Chemometric development using portable molecular vibrational spectrometers for rapid evaluation of AVC (Valsa mali Miyabe et Yamada) infection of apple trees

Abstract

Detecting of apple valsa canker (AVC), which is caused by Valsa mali Miyabe et Yamada, at early infection stage is beneficial to disease prevention and control for ensuring yield and quality of apples. Applying simple, economical and non-destructive rapid detection method for the early diagnosis of AVC is of great significance for precision management of orchards. In this research, near-infrared (NIR) spectroscopy (900−1700 nm) and Raman scattering (0−2000 cm−1) combined with machine learning algorithms were employed to diagnose 3 infection degrees of AVC (healthy, disease 1, disease 2) based on optimal variables which were selected by chemometric methods. NIR and Raman spectroscopy were obtained using portable spectrometers in reflection mode, respectively. Firstly, adaptive iterative reweighting partial least squares (air-PLS) was utilized to remove fluorescence background in Raman spectra. Secondly, clustering analysis was developed using principal component analysis (PCA). After that, optimal variables were selected by x-loadings (XLs) of PCA and competitive adaptive reweighted sampling (CARS) algorithm, respectively. Four optimal wavelengths at 983, 1156, 1395, and 1457 nm from NIR spectra and five optimal wavenumbers at 175, 229, 326, 409, and 1523 cm−1 from Raman spectra were identified by XLs. Six optimal wavelengths at 943, 949, 967, 1240, 1632, 1666 nm from NIR spectra and eight optimal wavelengths at 412, 1500, 1585, 1596, 1599, 1602, 1672 and 1709 cm−1 from Raman spectra were selected according to CARS. Finally, AVC diagnosing models were developed using least square support vector machine (LS-SVM), and receiver operating characteristic (ROC) curves were applied to evaluate the performance of the classification models. Overall, the relatively good classification results on optimal variables (94.67% and 97.33% in NIR, 89.33% and 89.33% in Raman) were obtained using LS-SVM. This study confirmed that molecular vibrational spectroscopy techniques (NIR and Raman) is promising for early detecting AVC and providing a practical way for diagnosing diseases in large-scale orchards.