UM PACOTE COMPUTACIONAL PARA DETERMINAÇÃO POR IMAGEM DE PARÂMETROS DE PULVERIZAÇÃO AGRÍCOLA SOBRE PAPÉIS HIDROSSENSÍVEIS

Abstract

To properly assess the quality of agricultural spraying, it is necessary to estimate parameters obtained through analysis of water-sensitive papers, such as droplet density and diameter. Software analysis is efficient, but generally limited by the handling of paper for scanning, high air humidity, luminosity and identification of the smallest droplets. With this work, a computational package in R language (free) was developed for the optimized analysis of water-sensitive papers, which allows the processing of images of the papers captured in the field. Comparative tests of the package's performance with the system available on the market were carried out. The test images were captured in an area cultivated with corn, uniformly distributing 38 water-sensitive papers on the leaves of the plants and then manually sprayed with a CO2 sprayer. The papers were also taken to the laboratory to capture images under controlled artificial lighting conditions. For field imaging, the package has an automatic water-sensitive paper detector. To estimate the spray parameters, tests were performed with different segmentation methods (Hue, ExB+Otsu, ExB+Adaptive Threshold and Artificial Neural Network). The neural network promoted values close to the software taken as a reference, and in the cross-validation it presented 50% accuracy against 10% of the reference, showing robustness to variations in the field and laboratory images. The package is available from: https://arsilva87.github.io/hydropaper