| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.71 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Atualmente, é possível visualizar uma abundância de pesquisadores em busca de estratégias aprimoradas para identificação de cores e texturas em imagens, seja por métodos analíticos, espectrais ou estruturais. Nesse contexto, a presente dissertação propõe uma análise teórica e aprofundada dos modelos de redes convolucionais, com foco nas camadas mais profundas da rede neural residual ResNet50. O principal objetivo é identificar as diferentes sensibilidades existentes sobre cada um dos blocos residuais presentes no modelo de rede neural ResNet50, utilizando diferentes técnicas de pré-processamento sobre o conjunto de imagens de Folhas de Oliveiras (saudáveis e doentes). Para atingir esse propósito, a pesquisa não apenas irá apresentar um estudo teórico das metodologias de Inteligência Artificial, mas também apresentará uma análise aplicando transformações sobre as diversas tonalidades presentes nas imagens. Neste contexto, será utilizado essa metodologia visando aprimorar o entendimento das características visuais das folhas de oliveira, proporcionando uma análise mais completa sobre a eficácia das camadas de convolução no processo de classificação. Além disso, a dissertação visa contribuir para o avanço da pesquisa ao explorar diferentes estratégias na análise de texturas, aprimorando a capacidade de diferenciação entre folhas saudáveis e doentes, com potenciais aplicações em monitoramento agrícola e diagnóstico precoce de doenças.
Currently, it is possible to see a large number of researchers looking for improved strategies for identifying colors and textures in images, whether through analytical, spectral or structural methods. In this context, this dissertation proposes an in-depth theoretical analysis of convolutional network models, focusing on the deepest layers of the ResNet50 residual neural network. The main objective is to identify, through a set of images, the textures present and categorize the layers according to their performance in classifying olive leaves (healthy and diseased). To achieve this purpose, the research will not only present a theoretical study of Artificial Intelligence methodologies, but will also present an analysis applying transformations to the different tones present in the images. In this context, this methodology will be used with the aim of improving the understanding of the visual characteristics of olive leaves, providing a more complete analysis of the effectiveness of convolution layers in the classification process. Furthermore, the dissertation seeks to contribute to the advancement of research by exploring different strategies in texture analysis, improving the ability to differentiate between healthy and diseased leaves, with potential applications in agricultural monitoring and early diagnosis of diseases.
Currently, it is possible to see a large number of researchers looking for improved strategies for identifying colors and textures in images, whether through analytical, spectral or structural methods. In this context, this dissertation proposes an in-depth theoretical analysis of convolutional network models, focusing on the deepest layers of the ResNet50 residual neural network. The main objective is to identify, through a set of images, the textures present and categorize the layers according to their performance in classifying olive leaves (healthy and diseased). To achieve this purpose, the research will not only present a theoretical study of Artificial Intelligence methodologies, but will also present an analysis applying transformations to the different tones present in the images. In this context, this methodology will be used with the aim of improving the understanding of the visual characteristics of olive leaves, providing a more complete analysis of the effectiveness of convolution layers in the classification process. Furthermore, the dissertation seeks to contribute to the advancement of research by exploring different strategies in texture analysis, improving the ability to differentiate between healthy and diseased leaves, with potential applications in agricultural monitoring and early diagnosis of diseases.
Description
Mestrado de dupla diplomação com a UTFPR - Universidade Tecnológica Federal do Paraná
Keywords
Redes convolucionais ResNet50 Folhas de oliveira Textura e classificação