Estudos de casos de classificação de áreas cultivadas com café por meio de descritores de textura
Data
2016-10
Título da Revista
ISSN da Revista
Título de Volume
Editor
Editora UFLA
Resumo
O objetivo neste trabalho foi desenvolver um sistema para identificar áreas cultivadas com café utilizando Redes Neurais Artificiais (RNAs) tendo como variáveis de entrada os descritores de textura de Haralick. Utilizou-se o algoritmo de treinamento do tipo retro-propagação do erro (backpropagation) e o método de Levenberg-Marquardt. Foram realizados dois estudos de casos: no primeiro, as RNAs foram desenvolvidas para discriminar entre as classes café, mata, água, solo exposto, pastagem e área urbana; no segundo, as RNAs foram desenvolvidas para classificar as plantações de café de acordo com a idade e com a data de recepa. Para a avaliação do desempenho de classificação das RNAs empregou-se um mapa de referência de uso e ocupação do solo elaborado por meio do Sistema de Informações Geográficas. A concordância entre os mapas temáticos, classificados pela RNA, e o mapa de referência foi avaliada pelo coeficiente Kappa. Verificou-se que o coeficiente Kappa para discriminar a região cafeeira das outras classes temáticas foi de 0,652 no primeiro estudo de caso, desempenho considerado muito bom. Para classificar os plantios de café em função da idade e data de recepa o índice Kappa foi variável (0,675 a 0,4783), sendo considerado muito bom para a fazenda Itatiaia e razoável para a fazenda Pedra Redonda.
The objective of this work is to develop a system to identify areas cultivated with coffee using ANNs having as input variables descriptors Haralick. We used the training algorithm Back-propagation and Levenberg -Marquardt method. There were two cases of study: in the first step, the ANN was trained with representative samples of each class of interest (coffee, forest, water, bare soil, and urban area), thus verifying the potential to discriminate output classes; in the second step the objective was to classify the coffee plantations accordingly with the age. For the evaluation of the classification performance of ANNs was employed a reference map and land use through the Geographic Information System. The concordance between the thematic maps, classified by ANN, and the reference map was evaluated by Kappa index. It was verified that Kappa index for discriminating the coffee region of the other class of interest was 0,652 in the first case study, performance as very good. To classify the coffee plantations accordingly with the age, Kappa index was variable (0.675 to 0.4783), very good for Itatiaia farm and reasonable to Pedra Redonda farm.
The objective of this work is to develop a system to identify areas cultivated with coffee using ANNs having as input variables descriptors Haralick. We used the training algorithm Back-propagation and Levenberg -Marquardt method. There were two cases of study: in the first step, the ANN was trained with representative samples of each class of interest (coffee, forest, water, bare soil, and urban area), thus verifying the potential to discriminate output classes; in the second step the objective was to classify the coffee plantations accordingly with the age. For the evaluation of the classification performance of ANNs was employed a reference map and land use through the Geographic Information System. The concordance between the thematic maps, classified by ANN, and the reference map was evaluated by Kappa index. It was verified that Kappa index for discriminating the coffee region of the other class of interest was 0,652 in the first case study, performance as very good. To classify the coffee plantations accordingly with the age, Kappa index was variable (0.675 to 0.4783), very good for Itatiaia farm and reasonable to Pedra Redonda farm.
Descrição
Palavras-chave
Redes neurais artificiais, Sensoriamento remoto, Classificação supervisionada
Citação
SILVEIRA, L. S. et al. Estudos de casos de classificação de áreas cultivadas com café por meio de descritores de textura. Coffee Science, Lavras, v. 11, n. 4, p. 502-511, out./dez. 2016.