O uso de veículos aéreos não tripulados (VANT) na identificação do percevejo marrom em lavouras de soja usando técnicas de reconhecimento de padrões e aprendizado de máquinas

Abstract

The increasing evolution of technology is increasingly influencing people's lives. Due to the speed of information propagation, thus aiding a faster and more efficient decision making. One of the segments where technology is increasingly present is in agriculture, specifically to intensify increased production and make farming healthier, thereby reducing the use of insecticides and pesticides. Pest control in soybean crops, especially the brown stink bug, has always been of great concern to the farmer because of the difficulty of locating the source of infestation. One of the methods for locating the brown stink bug is the cloth, but it is necessary for the producer to walk the property to take samples, requiring a lot of time and influencing the decision making based on isolated parts of the property. The use of unmanned aerial vehicles (UAVs) has been increasing and gaining more importance, due to the benefits that it has been adding in decision making. The UAVs were initially developed for the military market and in recent years started to gain other markets, starting to be used in activities that were previously performed only by humans. In agriculture its use has enabled the producer to monitor difficult access places, analyzing its property as a whole. In addition, eliminating the need to tread plowing or even failure, bringing a gain in productivity and cost savings. The methodology presented is performed through the digital image processing, where they were captured in the soybean crop with the use of a UAV. After this step the image undergoes a computational treatment with the purpose of identifying points with possible infestations, it was also used the supervised machine learning, with the objective of identifying the brown bug in the soybean and its location through the geographical coordinates of the image. For validation of the classification process, two algorithms of pattern recognition, linear discriminant analysis and logistic regression were used. To validate the results, classification tests were performed with the original data and with the use of factorial analysis. After the tests, the algorithm that presented the highest assertiveness in the classification of the brown bug was the logistic regression with the use of the factorial analysis, reaching 97.22% accuracy, thus proving the use of the proposed methodology, allowing the producer to control insects in the soybean crop by making insecticides only in the areas with an infestation focus, thus reducing production costs and contributing to a healthier production.