagens aéreas com drones para identificação de plantas daninhas

Abstract

The use of remotely piloted aircraft, or drones, in agriculture is considered one of the most recent and innovative technologies being used. The present work aimed to analyze the behavior of data from RGB images obtained by drone to estimate weed areas in a property in the region of Caarapó-MS through vegetation indices and classification. The images were collected and processed in a free software WEBODM to make the orthomosaic. After this step, they were forwarded for processing in which the following analyses were performed in the QGIS software: calculation of the GLI indexes (Green leaf index); pGRVI (Passive Green Red Vegetation Index); and RI (Normalized Difference Red/Green Redness Index) as well as supervised analysis with the Dzetsaka complement in the Randon Forest, Support Vector Machines and K-Nearest Neighbors algorithms, to define the area of vegetation, straw, and exposed soil. Subsequently, the results obtained from the classification were compared by using the Kappa index, in wich demonstrate that using the KNN as an witness the classifications with higher similarity levels was between the RF and KNN, with 98,66 % hit rate between then.