MACHINE LEARNING E PROCESSAMENTO DIGITAL DE IMAGENS UAV: UMA ABORDAGEM PARA ESTIMAR DISTRIBUIÇÃO LONGITUDINAL DE PLANTAS DE SOJA

Abstract

It is possible to achieve high productivity in soybean crops, by sowing with adequate and and uniform local distribution of the seeds.  To attain this, the use of technologies is important - for example the application of Digital Image Processing that allows treatment of collected images and improvement for human interpretation, and then the automatic analysis by the computer, based on pattern recognition classification. The objective of this research was to test Machine Learning methods to estimate the distribution of plants in the soybean planting line. The Random Forest model showed the best result, where the resulting accuracy was 65% on average, but the algorithm did not obtain a good result. It is possible to conclude that the difficulty of classifying the distances between soybean plants with the model used may be associated with the variables of image quality, overlapping of soybean plants and the precision of the model.