Classificação de áreas cafeeiras em imagens multiespectrais através de redes neurais artificiais e extratores de textura

• Autor(a) principal: Silva, Alexsandro Cândido de Oliveira
• Data de Publicação: 2013
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/5192
• topic: Imagens de Satélite; Características de Textura; Redes Neurais Artificiais; Classificação Automática.

Remote Sensing images become increasingly propitious to the study of spatial and temporal variation of the land surface. However, just the spectral information of these images is not enough to classify objects that have similar pixels values like coffee crops and native forests. These two classes have a spectral pattern very close which makes difficult the automatic classification based on the spectral data. The application of Artificial Neural Network (ANN) in the classification of Remote Sensing images has shown a promising approach for discrimination classes. But the neural network doesn’t produce a satisfactory discrimination between coffee areas and forests using spectral data. In this work it was studied three regions with distinct features of the south of the state Minas Gerais where the coffee is an activity of paramount importance. This work proposes the extraction of spatial attributes like a texture of images of these regions to differentiate classes with similar spectral features. The use of texture attributes assists the classification process made by ANN. The methodology applied was the same for the three study areas. With the software Envi, it collected representing polygons each class of the land use for training and validation of the ANN. For extraction of the texture measures these polygons it used the software Texture and the neural network was implemented in the software Scilab. The automatic classifier built with one ANN and one Extractor Texture showed satisfactory and superior accuracy indices to other classifiers found literature, it shown a good procedure in the Remote Sensing data classification.