Uso de Inteligência Artificial para escolha de sementes visando a Sustentabilidade na Agricultura.

Abstract

Brazil has an outstanding agriculture in the last forty years, with Artificial Intelligence drawing attention from researchers in general. However, there is a worldwide concern that values more the sustainable product, discarding from the purchase options when they do not meet these standards. Thus, the objective of this study was to use machine learning to enable the classification of soybean seeds with greater accuracy and efficiency, providing the rural producer with high quality seeds, reducing dependence on agrochemicals to achieve high productivity, in order to make the process safer and more environmentally sustainable. Data from tests performed in the seed analysis laboratories of the Federal University of Pelotas - RS and also from seed companies were used. The data were divided into two sets, C1 with three forms of classification, and C2 with fewer variables. The classifiers used were J48, Random Forest, CVR, lBk, MLP and NäiveBayes. To certify the accuracy of the algorithms, the following accuracy metrics were used: precision, recall, F-Measure, and ROC Area. The MLP and Random Forest algorithms obtained the best responses in the 2 classification forms of C1 and in C2, with more than 95% of assertiveness. All algorithms were able to classify the plots with hit rates higher than 75%, with some getting 99%. The MLP presented the best results, even when the data were classified from a single variable, which was later removed so as not to bias the classifier. With the results obtained it was possible to perform the evaluation and ranking with precision and accuracy, using artificial intelligence on soybean seed lots. The physical attributes were not used as main variables when the ranking was performed based on the second viability test by TZ. With C2, Vigor and Germination were shown to be paramount for accurate classification. However, some scientists still point out that for a good characterization and ranking of the batch it is necessary to perform other tests. The environmental impacts caused by excessive use of agrochemicals in the implementation of the crop can be reduced due to the elevation of physiological quality and, thus, make the production chain more sustainable by using information technology as a tool, aiming to improve the quality of life of people and pay attention to the environment, so that the entire ecosystem receives proper care.