LAUSR

ABSTRACT Asian soybean rust (Phakopsora pachyrhizi) is the most severe disease in soybean crops production. The early detection of the disease by traditional methods involves visual inspection of the symptoms present in the leaves and is expensive and time-consuming. The limitations of visual detection have led to an interest in the development of spectroscopically based detection techniques for the rapid diagnosis of this disease. Thus, this work aimed to develop a procedure for early and accurate detection and differentiation of soybean under different levels of Asian rust disease, based on spectral analysis and linear discriminant analysis (LDA), with optimum wavelengths selection by a stepwise procedure. Reflectance spectroscopy ranging from the visible (Vis) to the near-infrared (NIR) region (350–2,500 nm) was obtained by a Fieldspec 3 Jr. hyperspectral sensor through the spectral measurement of soybean leaves with different levels of disease that had the following treatments: uninfected (T1), severity 0.6% (T2), severity 2.0% (T3), severity 7.0% (T4), severity 18.0% (T5), and severity 42.0% (T6). There were 15 spectral curves measured in each treatment, totalling 90 spectral samples. Principal component analysis (PCA) was applied as an indicator of the explained variance of the reflectance spectra among the different disease progressions. The spectral signature of the leaves showed the existence of a strong increase in reflectance in the Vis region when the levels of disease increased, associated with a lower concentration of pigments. The PCA explained over 97.00% of the spectral variance in the first and second principal components and the stepwise procedure selected from 87 spectral bands. The LDA achieved global accuracies of 100.00% and 82.51%, in the calibration and validation procedures, respectively. These results suggest the spectral reflectance technique as a promising tool for cost-effective, fast analysis and a non-destructive method for diagnosis Asian soybean rust.