Evaluation of soybean plants affected by Aphelenchoides besseyi using remote sensing and machine learning techniques

Abstract

Soybeans (Glycine max (L.) Merrill) are a major player in food security, and pest loss control is a major focus of research and technological development by the agricultural sector. Among these pests, Aphelenchoides besseyi contaminates the aerial part of the plant, which can be detected in the leaf's spectral response, based on in situ hyperspectral sensors with the adoption of remote sensing techniques, such as spectral models. Assessing such data using machine learning allows the identification of optimal computational conditions to evaluate different levels of infection by the green stem nematode in soybeans. Thus, this research aimed to (i) discriminate the spectral bands most sensitive to nematode infection, (ii) identify the spectral model with the greatest accuracy for distinguishing different levels of nematode infection according to reflectance, and (iii) verify the resilience to the impact of A. besseyi on soybeans. From this approach, the near and short-wave infrared spectral portions contributed most to discriminating different amounts of nematodes in the plant, in a scenario in which the logistic regression algorithm had greater performance. Finally, this evaluation suggests that the best discrimination conditions occur with data obtained in the final half of the soybean cultivation cycle.