Hyperspectral imaging with different illumination patterns for the hollowness classification of white radish

This study presented the detection of hollowness in the worldwide important vegetable crop white radish (Raphanus sativus L.) by using hyperspectral imaging covering the spectral range of 400-1000 nm. The hyperspectral images based on the three illumination patterns of reflectance, transmittance, and semi-transmittance were acquired from white radishes. The successive projections algorithm (SPA) was used to identify the optimal wavelengths from the three patterns of spectra. Two classifiers of partial least square discrimination analysis (PLS-DA) and back propagation artificial neural network (BPANN) were established based on the full wavelengths and selected wavelengths. Discrimination models were performed for the two-class, three-class, and five-class hollowness classifications using the mean spectra from the regions of interest (ROI) in the spectra images. The classification results showed that hyperspectral semi-transmittance imaging combined with the BPANN model performed the best classification accuracy for the two-class hollowness classification based on the full and selected wavelengths reaching 98% and 97% for the calibration and the prediction sets, respectively. Lower accuracies were obtained for the three-class and five-class hollowness classifications based on the combination of classifiers and illumination modes. The results demonstrated that hyperspectral semi-transmittance imaging was potentially useful as a non-invasive method to identify the hollowness in white radishes.