Potencial de Imagens Multiespectrais Sentinel 2 na Detecção de Nematoides na Cultura Cafeeira

In this paper we have evaluated the potential of using free multispectral images for identifying nematode infection in a coffee crop. First, we have adopted a study area within a known condition of the spatial distribution of Nematode infection (N. paranaenses ) on the coffee crop. Secondly, we analyzed the variation of spectral response of infected and non-infected coffee trees, on different bands of the Sentinel 2 satellite. As hypothesis we expected that healthy trees would respond differently of those non-healthy. Due this first analysis, we have detected bands whose variation on the reflectance could aid us on the image classification process. Thus, we detected those variations, while observing the Red, NIR, Red Edge 3-4 bands. In the next step, we have made the image classification (neural network) by applying different combinations of images sources, considering the results from the previous step, plus an image representing the NDVI index. The combination of Red, NIR, and NDVI bands as classification input gave us the best result when compared to the other combination. This combination allowed us to detect Nematode infection areas, and to perform the image classification with 97.91% of accuracy. Therefore, we have demonstrated the positive potential of using free images from the Sentinel 2 for identifying Nematode infections in coffee crops. This is a remarkable result, once we have produced an innovative, low-cost and confident solution.