Rapid nondestructive hardness detection of black highland Barley Kernels via hyperspectral imaging

The objective of this study was to propose a rapid and nondestructive method for quantitatively detecting the hardness of black highland barley kernels using hyperspectral imaging. Initially, a regression model was established to predict hardness based on beta-glucan content. Spectral reflectance within the 400-1000 nm wavelength range was gathered for black highland barley, and six preprocessing techniques were applied. Once preprocessing was completed, three characteristic wavelength screening methods were employed. Finally, three different models were utilized to construct a dependable prediction model for beta-glucan content. The results indicated that the one-dimensional convolutional neural network (1D-CNN), in combination with the moving average (MA) preprocessing method, exhibited the best performance. To validate the hardness prediction model, the beta-glucan content prediction model was integrated with the hardness regression model. The hardness pre- diction model attained a coefficient of determination (R-2) value of 0.8093 and root mean square error (RMSE) of 0.2643 kg. The visual images exhibit characteristics feature of hardness in different varieties of black highland barley. These findings offer insights into the feasibility of designing a noncontact system to monitor the quality of black highland barley.