Hyperspectral imaging-based quality assessment of salted radish with spectral feature selection

This study aimed to use hyperspectral imaging in nondestructive monitoring of the changes in radish quality during salting process and to enhance the efficiency and accuracy of the prediction models through feature selection techniques. Hyperspectral imaging was utilized to predict the salinity, moisture content, and work of penetration (WOP) of radishes. Salinity increased with the prolonged salting time, whereas the moisture content and WOP decreased. Prediction using a partial least squares regression (PLSR) model based on full-wavelength hyperspectral data, resulted in the R_p² values for salinity, moisture content, and WOP being 0.909, 0.725, and 0.705, respectively. Feature selection methods, including competitive adaptive reweighted sampling (CARS), uninformative variable elimination (UVE), and their combinations, were applied to extract informative wavelengths for each quality parameter. With UVE + CARS, high R_p² values were achieved for salinity (0.934) and moisture content (0.846) using only 32.2 % and 25.7 % of the full-wavelengths, respectively. Similarly, R_p² values for WOP (0.717) improved with CARS, utilizing 31.8 % of the full-wavelengths. Hyperspectral imaging, coupled with suitable feature selection, not only improved the efficiency and accuracy of quality assessment during the salting process but also enabled the simultaneous prediction of key chemical and physical attributes of salted radishes with informative wavelength selection. This approach provides an efficient strategy for monitoring the quality of salted vegetables, offering valuable insights for optimizing industrial salting processes and advancing real-time quality control applications.