| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 9.12 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Atualmente, a monitorização de pragas e doenças, bem como a identificação de espécies
em campos agrícolas, implica custos elevados de deslocação aos terrenos por parte de
profissionais especializados. Particularmente para pequenos produtores, isto pode ser
um entrave ao diagnóstico de problemas como doenças e pragas em culturas agrícolas,
dificultando o seu tratamento atempado.
Para tentar minimizar este problema, é benéfico fazer uso dos recursos tecnológicos
atuais, criando modelos de aprendizagem automática capazes de identificar problemas a
partir de fotografias tiradas no local. Além do mais, é importante garantir que se conseguem
partilhar esses dados entre os técnicos agrícolas, para que o seu trabalho conjunto possa
garantir a validade das classificações.
Por outro lado, o universo da aprendizagem automática tem vindo a crescer nos últimos
anos, estando a ser criados modelos cada vez mais avançados e eficazes para determinadas
áreas. Desta forma, pretende-se também garantir que a forma como a classificação é feita
possa evoluir ao longo do tempo, permitindo a profissionais da área de informática o
desenvolvimento ou adição de novos modelos de aprendizagem.
Assim, esta dissertação tem como objetivo o desenvolvimento, estudo e testagem de
uma plataforma desenvolvida usando os serviços da Google Cloud Platform para treinar,
visualizar e utilizar modelos de aprendizagem automática. Esta permite o armazenamento
de imagens em repositórios, de uma forma organizada que permite a anotação automática
dos dados. Por outro lado, permite o carregamento de templates de código por parte
de profissionais qualificados, bem como a sua instanciação com parâmetros relevantes.
Estes permitem desenvolver pipelines que serão responsáveis por toda uma cadeia de
procedimentos capazes de treinar modelos de aprendizagem automática. A plataforma
permite executar os templates instanciados de forma a desenvolver esses modelos, que
ficam disponíveis para uso de qualquer interveniente e para os quais se podem visualizar
métricas de avaliação relevantes.
Currently, the monitoring of pests and diseases, as well as the identification of species in agricultural fields, involves high costs of traveling to the land by specialized professionals. Particularly for small producers, this can be an obstacle to diagnosing problems such as diseases and pests in agricultural crops, making their timely treatment difficult. To try to minimize this problem, it is beneficial to make use of current technological resources, creating machine learning models capable of identifying problems from pho- tographs taken on site. Furthermore, it is important to ensure that this data can be shared between agricultural technicians, so that their joint work can guarantee the validity of the classifications. On the other hand, the world of machine learning has been growing in recent years, with increasingly advanced and effective models being created for certain areas. In this way, it is also intended to ensure that the way in which classification is carried out can evolve over time, allowing IT professionals to develop or add new learning models. Therefore, this dissertation aims to develop, study and test a platform developed using Google Cloud Platform services to train, visualize and use machine learning models. This allows the storage of images in repositories, in an organized way that allows automatic annotation of data. On the other hand, it allows the loading of code templates by qualified professionals, as well as their instantiation with relevant parameters. These allow the development of pipelines that will be responsible for an entire chain of procedures capable of training machine learning models. The platform allows you to run instantiated templates in order to develop these models, which are available for use by any stakeholder and for which relevant evaluation metrics can be viewed.
Currently, the monitoring of pests and diseases, as well as the identification of species in agricultural fields, involves high costs of traveling to the land by specialized professionals. Particularly for small producers, this can be an obstacle to diagnosing problems such as diseases and pests in agricultural crops, making their timely treatment difficult. To try to minimize this problem, it is beneficial to make use of current technological resources, creating machine learning models capable of identifying problems from pho- tographs taken on site. Furthermore, it is important to ensure that this data can be shared between agricultural technicians, so that their joint work can guarantee the validity of the classifications. On the other hand, the world of machine learning has been growing in recent years, with increasingly advanced and effective models being created for certain areas. In this way, it is also intended to ensure that the way in which classification is carried out can evolve over time, allowing IT professionals to develop or add new learning models. Therefore, this dissertation aims to develop, study and test a platform developed using Google Cloud Platform services to train, visualize and use machine learning models. This allows the storage of images in repositories, in an organized way that allows automatic annotation of data. On the other hand, it allows the loading of code templates by qualified professionals, as well as their instantiation with relevant parameters. These allow the development of pipelines that will be responsible for an entire chain of procedures capable of training machine learning models. The platform allows you to run instantiated templates in order to develop these models, which are available for use by any stakeholder and for which relevant evaluation metrics can be viewed.
Descrição
Palavras-chave
Aprendizagem Automática Computação em Cloud Banco de Dados Agricultura
