versión impresa ISSN 0100-6916
CRUZ, Eleandro S. et al. Comparison of digital image classifiers for soil cover determination. Eng. Agríc. [online]. 2008, vol.28, n.2, pp. 237-244. ISSN 0100-6916. http://dx.doi.org/10.1590/S0100-69162008000200004.
In order to compare two image classifiers for soil cover estimation under both clear sky conditions and sky with clouds, it was evaluated the soil cover by grass and bean crops cultivated in different densities and tillage systems. The experiment was conducted in four soil loss experimental plots of 22.0 m by 3.5 m, in a Red-yellow argil soil. The four treatments consisted of sowing the velvet bean (Mucuna pruriens), sun hemp (Crotalaria juncea) and corn (Zea mays L.), all three following the contour lines, and corn following the slope line. The plot images were acquired from 15 to 85 days after sowing. The soil cover was estimated by off-shelf software (SIARCS) and a proposed algorithm based on green and red bad reflectance (SEROBIN). The highest soil cover was obtained in the sun hemp plot (85.8%), which was also obtained sooner (56 days after sowing). On the other hand, the lowest soil covers were obtained in both corn plots, following the contour lines and the slope line (38.6 and 35.2%, respectively). The overall classification accuracies were 0.96 for SIARCS and 0.92 for SEROBIN. There was no statistical difference between the classifiers using the Z test at the 5% significance level.
Palabras llave : soil erosion; soil cover; soil management; digital image processing.