Detection of Powdery Mildew of Bitter Gourd Based on NIR/Fluorescence Spectra

Purpose Powdery mildew as one of the common vegetable diseases has very rapid infection. Its outbreak will bring about disastrous consequences to vegetable output; thus, it is of key importance to do rapid identification and prevention of powdery mildew. Methods In this test, 100 bitter gourd leaves were collected as research samples, and the data of near-infrared spectra, fluorescence spectra, and chromatic values L*a*b* , and the classic K-S algorithm was adopted to divide the sample sets; then, the quantitative forecasting and qualitative discrimination models were established. First, Pearson’s correlation analysis was carried out to find the feasibility of taking a * as the modeling parameter, through cross-validation; the quantitative forecasting model was optimized by the PLSR (partial least squares regression) method. The model is also optimized by extracting the spectral feature bands using the continuous projection SPA method. Results The optimization results showed that the MSC + SPA + PLSR quantitative forecasting model of near-infrared spectra could effectively improve model precision, which was significantly higher than that of fluorescence spectra. Classification Leaner was used to establish the quantitative forecasting model. Compared with the model of near-infrared spectra, the SPA + SVM qualitative discrimination model of fluorescence spectra could improve the identification precision of powdery mildew of bitter gourd as high as 98% through training. Conclusion This study proposed different combination methods based on quantitative forecasting and qualitative discrimination and could provide a method and reference to the identification of powdery mildew of bitter gourd.