Mostrar registro simples

dc.contributor.advisorPellico Netto, Sylvio, 1941-pt_BR
dc.contributor.authorHentz, Ângela Maria Klein, 1990-pt_BR
dc.contributor.otherDalla Corte, Ana Paula, 1980-pt_BR
dc.contributor.otherStrager, Michael P.pt_BR
dc.contributor.otherUniversidade Federal do Paraná. Setor de Ciências Agrárias. Programa de Pós-Graduação em Engenharia Florestalpt_BR
dc.date.accessioned2022-05-11T14:32:59Z
dc.date.available2022-05-11T14:32:59Z
dc.date.issued2018pt_BR
dc.identifier.urihttps://hdl.handle.net/1884/60697
dc.descriptionOrientador: Prof. Dr. Sylvio Péllico Nettopt_BR
dc.descriptionCoorientador: Profa. Dra. Ana Paula Dalla Corte, Prof. Dr. Michael P. Stragerpt_BR
dc.descriptionTese (doutorado) - Universidade Federal do Paraná, Setor de Ciências Agrárias, Programa de Pós-Graduação em Engenharia Florestal. Defesa : Curitiba, 10/12/2018pt_BR
dc.descriptionInclui referências: p.200-221pt_BR
dc.descriptionÁrea de concentração:pt_BR
dc.description.abstractResumo: Este trabalho teve como objetivo analisar o potencial do uso de dados obtidos com sensores passivos embarcados em Veículo Aéreo Não Tripulado (VANT) para obtenção de variáveis de inventário florestal. Para tanto, um talhão de Eucalyptus urograndis com 5 anos e dois de Pinus taeda com 7 e 16 anos, todos com aproximadamente 3 ha, foram inventariados em censo. Foram mensurados o dap (diâmetro à altura do peito - 1,3 m) e altura, e obteve-se o volume individual com equações polinomiais de 5° grau. Tomou-se a linha e posição de cada árvore na linha, e realizou-se a alocação de todas as árvores em um sistema de coordenadas com apoio de ortomosaicos. Voos com o VANT eBee e câmeras RGB, NIR e Multiespectral foram realizados, objetivando 5 cm de resolução e sobreposição de 80%. As imagens foram processadas com o software Pix4D, obtendo-se um ortomosaico, um modelo digital de superfície (MDS) e uma nuvem de pontos para cada câmera. As resoluções dos ortomosaicos variaram entre 5-7 cm para as câmeras RGB e NIR, e entre 10-14 cm para a câmera Multiespectral. Os MDS de cada câmera foram normalizados a partir de dados LiDAR, resultando em um modelo digital de altura de copa (CHM). Os produtos VANT foram utilizados para a detecção individual de árvores, executada a partir de uma ferramenta desenvolvida neste trabalho, chamada TreeDetect e outros três métodos de detecção. Todos os métodos de detecção mostraram-se promissores, porém os resultados da detecção variaram em função dos 3 talhões. A ferramenta TreeDetect apresentou os melhores resultados pela análise de qualidade em todos os talhões, principalmente quando foi aplicada com a banda espectral selecionada (NIR), se comparado ao seu uso com CHM. Para as estimativas das variáveis dendrométricas dap, altura e volume, as copas de todas as árvores alocadas manualmente foram delimitadas, utilizando-se um algoritmo disponível no pacote rLiDAR. Para cada copa foram extraídas informações derivadas dos produtos VANT, classificados como produtos estruturais (CHM e MDS), espectrais (bandas e índices), e de textura GLCM. Essas variáveis foram aplicadas em modelos de regressão múltipla com seleção stepwise, em sete combinações. As melhores equações estimativas resultaram em R2aj. e Syx% variando entre: 0,27-0,58 e 8,98-16,41% para dap, 0,34-0,52 e 5,94-13,87% para altura, e 0,37-0,59 e 18,57-36,99% para volume. O talhão Eucalyptus apresentou os melhores resultados e o Pinus com 16 anos os piores. A combinação de todas as variáveis apresentou os melhores ajustes em todas as situações. Todas as equações apresentaram resíduos tendenciosos, superestimando as árvores menores e subestimando as maiores, porém a inclusão de um fator de correção calculado em classes de tamanho, permitiu a redução das tendências e melhoria dos ajustes, que atingiram valores de R2aj. acima de 0,70 na maioria dos casos. A aplicação dos modelos estimativos de volume nas árvores detectadas pela ferramenta TreeDetect apresentou resultados muito bons, com erro máximo de 9,09% do volume total do talhão. Portanto, observou-se que dados de VANT podem ser aplicados com sucesso para a detecção de árvores individuais, e subsequente estimativa de variáveis dendrométricas. Palavras-chave: Drone. Árvore individual. Detecção. LiDAR. Regressão Múltipla.pt_BR
dc.description.abstractAbstract: This project had as main objective to evaluate the potential of using Unmanned Aerial Vehicle (UAV) data, and passive sensors, to obtain forest inventory variables. To accomplish this, one Eucalyptus urograndis stand with 5 years, and two Pinus taeda stands with 7 and 16 years, with approximately 3 ha each, were inventoried at census level. The diameter at breast height (DBH - 1.3 m) and total height of all trees were measured, and the individual volume was obtained using a fifth-degree polynomial equation. The line and position of each tree in the line was also recorded and, with this information each tree was plotted into a coordinate system over an orthomosaic. Flights were made with the UAV eBee and cameras RGB, NIR and Multispectral at an elevation to obtain 5 cm GSD and 80% overlap. The images were processed with Pix4D software, and an orthomosaic, a digital surface model (DSM), and a point cloud were obtained from each camera. The orthomosaic resolutions ranged from 5-7 cm for RGB and NIR cameras, and 10-14 cm for the Multispectral camera. Each camera DSM was normalized with LiDAR data, resulting in a canopy height model (CHM). The UAV products were applied to individual tree detection, performed using a tool called TreeDetect, developed for this project, and three other detection methods. Every detection method presented promising results, but the detection results were variable depending on the three stands. The TreeDetect tool presented the best results considering the quality assessment in all stands, especially when the tool was applied using the spectral band selected (NIR), in comparison with the TreeDetect with the CHM. The crowns of each plotted tree were delimited, using an algorithm available in the rLiDAR package, to predict the variables DBH, height and volume. From each crown, UAV derived metrics were computed, considering structural (CHM and DSM), spectral (bands and indexes) and GLCM textural products. The variables were applied into multiple regression models, with stepwise selection, in seven combinations. The developed equations resulted in R2aj. and Syx% ranging from 0.27-0.58 and 8.98- 16.41% for DBH, 0.34-0.52 and 5.94-13.87% for height, and 0.37-0.59 and 18.57- 36.99% for volume. The Eucalyptus stand presented the best results and the Pinus with 16 years presented the worse results. The combination of all variables provided the best model fit in all situations. All equations presented tendency in the residuals, overestimating the smallest trees and underestimating the largest, therefore the addition of a correction coefficient based in size classes resulted in reduction of those trends and in better fitting values for the equations, reaching R2aj. above 0.70 in most of the cases. Yet, the use of those estimative equations using the detected trees from the TreeDetect tool presented very good results, with maximum error of 9.09% of total stand volume. Considering the above evidences, it is visible that UAV data can be applied with success for individual tree detection and subsequent prediction of dendrometric variables. Keywords: Drone. Individual tree. Detection. LiDAR. Multiple regression.pt_BR
dc.format.extent1 recurso online : PDF.pt_BR
dc.format.mimetypeapplication/pdfpt_BR
dc.languagePortuguêspt_BR
dc.subjectLevantamentos florestaispt_BR
dc.subjectMapeamento florestalpt_BR
dc.subjectFotogrametria aereapt_BR
dc.subjectSensoriamento remotopt_BR
dc.subjectRecursos Florestais e Engenharia Florestalpt_BR
dc.subjectDronept_BR
dc.subjectLaserspt_BR
dc.titleObtenção de informações dendrométricas para inventário florestal automatizado por meio de veículo aéreo não tripulado (VANT)pt_BR
dc.typeTese Digitalpt_BR


Arquivos deste item

Thumbnail

Este item aparece na(s) seguinte(s) coleção(s)

Mostrar registro simples