| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 12.47 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Atualmente, com o crescimento da população, os pequenos agricultores das zonas rurais
transmontanas têm ficado cada vez mais desprotegidos, não possuindo meios para aumentar
a produtividade das suas explorações. Consequentemente, a fraca produtividade,
ligada à presença de carências nutritivas e doenças associadas, num ambiente cada vez
mais competitivo, poderá levar ao abandono dos terrenos podendo acentuar a desertificação
vivida em Portugal. Para tentar mitigar este problema esta dissertação pretende
desenvolver uma solução de baixo custo, automática e fiável, para combater a presença de
carências severas de nutrientes e doenças nas plantas, particularmente na oliveira, árvore
relevante em Portugal. Neste sentido, foi desenvolvida uma aplicação móvel que utiliza o
Transfer Learning para a classificação de doenças, visíveis na folha, obtidas com recurso à
fotografia. Deste modo, construiu-se um dataset de imagens contendo quatro classes, oliveira
saudável, carência de boro, carência de potássio e olho de pavão, que serviu de base
de treino ao modelo usado. Este foi treinado recorrendo à rede neuronal MobileNetV2,
possuindo uma precisão de 99%, com perdas de 1%. Os resultados em ambiente real
mostraram-se bastante bons pois a taxa de acerto é de 85%, sendo uma solução bastante
fiável.
Currently, with the growth of the population, small farmers in rural areas in Trás-os- Montes have become increasingly unprotected, not having the means to increase the productivity of their farms. Consequently, the low productivity, linked to the presence of nutritional deficiencies and associated diseases, in an increasingly competitive environment, may lead to the abandonment of land and may accentuate the desertification experienced in Portugal. To try to mitigate this problem, this dissertation intends to develop a low-cost, automatic and reliable solution to combat the presence of severe nutrient deficiencies and diseases in plants, particularly in the olive tree, a relevant tree in Portugal. In this sense, a mobile application was developed that uses Transfer Learning for the classification of diseases, visible on the leaf, obtained using photography. In this way, an image dataset was built containing four classes, healthy olive, boron deficiency, potassium deficiency, and peacock spot, which served as the training base for the model used. This was trained using the MobileNetV2 neuronal network, with an accuracy of 99%, with losses of 1%. The results in real environment proved to be quite good since the hit rate is 85%, being a very reliable solution.
Currently, with the growth of the population, small farmers in rural areas in Trás-os- Montes have become increasingly unprotected, not having the means to increase the productivity of their farms. Consequently, the low productivity, linked to the presence of nutritional deficiencies and associated diseases, in an increasingly competitive environment, may lead to the abandonment of land and may accentuate the desertification experienced in Portugal. To try to mitigate this problem, this dissertation intends to develop a low-cost, automatic and reliable solution to combat the presence of severe nutrient deficiencies and diseases in plants, particularly in the olive tree, a relevant tree in Portugal. In this sense, a mobile application was developed that uses Transfer Learning for the classification of diseases, visible on the leaf, obtained using photography. In this way, an image dataset was built containing four classes, healthy olive, boron deficiency, potassium deficiency, and peacock spot, which served as the training base for the model used. This was trained using the MobileNetV2 neuronal network, with an accuracy of 99%, with losses of 1%. The results in real environment proved to be quite good since the hit rate is 85%, being a very reliable solution.
Description
Keywords
Doenças em plantas Transfer learning App para classificação de imagens