Smart agriculture framework from imagery based on representation learning

Nowadays, with the increasing need for food due to the high population, there are needed new alternatives in agriculture that avoid yield losses, while augmenting food security. Smart agriculture arises as a solution that gathers technology and agronomics, including aerial imagery, digital surface m...

Full description

Autores:: García Murillo, Daniel Guillermo

Tipo de recurso:

Fecha de publicación:: 2019

Institución:: Universidad Nacional de Colombia

Repositorio:: Universidad Nacional de Colombia

Idioma:: spa

Description
Summary:	Nowadays, with the increasing need for food due to the high population, there are needed new alternatives in agriculture that avoid yield losses, while augmenting food security. Smart agriculture arises as a solution that gathers technology and agronomics, including aerial imagery, digital surface models, meteorological stations, and machine learning techniques, to improve agricultural management, improving traditional techniques as human interpretation. Nevertheless, we identify two main problems: First, the lack of information about tree inventory due to not accurate maps. Second, the absence of accurate weed management derived from inefficient weed vs crop discrimination. In this work, we propose a new morphological transformation to deal with not accurate digital surface models, improving tree identification, and thus, generating tree inventories. Furthermore, we introduce a sparse feature extraction approach that remarkably separates overlapped classes. Finally, we propose a feature selection algorithm that extracts relevant features from a matrix projection. As a result, this work uses a smart agriculture framework from imagery based on representation learning to improve tree identification and weed/crop discrimination

Smart agriculture framework from imagery based on representation learning

Publicaciones similares