Using unmanned aerial vehicle and machine learning algorithm to monitor leaf nitrogen in coffee

Abstract

Nitrogen is an essential element for coffee production. However, when fertilization do not consider the spatial variability of the agricultural parameters, it can generate economic losses, and environmental impacts. Thus, the monitoring of nitrogen is essential to the fertilizing management, and remote sensing based on unmanned aerial vehicles imagery has been evaluated for this task. This work aimed to analyze the potential of vegetation indices of the visible range, obtained with such vehicles, to monitor the nitrogen content of coffee plants in southern Minas Gerais, Brazil. Therefore, we performed leaf analysis using the Kjeldahl method, and we processed the images to produce the vegetation indices using Geographic Information Systems and photogrammetry software. Moreover, the images were classified using the Color Index of Vegetation and the Maximum Likelihood Classifier. As estimator tool, we created Random Forest models of classification and regression. We also evaluated the Pearson correlation coefficient between the nitrogen and the vegetation indices, and we performed the analysis of variance and the Tukey-Kramer test to assess whether there is a significant difference between the averages of these indices in relation to nitrogen levels. However, the models were not able to predict the nitrogen. The regression model obtained a R2 = 0.01. The classification model achieved an overall accuracy of 0.33 (33%), but it did not distinguish between the different levels of nitrogen. The correlation tests revealed that the vegetation indices are not correlated with the nitrogen, since the best index was the Green Leaf Index (R = 0.21). However, the image classification achieved a Kappa coefficient of 0.92, indicating that the tested index is efficient. Therefore, visible indices were not able to monitor the nitrogen in this case, but they should continue to be explored, since they could represent a less expensive alternative.