Resposta hiperespectral da cultura da soja em função da severidade da mancha alvo (Corynespora cassiicola)

Abstract

In the current context of agriculture, productive efficiency is fundamental for farmers, but diseases, such as target spot, continue to harm soybean productivity. Remote sensing, especially hyperspectral sensing, can detect these diseases, but it has disadvantages such as cost and complexity. The objectives of this work were: to identify the input variable (Bands, Vegetation Indices and Reflectance) most appropriate for the metrics worked on (Correct Classification, Kappa and F-score) and to identify whether there is a relationship between the spectral bands and vegetation indices with target stain severity levels, yield and vigrain mass. The experiment was carried out in the 2022/23 harvest on a farm in Costa Rica/MS/BR, conducted with different fungicide treatments, to obtain different levels of disease severity. A spectroradiometer and remotely piloted aircraft imaging were used to collect spectral data from the leaves. The data was subjected to machine learning analysis using different algorithms. The RF (Random Forest) and SVM (Support Vector Machine) algorithms showed better performance in classifying the severity levels of the target spot, using reflectance. Multivariate analysis showed that healthy leaves stood out at specific wavelengths, while diseased leaves showed different spectral patterns. Disease detection using hyperspectral sensors has enabled detailed information acquisition. The study demonstrated that remote sensing, especially hyperspectral sensors and machine learning techniques can be effective in early detection and monitoring of target spot in soybean crops, allowing rapid action to control and prevent productivity losses.

Keywords: Glycine max. Remote Sensing. Precision Agriculture.