Sistema de visão artificial para discriminação entre plantas daninhas e milho.

Abstract

The aim of the present work is to verify the effect of the number of the pixels values levels of the images (4, 8, 16 levels) and days after the emergency (22, 25 and 29 DAE), on a system for identification of weed and maize to be developed, using a green excess and monochromatic digital image and textural analyses technique. The classification was done by Bayes classificatory and the verification of the exactness was done by the global exactness matrix. The classification of the plants showed for the monochromatic images: a) the addition of the pixels levels values (4,8 and 16), for 29 DAE, resulted in exactness values of 87.6; 92 and 93%, respectively; b) the values of global exactness for DAE were 86.2; 83.4 and 92%, respectively for 22, 25 and 29 DAE. For green excess image (EXV), it was observed that: a) the values of global exactness observed were 47.2; 79.4 and 83.6%, respectively for the pixels levels 4, 8 and 16; b) The values obtained were 83.4; 80.2 and 83.6% respectively for 22, 25 and 29 DAE. The values of global exactness obtained, with EXV images, were inferior to that obtained with monochromatic image. The reduction of the levels of the pixels values represent reduction on the time processing of the images, which permits the use of the technique in real time. The monochromatic camera is low cost compared to the color camera.