The use of UAVs in monitoring yellow sigatoka in banana

Abstract

Monitoring pests and diseases is an extremely important activity for increasing productivity in agriculture. In this scenario, remote sensing, coupled with techniques of machine learning, offer new prospects for monitoring and identifying characteristic specific patterns, such as manifestations of diseases, pests, and water and nutritional stress. The aim was to use high spatial resolution aerial images to monitor the extent of an attack of yellow sigatoka in a banana crop, following the basic assumptions of identification, classification, quantification and prediction of phenotypic factors. Monthly flights were carried out on a commercial banana plantation using an unmanned aerial vehicle, equipped with a 16-megapixel RGB camera (GSD of 0.016781 m pixel−1). Five classification algorithms were used to identify and quantify the disease while field evaluations were also made following traditional methodology. The results showed that, for September 2017, the Support Vector Machine algorithm achieved the best performance (99.28% overall accuracy and 97.13 Kappa Index), followed by the Artificial Neural Network and Minimum Distance algorithms. In quantifying the disease, the SVM algorithm was more effective than other algorithms compared to the conventional methodology used to estimate the extent of yellow sigatoka, demonstrating that the tools used for monitoring leaf spots can be handled by remote sensing, machine learning and high spatial-resolution RGB images.