Abstract:

Hollowness is a common defect found in radish postharvest storage. In the present study, a prototype hyperspectral
imaging system was designed for evaluating the internal quality of white radish. Three different detection models including
semi-transmittance, reflectance and transmittance were evaluated and used to extract the hyperspectral imaging data of
white radish, partial least squares discriminant analysis (PLS-DA), support vector machine (SVM), and artificial neural
network (ANN) algorithms were then used to establish the hollowness model for radish identification and the recognition
accuracy was calculated. The prediction accuracies based on PLS-DA, SVM, and ANN were 72.5%, 72.5% and 83.3% in
semi-transmittance mode, 82.5%, 82.5% and 92.3% in reflectance mode, and 90.0%, 90.0% and 94.3% in transmittance
mode, respectively. The results showed that hyperspectral transmittance imaging achieved the best prediction results among
the three different detection models, ANN algorithm was the optimal algorithm to build hollowness discrimination model.
Hyperspectral transmittance imaging in the combination with ANN gave the best results with a prediction accuracy of 94.3%
for detecting the internal hollowness of white radish. Therefore, it was feasible to use hyperspectral transmittance imaging
system for detecting the hollowness of white radish in postharvest storage.

Key words: hyperspectral imaging, detecting model, white radish, hollowness