Continuum removal method for monitoring Fulvia fulva morbidity using hyperspectral data

Fulvia fulva is a major disease in tomato cultivation. Compared with traditional laboratory analysis method, hyperspectral remote-sensing technology can provide simple, cost effective and non-destructive information that can offer processing methods for diagnosing and quantifying plant health. However, there are many limitations (e.g., large volume of data, redundant information and complex spectral) in dealing with hyperspectral data. This paper aimed to clarify the spectrum characteristics of tomato leaf infected by F. fulva and estimate its morbidity degree to provide theoretic basis for large-scale monitoring of F. fulva using hyperspectral remote sensing. To this end, experiments were carried out in 2016 in with disease nursery of tomato F. fulva in Shangqiu. In the research, leaf spectral reflectance of tomato was acquired via ASD FieldSpec 3 spectrometer (350-2 500 nm). The continuum removal method was adopted to process the original spectrum reflectance of tomato leaf with different morbidity degrees of F. fulva. The bands sensitive to F. fulva morbidity degree were selected and an inversion model of morbidity degree established based on absorption parameters of the spectrum features. The results showed that spectral reflectance of healthy tomato plants was higher than that of disease plants in the wavelength range of 350-2 500 nm. Besides, the reflectance, spectral sensitivity and relative reflectance decreased with increasing F. fulva morbidity degree. The most sensitive wave bands for distinguishing F. fulva severity were located in the visible region (550-730 nm) and shortwave infrared region (1 860-2 260 nm). With increasing F. fulva morbidity degree, the absorption position (λ) of both visible spectrum and shortwave infrared spectrum moved to the short wavelength band, while the maximum absorption depth (D_c) and area (A) increased. Particularly, the morbidity degree had a very significant correlation with maximum absorption depth in visible band (D_c1), maximum absorption area in shortwave infrared band (A₂), maximum absorption depth in shortwave infrared band (D_c2), position of maximum absorption depth in visible band (λ₁) and position of maximum absorption depth in shortwave infrared band (λ₂). Consequently, a stepwise regression model for F. fulva morbidity degree was built based on the spectral absorption parameters. The model had good validation results, with determination coefficient (R²) of 0.81. The results of the study not only contributed to the estimation of F. fulva morbidity degree using hyperspectral remote-sensing data, but also had promising values of practical application in monitoring and preventing crop diseases.