Desenvolvimento e avaliação de um sistema para classificar grãos de culturas anuais por processamento de imagem digital

• Ano de defesa: 2012
• Autor(a) principal: Gallon, Rogério Antonio
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Mato Grosso; Brasil; Faculdade de Agronomia, Medicina Veterinária e Zootecnia (FAMEVZ); UFMT CUC - Cuiabá; Programa de Pós-Graduação em Agricultura Tropical
• Palavras-chave em Português: Classificação de grãos; Automação agrícola; Visão de máquina; CNPQ::CIENCIAS AGRARIAS; Classification of grain; Agricultural automation; Machine vision
• Link de acesso: http://ri.ufmt.br/handle/1/1330

The objective of this study was to develop a system to classify grains using digital image processing, to develop and evaluate the system. We used grains of annual agricultural crops, corn (Zea mays L.), soybean (Glycine max L.), rice (Oryza sativa L.), cotton (Gossypium hirsutum L.), sunflower (Helianthus annus), bean (Phaseolus vulgaris L.), produced in the State of Mato Grosso. The work was executed at the Laboratory of Remote Sensing and Geoinformation (Sergeo) int the Federal University of Mato Grosso. We used a PC-type computer, video camera and analytical box to position the equipment needed for the collection of images (lights, support for the video camera and a place to accommodate the grains of the six species). To illuminate the target, three electronic lamps were disposed below the grain sample. The construction of the analytical box to position the lamps was useful in recording and processing the images. A computer routine capable obtaining the image, processing the information and providing the results was developed using the Matlab® and the specific module for image processing. The images were obtained using a ‘webcam’ type video camera kept in a fixed position with the same distance from the target throughout the experiment. After obtaining the images, we proceeded to the geometric calibration. The captured image was corrected radiometricaly using filters to eliminate noise. The first color composite image (RGB) were converted to binary. For the binarization method was used for optimal Otsu thresholding. Then, the extracted data required for calculation of the geometrical measurements of area, major axis, minor axis and eccentricity. The measures of axis obtained by digital processing were compared with a digital caliper and the coefficient of correlation (Pearson) was determined as r = 0.98, 0.98, 0.99, 0.99, 0.97 respectively for corn, sunflower, beans, soybeans and cotton. For species identification we used a classifier that used values belonging to a range of minimum and maximum for each culture. These values were previously identified for the four traits and fixed in the routine of the program. The total success of the program in the identification of individual species, compared with visual assessment for the soybean, rice and sunflower was 11 100% and cotton, beans and maize, 98%, 89.4% and 90.4%, respectively. The accuracy of the program for evaluation of the six species, using the confusion matrix was 86%. For a better usage of the image classifier, a graphical interface was developed and an executable program was created. The software has proved useful in the automatic identification of annual grain crops. The advantages of using digital processing in the classification of grains is the speed in obtaining results, the high accuracy of results, reducing costs and permanent record of the results.