Detecção de manchas em lavoura de soja ocasionadas por patógenos do solo, com base em dados espectrais no estado do Paraná

Abstract

The identification of pathogens in agricultural production areas is a tool that helps in localized control and its dissemination, reducing the cost of control and in a less aggressive way to the environment. The objective of this work was to evaluate the use of images obtained by RPA to identify spots in soybean (Glycine max) crops caused by soil borne pathogens in two harvests in the region of Ponta Grossa and one harvest in Wenceslau Braz, State of Paraná. Within the two study sites, an area of 4 hectares was selected to carry out the data collection in the field, where they were divided into 80 sample grids, measuring 500m2 each sample grid with the dimensioning of 20 x 25 m. The RPA used was Inspire 2 from DJI with the Sentera Double 4K camera (multispectral). The identification of symptoms of stain in soybean crops, caused by the soil borne pathogens, identification of pathogens, plant height, population, and yield in kg ha-1 were evaluated in the field in Weka and SynthesisFS software. The verification of the best algorithm was based on the highest correlation coefficient index in relation to the attribute of plants with and without pathogens, using the Shapiro-Wilk normality test to verify if the data that were worked were parametric or not, in according to results of the analyses. For the non-parametric values, the Wilcoxon test was used to verify the difference between the populations of the quadrants with and without soil pathogens and, for the parametric data, the Student's T test was used. The pathogens found in the areas during the two seasons, in the two study sites, were Fusarium spp, Macrophomina phaseolina and Phomopsis spp. Which affected productivity. The algorithms showed the importance of using the reflectance values of the bands in the two study sites with emphasis on the NDVI and NDRE indices. It was possible to

generate a classification model with the J48 algorithm, based on trees, with cross- validation of 5 folds with hit rates above 80% in both places, considered efficient to

identify the ruffles caused by soil borne pathogens in the soybean crop.