Use este identificador para citar ou linkar para este item: http://repositorio.ufla.br/jspui/handle/1/46267
Título: Aplicação de imagens RGB obtidas por VANT na cafeicultura de precisão
Título(s) alternativo(s): Application of RGB images obtained by UAV in precision agriculture
Autores: Ferraz, Gabriel Araújo e Silva
Silva, Fábio Moreira da
Guimarães, Rubens José
Silva, Flávio Castro da
Carvalho, Luis Marcelo Tavares de
Palavras-chave: Sensoriamento remoto
Aprendizagem de máquina
Café - Previsão de produtividade
Café - Agricultura de precisão
Machine learning
Remote sensing
Coffee - Yield estimate
Coffee - Precision agriculture
Unmanned aerial vehicle images
Data do documento: 14-Mai-2021
Editor: Universidade Federal de Lavras
Citação: BARBOSA, B. D. S. Aplicação de imagens RGB obtidas por VANT na cafeicultura de precisão. 2020. 81 p. Tese (Doutorado em Engenharia Agrícola) – Universidade Federal de Lavras, Lavras, 2021.
Resumo: Coffee is a commodity of significant importance in the Brazilian trade balance. To increase coffee productivity, new technologies such as soil sensors, development of cultivars that are more resistant to pests and water deficit, application at variable rates, and crop monitoring through images are used. Agricultural monitoring by images together as remote sensing techniques brings significant benefits to the producer by providing information in the desired time and with adequate resolution to identify and solve anomalies in the field, saving resources and ensuring the quality of production. In this scenario, the use of UAV (unmanned aerial vehicle) provides the farmer with accurate and timely monitoring. This work aimed to evaluate the use and application of a UAV and RGB images in the monitoring of a coffee crop, evaluating the behavior of Vegetation Indexes (IV) and the productivity prediction through products derived from RGB images. In article 1, the potential for practical application of UAV and IV RGB in monitoring a coffee crop during a production cycle was evaluated. Nine IV RGB were evaluated, using Pearson's correlation with the Leaf Area Index (LAI) as the metric for selecting IV. The results show that among the IV evaluated, the GLI and MPRI showed a higher correlation with the LAI (estimated through biophysical parameters derived from the images), but both were weak. During the phases of the coffee production cycle, both IVs illustrate the variability of the crop and soil. The increase of weeds was also observed in the marginal areas and between the planting lines. These results show that the use of a low-cost UAV and RGB camera has the potential for practical use for monitoring coffee in its production cycle, allowing the manager to decide for more assertive crop management quickly and simply. Given the potential for using UAV and RGB images described above, article 2 further investigates the potential for predicting coffee production using RGB images, using the variables: height, crown diameter (used to estimate the LAI) and brightness values of the RGB bands and multiple regression algorithms in 12 months, in which four different regression algorithms were used - Linear Support Vector Machines (SVM), Gradient Boosting Regression (GBR), Random Forest Regression (RFR) and Partial Minimum Square Regression (PLSR) - and a genetic algorithm - NEAT (Neuroevolution of augmenting topologies). The best result was obtained with a mean absolute percentage error (MAPE) of 31% for NEAT. Feature selections suggest that a December data set was the most important for the yield model, thus reducing the need for extensive data collection for all twelve months.
URI: http://repositorio.ufla.br/jspui/handle/1/46267
Aparece nas coleções:Engenharia Agrícola - Doutorado (Teses)

Arquivos associados a este item:
Arquivo Descrição TamanhoFormato 
TESE_Aplicação de imagens RGB obtidas por VANT na cafeicultura de precisão.pdf3,93 MBAdobe PDFVisualizar/Abrir


Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.