| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.12 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Este trabalho de projeto incide sobre a área geográfica do Parque Natural da Arrábida (PNArr), uma zona muito frequentada pela população, especialmente nas atividades de ciclismo/BTT e passeios pedestres. A grande afluência ao parque a juntar à falta de sinalização, levou a uma proliferação dos trilhos existentes, destruindo desta forma uma flora rica e única. O principal objetivo deste trabalho de projeto consiste na criação de uma rede viária que contemple todos os trilhos existentes, bem como a rede viária já existente e cartografada. Para se atingir o objetivo proposto, foram utilizados dados geográficos voluntários, técnicas de deteção remota e uma imagem do satélite Sentinel-2, nas bandas do azul, verde, vermelho e infravermelho próximo, todas com uma resolução espacial de 10m. Os dados geográficos voluntários representam os trilhos efetuados por cada utilizador. Foi efetuada uma máscara com esses mesmos trilhos, com o intuito de auxiliar na classificação da imagem de satélite, recorrendo a quatro algoritmos distintos: Máxima verossimilhança; SVM; Rede Neuronal; Árvore de decisão. Para cada algoritmo com exceção da árvore de decisão foram efetuadas duas classificações, uma sem o auxílio da máscara com os trilhos, outra com o auxílio desses dados. Após as classificações, foram realizadas as matrizes de confusão, de modo, a analisar os resultados obtidos, com o âmbito de verificar se realmente existia uma melhoria na classificação utilizando a máscara com os trilhos provenientes dos dados geográficos voluntários. Os resultados obtidos são satisfatórios, mas existe margem para melhorar, tanta na classificação das imagens, como na resolução espacial da imagem de satélite.
This project work focuses on the geographical area of Arrabida Natural Park, an area with a huge demand specially for mountain bike and hiking lovers. The large affluence to the park coupled with the lack of signage led to a proliferation of existing trails, thus destroying a unique and rich flora. The main goal of this project work is to create a road network that covers all the existing trails as well as the existing and mapped network. To achieve the proposed objective, voluntary geographic information, remote sensing techniques and a Sentinel-2 satellite imagery were used, four bands were used: Blue, green, red and near infrared, all with a 10m special resolution. The voluntary geographic information represents the trails made by each user. A mask was created with these same trails, in order to assist the classification of the satellite imagery, using four distinct algorithms: Maximum Likelihood; SVM; Neural Network; Decision Tree. For each algorithm except the decision tree, two classifications were made, one without aid of the mask with the trails, the other one with the aid of the mask. After the classifications, confusions matrices were performed in order to analyze the results obtained and if there really was an improvement in the classification where the mask with the trails were used. The results were reasonable, but like always there is room to improve, both on image classification and on the spatial resolution of the imagery.
This project work focuses on the geographical area of Arrabida Natural Park, an area with a huge demand specially for mountain bike and hiking lovers. The large affluence to the park coupled with the lack of signage led to a proliferation of existing trails, thus destroying a unique and rich flora. The main goal of this project work is to create a road network that covers all the existing trails as well as the existing and mapped network. To achieve the proposed objective, voluntary geographic information, remote sensing techniques and a Sentinel-2 satellite imagery were used, four bands were used: Blue, green, red and near infrared, all with a 10m special resolution. The voluntary geographic information represents the trails made by each user. A mask was created with these same trails, in order to assist the classification of the satellite imagery, using four distinct algorithms: Maximum Likelihood; SVM; Neural Network; Decision Tree. For each algorithm except the decision tree, two classifications were made, one without aid of the mask with the trails, the other one with the aid of the mask. After the classifications, confusions matrices were performed in order to analyze the results obtained and if there really was an improvement in the classification where the mask with the trails were used. The results were reasonable, but like always there is room to improve, both on image classification and on the spatial resolution of the imagery.
Descrição
Tese de mestrado, Sistemas de Informação Geográfica (Tecnologias e Aplicações), Universidade de Lisboa, Faculdade de Ciências, 2019
Palavras-chave
Dados Geográficos Voluntários Deteção Remota Sentinel-2 Algoritmos de classificação de imagem Parque Natural da Arrábida Teses de mestrado - 2019
