A carregar...
Publicação
Classificação Automática de Imagens Aéreas Multiespectrais para Agricultura de Precisão
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.22 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Ao longo dos anos, a utilização dos Veículos Aéreos não Tripulados (VANT) tem vindo a aumentar quer para fins recreativos, quer para fins académicos. As grandes vantagens económicas em relação aos seus directos competidores, satélites e aeronaves, permite que esta tecnologia seja amplamente utilizada em diversas áreas de estudo, como por exemplo a agricultura de precisão. As redes neuronais convolucionais (RNCs) já provaram ser excelentes em tarefas de classificação de imagens, e quando combinadas com os VANTs, estes permitem resolver alguns dos seus principais obstáculos relacionados com a necessidade de grandes conjuntos de dados ou grande capacidade computacional. Nesta dissertação é proposto um modelo de classificação de espécies de vegetação com recurso a imagens multiespectrais de alta resolução, utilizando redes neuronais convolucionais. O conjunto de dados utilizado foi adquiridos ao longo de 3 anos em diferentes ambientes e com diferentes culturas de vegetação. Inicialmente foram realizadas algumas operações de pré processamento dos dados tais como,alinhamento e remoção da distorção. Durante a implementação foram testados diversos parâmetros de treino da RNC, e utilizados algoritmos de Machine Learning (ML), nomeadamente Suport Vector Machines (SVMs), tendo sido obtidos resultados superiores a 80% de precisão.
Over the years, the use of unmanned aerial vehicles (UAVs) has been increasing for both recreational and academic purposes. The great economic advantages inrelation to its direct competitors, satellites and aircraft, allows this technology tobe widely used in several areas of study, such as precision agriculture. Convolutional neural networks have already proven to be excellent in image classification tasks, which when combined with UAVs, allow to solve some of the main obstacles related to the need for large data sets or large computational capacity. In this dissertation, a classification model of vegetation species using high resolution multispectral images is proposed, using convolutional neural networks. The data set used was acquired over 3 years in different environments with different vegetation cultures. Initially, some data pre-processing operations were performed,such as alignment and distortion removal. During the implementation, several training parameters of the RNC were tested, and Machine Learning (ML) algorithms were used, namely Support Vector Machines (SVMs), with results above 80 % accurate.
Over the years, the use of unmanned aerial vehicles (UAVs) has been increasing for both recreational and academic purposes. The great economic advantages inrelation to its direct competitors, satellites and aircraft, allows this technology tobe widely used in several areas of study, such as precision agriculture. Convolutional neural networks have already proven to be excellent in image classification tasks, which when combined with UAVs, allow to solve some of the main obstacles related to the need for large data sets or large computational capacity. In this dissertation, a classification model of vegetation species using high resolution multispectral images is proposed, using convolutional neural networks. The data set used was acquired over 3 years in different environments with different vegetation cultures. Initially, some data pre-processing operations were performed,such as alignment and distortion removal. During the implementation, several training parameters of the RNC were tested, and Machine Learning (ML) algorithms were used, namely Support Vector Machines (SVMs), with results above 80 % accurate.
Descrição
Palavras-chave
Agricultura de Precisão Veículos Aéreos não Tripulados Redes Neuronais Convolucionais Imagens Multiespectrais Classificação de Espécies de Vegetação Machine Learning