Inspecção visual em aplicações de viticultura de precisão
Ficheiros
Data
2009
Autores
Título da revista
ISSN da revista
Título do Volume
Editora
Resumo
Face aos exigentes níveis de qualidade dos produtos e à necessidade de rentabilização das explorações agrícolas, a utilização de sistemas robóticos autónomos tem sofrido um forte incremento nas últimas décadas. São cada vez mais os exemplos onde são reunidos esforços no sentido de se desenvolverem aplicações deste tipo na agricultura, incluindo a localização de frutos, apanha, controlo de qualidade, entre outros. Comparativamente às soluções tradicionais, recorrentes ao trabalho manual, estes sistemas modernos contribuem para uma melhoria na qualidade dos produtos e, ao mesmo tempo, para o aumento da rentabilidade, já que apresentam uma maior disponibilidade.
Apesar da existência de um conjunto alargado de sistemas robóticos, e de inspecção visual aplicados à agricultura, é ainda deficiente a oferta destes sistemas para a cultura da vinha e do vinho, especialmente na região de Trás-os-Montes e Alto Douro pelas características e condições naturais únicas desta região.
Neste trabalho foi desenvolvido um sistema de inspecção visual capaz de localizar as uvas em ambiente natural, tendo em conta a possível aplicação a qualquer exploração vitícola. O recurso à visão para localização de uvas dota este sistema com a capacidade de se adaptar ao ambiente em que se encontra.
O sistema de inspecção visual implementado consiste num computador equipado com o software Sherlock em conjunto com uma aplicação desenvolvida em Visual Basic, devendo ser-lhe fornecidas, para processamento e análise, imagens obtidas com uma máquina digital comum. Este sistema permite a aplicação em ambos os tipos de castas (brancas e tintas), detectando e localizando os cachos presentes nas imagens, ao mesmo tempo que é calculada a possível localização da haste do cacho que permite, no futuro, a conjugação com um sistema mecânico capaz de realizar a vindima de forma totalmente automática e sem intervenção humana.
Do conjunto de testes realizados destaca-se, desde já, a detecção correcta entre 81% e 91% para uvas brancas e entre 94% a 97% para uvas tintas, situando-se as falsas detecções entre os 19% e os 9% para uvas brancas e entre 6% e 3% para uvas tintas.
In response to the high demanding levels of the products quality and to the needs of agricultural exploitations profitability, the use of autonomous robotic systems have increased in the past decades. There are more and more examples where efforts are being combined to develop this kind of application in agriculture, including fruits location, harvesting, quality control, among others. Comparatively to the traditional solutions, appealing to manual work, these modern systems contribute to the improvement of the products quality and, at the same time, to the increase of the profitability, due to their higher availability. Despite the existence of a large group of robotic systems, and of machine vision, applied to agriculture, the offer of those systems to the vine and the wine culture is deficient, especially in the region of Trás-os-Montes e Alto Douro due to its unique characteristics and conditions. In this work, a system for machine vision capable of locating grapes in natural environment has been developed, in prevision to a possible application in any kind of vine exploitation. The use of vision to locate grapes gives this system the capability to adapt itself to the surrounding environment. The implemented system consists of a computer equipped with the Sherlock software and combined with an application developed in Visual Basic, where the images obtained by a common camera are provided to the system for processing and analysis. This system allows the identification of both types of castes (white and red), detecting and locating the grapes present in the images, and at the same time it also calculates the location of the bunch stem, allowing, in the future, an union with a mechanical system capable of grape harvesting totally automated and without human intervention. From the group of tests implemented, we can highlight a correct detection between 81% and 91% for the white grapes and between 94% and 97% for the red ones, with a rate of false detections standing between 19% and 9% for the white grapes and 6% and 3% for the red ones.
In response to the high demanding levels of the products quality and to the needs of agricultural exploitations profitability, the use of autonomous robotic systems have increased in the past decades. There are more and more examples where efforts are being combined to develop this kind of application in agriculture, including fruits location, harvesting, quality control, among others. Comparatively to the traditional solutions, appealing to manual work, these modern systems contribute to the improvement of the products quality and, at the same time, to the increase of the profitability, due to their higher availability. Despite the existence of a large group of robotic systems, and of machine vision, applied to agriculture, the offer of those systems to the vine and the wine culture is deficient, especially in the region of Trás-os-Montes e Alto Douro due to its unique characteristics and conditions. In this work, a system for machine vision capable of locating grapes in natural environment has been developed, in prevision to a possible application in any kind of vine exploitation. The use of vision to locate grapes gives this system the capability to adapt itself to the surrounding environment. The implemented system consists of a computer equipped with the Sherlock software and combined with an application developed in Visual Basic, where the images obtained by a common camera are provided to the system for processing and analysis. This system allows the identification of both types of castes (white and red), detecting and locating the grapes present in the images, and at the same time it also calculates the location of the bunch stem, allowing, in the future, an union with a mechanical system capable of grape harvesting totally automated and without human intervention. From the group of tests implemented, we can highlight a correct detection between 81% and 91% for the white grapes and between 94% and 97% for the red ones, with a rate of false detections standing between 19% and 9% for the white grapes and 6% and 3% for the red ones.
Descrição
Dissertação de Mestrado em Engenharia Electrotécnica e de Computadores
Palavras-chave
Inspecção visual , Detecção automática , Robótica móvel , Agricultura