Classificação de tábuas de madeira usando processamento de imagens digitais e aprendizado de máquina
Carregando...
Arquivos
Data
2014-12-02
Autores
Almeida, Osvaldo Cesar Pinheiro de [UNESP]
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O setor madeireiro no Brasil representa um forte componente da economia nacional, participando significativamente no Produto Interno Bruto (PIB) brasileiro. O segmento de madeira processada mecanicamente faz parte desse setor e tem a madeira de Pinus como principal espécie florestal plantada destinada ao seu processo produtivo. A madeira serrada desse gênero é usada em larga escala pela indústria madeireira e, devido à presença de defeitos, pode ser classificada em diferentes escalas de qualidade. A Associação Brasileira de Normas Técnicas (ABNT) define um padrão para a classificação visual de tábuas de madeira serrada de coníferas. Contudo, a graduação manual em um processo produtivo pode se tornar exaustivo, elevando a falha de classificação. Por essa razão, a automatização do processo de classificação de tábuas de madeira têm um papel importante na evolução tecnológica dos processos produtivos de serrarias. O objetivo desse trabalho foi o desenvolvimento de um sistema de classificação de tábuas de madeira de coníferas usando técnicas de processamento de imagens e aprendizado de máquinas. A partir de imagens de tábuas de madeira de Pinus foram realizados pré-processamentos, de maneira que as imagens fossem subdivididas em imagens menores. Em seguida foram extraídas as principais informações da imagem por meio de técnicas de análise de cor, usando o percentil das bandas de cor, e de textura, usando wavelet de Gabor. Essas informações foram usadas para criar modelos de classificação dos defeitos da tábua a partir do aprendizado de máquinas SVM – Support Vector Machine e redes neurais, onde cada imagem foi classificada como sendo madeira limpa (com ausência de defeitos) ou com nó ...
The Brazilian timber sector is a strong national economy component, participating significantly in the Brazilian Gross Domestic Product (GDP). The mechanically processed wood segment is part of this sector and has Pinus wood as the main tree species intended for their production process. The Pinus timber is extensively used by the industry and can be classified in different quality scales, depending on the presence of defects. The Brazilian Association of Technical Standards (ABNT) defines a standard for the visual grading of sawn wood of softwood. However, manual degree in a productive process can become exhausting, bringing the fault classification. For this reason, automation of the classification process of wooden boards plays an important role in the technological development in sawmill production process. The aim of this study was to develop a classification system of boards of softwood using techniques of image processing and machine learning. The boards images of wood planks preprocessing were performed so that the images were subdivided into smaller images. Then we extracted the main image information through color analysis techniques, using the percentiles of bands of color, and texture, using Gabor wavelet. This information was used to create a classification model of the board defects from the machine learning SVM - Support Vector Machine and neural networks, where each image was classified as clean wood (no defects) and with knot. The consolidation of defects identified on board served as the basis for creating models of quality grade board through the learning SVM, neural networks and the C5.0 algorithm. The machine learning SVM and neural network applied to 32x32 images showed an ...
The Brazilian timber sector is a strong national economy component, participating significantly in the Brazilian Gross Domestic Product (GDP). The mechanically processed wood segment is part of this sector and has Pinus wood as the main tree species intended for their production process. The Pinus timber is extensively used by the industry and can be classified in different quality scales, depending on the presence of defects. The Brazilian Association of Technical Standards (ABNT) defines a standard for the visual grading of sawn wood of softwood. However, manual degree in a productive process can become exhausting, bringing the fault classification. For this reason, automation of the classification process of wooden boards plays an important role in the technological development in sawmill production process. The aim of this study was to develop a classification system of boards of softwood using techniques of image processing and machine learning. The boards images of wood planks preprocessing were performed so that the images were subdivided into smaller images. Then we extracted the main image information through color analysis techniques, using the percentiles of bands of color, and texture, using Gabor wavelet. This information was used to create a classification model of the board defects from the machine learning SVM - Support Vector Machine and neural networks, where each image was classified as clean wood (no defects) and with knot. The consolidation of defects identified on board served as the basis for creating models of quality grade board through the learning SVM, neural networks and the C5.0 algorithm. The machine learning SVM and neural network applied to 32x32 images showed an ...
Descrição
Palavras-chave
Madeira serrada, Pinus elliottii, Pinus taeda, Madeira - Qualidade, Processamento de imagens, Inteligencia artificial, Wood
Como citar
ALMEIDA, Osvaldo Cesar Pinheiro de. Classificação de tábuas de madeira usando processamento de imagens digitais e aprendizado de máquina. 2014. xi, 107 f. Tese (doutorado) - Universidade Estadual Paulista Júlio de Mesquita Filho , Faculdade de Ciências Agronômicas de Botucatu, 2014.