Detecção e mapeamento da dispersão do Pinus elliottii Engelm. na planície litorânea do Rio Grande do Sul

Abstract

The monitoring of invasive alien species seeks information to understand their ability to interfere in ecosystems. Thus, this work aimed to generate a protocol to plan control and monitoring in areas with dispersal of Pinus elliottii Engelm. through the use of machine learning and high-resolution RGB images collected by drones. To this end, a protocol has been developed that provides a routine for the management of projects subject to compliance with conditions for the management and control of invasive exotics, with a focus on Pinus elliottii Engelm. The protocol proposes the use of geotechnical tools to obtain maps of the dispersal intensity of the target species. RGB images from drones were collected from two sample areas to apply methods for detecting Pinus elliottii plants with convolutional neural networks and interpolation of point data with the kernel density tool to generate zones of higher and lower dispersal density. The detection algorithm was trained and compared in terms of its evolution with data trained based on a pre-processed input of the species Citrus sinensis in the first sampling area and the insertion of samples of the target species itself in the second sampling area, and delivered in its best training an F-score of 94.1%. When estimating the kernel density, it was possible to obtain classes of dispersal coverage levels, providing a map with stratification of the invasion of the species Pinus elliottii. The management protocol for areas with dispersal has the potential to be used as a tool (System for Early Detection, Control and Monitoring of Invasive Exotic Species) to measure the effectiveness of efforts to control the dispersal of Pinus elliottii.