Support Vector Machine Coupled with Modular Visible-Near Infrared Spectroscopy for Chili (Capsicum annuum) Seed Viability Discrimination

This work investigated the performance of spectroscopic data in conjunction with a Gaussian support vector machine (SVM) for non-destructive discrimination of chili pepper (CP) seed viability. This present study also involved 2 wavelength selection methods, namely variable important in projection (VIP) and backward PLS (bPLS), which were realized and examined. The spectra data of CP seeds were collected at 2 regions: Visible-near infrared (Vis/NIR, 400 - 1000 nm) and shortwave near-infrared (SWNIR, 1000 - 1700 nm). The individual and mixed (generalized) discrimination models were then examined. This study demonstrated that a generalized model with effective Vis/NIR wavelengths through VIP achieved the optimum discrimination prediction accuracy (97.22 %). Thus, our findings successfully provided a general calibration model in a non-destructive way to discriminate CP seed viability. Our findings hold promises for practical implementation in the seed industry. HIGHLIGHTS Vis/NIR spectroscopy was used to discriminate the chili seed viability. Gaussian SVM produced 97.22 % accuracy. VIP method significantly reduced the number of wavelengths. GRAPHICAL ABSTRACT