Desenvolvimento de plataforma de manejo de pastagens através do uso da visão computacional e inteligência artificial

Abstract

Technological improvement has greatly contributed to the development of rural activities, both agricultural and livestock. In Brazil, extensive animal husbandry is predominant, a feature that highlights the importance of pasture management to optimize production in order to meet domestic and global demand, while maintaining sustainability. In this scenario, this work aimed to use the drone as a technological tool to, in a practical and efficient way, assist in the measurement of pasture, the food base of animal production. To analyze the feasibility of using drones for pasture management, pasture areas of Panicum were chosen at the Experimental Farm Capim Branco of the Federal University of Uberlândia, randomly, with different heights, so that forage mass were collected manually in these areas and images with the drone, thus allowing to compare the data obtained by the two methods. Having determined the paddocks to be evaluated, with the aid of a graduated ruler, 100 points of clump height were measured. The values were listed in ascending order to obtain five averages of the 100 points collected. Once the value of each average was determined, 4 points were found with this height value, for each average obtained, totaling twenty points per paddock. These points were identified with a 1m² PVC frame, 1.5 meters above the ground. Using the DJI Spark drone, 3 images were obtained for each point, 4 points for each average height value determined, for the 5 averages, totaling 60 images taken by the drone for each paddock. After collecting the images, the forage that was inside the PVC frame was removed from all points, close to the ground, stored in polyethylene bags, and weighed to obtain the green matter. The forage mass collected at the last point of each average was used as sub-green, which corresponds to a sample of all the material collected and which was homogenized, and then taken to the oven for 48 hours at 65ºC. After drying, the samples were weighed and through the relationship between dry mass and green mass of the subsample, the dry matter was calculated, and this value was correlated to the other points of the average. Thus, five sub-greens were made per paddock, one corresponding to each average. Based on all the data presented here, it was possible to conclude that the analysis in the software of at least two images is representative, with the estimate of forage mass similar to the average found by manual collection.