Sistema de visão artificial para identificação do estado nutricional de plantas

Abstract

The evaluation of the nutritional status of corn plants is usually done through chemical analysis or by visual diagnosis of the plant leaves. Visual diagnosis is subject to misinterpretation as the lack of some nutrient in the plant generates a specific pattern of change in the leaf surface that depends on the degree on which the nutrient is absent on the plant. The difficulties present in this process and its importance in agriculture creates the necessity to search automated systems for the assessment of nutritional status of plants. Thus, this dissertation had as main objective the development of an artificial vision system to verify the possibility of identifying levels of macronutrients calcium, sulfur, magnesium, potassium and nitrogen in corn plants by analyzing the surface of the leaves using computer vision methods. This project performs a review of the literature of the state of the art methods for feature extraction of color, grayscale and colored texture used in image processing. The high similarity between the symptoms caused by deficiencies and low similarity between samples of the same deficiency motivated the development of new methods for extracting features that could provide the data needed for a correct separation between classes. The results showed that the system enables the prediction of nutritional deficiencies in an initial stage of plant growth using only texture of the leaf surface as a source of information.

Use of artificial vision techniques for diagnostic of nitrogen nutritional status in maize plants

“…According to Zúñiga (2012), color and texture are the main visual characteristics associated to maize nutrition. The following methods proposed by Zúñiga (2012) for AVS were tested in the present study for the texture analysis: Volumetric Fractal Dimension (VFD), Gabor Wavelet (GW) and VFD with canonical analysis (VDFCA).…”

“…The following methods proposed by Zúñiga (2012) for AVS were tested in the present study for the texture analysis: Volumetric Fractal Dimension (VFD), Gabor Wavelet (GW) and VFD with canonical analysis (VDFCA).…”

Métodos de pré-processamento de texturas para otimizar o reconhecimento de padrões

“…Bases de dados de imagens reais são tomadas por diferenças em iluminação, posição, ruídos, entre fatores outros que tornam a eficiente extração de características um trabalho árduo. Muitos trabalhos se atentam em criar diferentes estratégias para a tarefa utilizando a taxa acerto no reconhecimento de padrões como avaliador dos seus novos métodos descritores BRUNO, 2013;GALLOWAY, 1975;ZÚÑIGA;. No entanto, poucos se preocupam em tratar a imagem da textura antes de aplicar estes extratores de características, transformando-as e consequentemente, evidenciando características importantes da textura.…”

Desenvolvimento de modelos de redes neurais convolucionais para avaliação nutricional da >i<Brachiaria brizantha>/i<

“…Inicialmente foi realizada a segmentação da imagem. O fundo da imagem original foi removido, deixando apenas a folha a ser analisada com um fundo branco (ZÚÑIGA, 2012).…”

“…Esse limiar separa os pixel acima do limiar e abaixo dele. Com isso é possível identificar os contornos e extrair somente a folha a ser analisada e, com essa informação, criado o conjunto de dados das imagens (CHITYALA;PUDIPEDDI, 2014;HAMUDA et al, 2016;ROMUALDO et al, 2014;WILLIAM;OLIVEIRA, 2015;XU et al, 2011;ZÚÑIGA, 2012). A ferramenta "Adobe Photoshop CC 2019" foi usada para ajustes nas imagens, quando necessário.…”