Detection of Verticillium infection in cotton leaves using ATR-FTIR spectroscopy coupled with machine learning algorithms

Abstract

Verticillium wilt (VW) is a soil-borne vascular disease that affects upland cotton and is caused by Verticillium dahliae Kleb. A rapid and user-friendly early diagnostic technique is essential for the preventing and controlling VW disease. In this study, Fourier transform infrared (FTIR) spectroscopy with attenuated total reflectance (ATR) technology was used to detect VW infection in cotton leaves. About 1800 FTIR spectra were obtained from 348 cotton leaves. The cotton leaves were collected from three categories: VW group, infected group and control group (non-infected). The vibrational peak of chitins at 1558 cm−1 was identified through mean and differential analysis of FTIR spectra as a criterion to differentiate the VW or infected group from the control group. Classification models were constructed using various machine learning algorithms. The support vector machines (SVM) model exhibited the highest predictive accuracy (>96 %) in each group and a total accuracy (>97 %) for the three groups. These results provide a new approach for detecting Verticillium infection in cotton leaves and shows a promising potential for the future applications of the method in plant science.