Accurate prediction of soluble solid content in dried Hami jujube using SWIR hyperspectral imaging with comparative analysis of models

Detection of Water pH Using Visible Near-Infrared Spectroscopy and One-Dimensional Convolutional Neural Network

“…Compared with the SPA-PLS model, the RMSEP is reduced by 22.37%, and the is increased by 12.95%. As shown in Figure 9 a, the 1D-CNN model shows that more sample points are closed to the ideal regression line, which indicates that the prediction performance of 1D-CNN is better than the other four traditional regression models [ 11 , 38 ].…”

“…In the 1D-CNN model training process, mean squared error (MSE) was employed as the loss function and calculated in Equation (2) [ 11 ]. where n is the number of water samples in the calibration set, and are the predicted pH and reference pH of the i th water samples, respectively.…”

“…The performance of the calibration models is evaluated according to the values of the root mean square error (RMSE) of calibration (RMSEC), prediction (RMSEP), and coefficient ( R 2 ) of determination calibration ( ) and prediction ( ) [ 11 , 33 ]. Generally, the lower value of RMSE and the closer the R 2 is to 1 indicate that the prediction result from the calibration model is more reliable and the calibration model has a better-predicted performance [ 11 ].…”

“…A lot of multivariate calibration algorithms have been developed to build the relationship between the Vis-NIR spectra and the target attributes, including partial least squares (PLS), least squares support vector machine (LS-SVM), artificial neural network (ANN), and extreme learning machine (ELM) [ 8 , 9 , 10 ], etc. Among these methods, the PLS (linear) and LS-SVM (nonlinear) are the most commonly used multivariate calibration methods [ 11 ]. The high dimension and high complexity are the main characteristics of modern spectroscopy.…”

“…Cui et al (2018) proposed a unified CNN structure used for multivariate regression, and the results indicated that the CNN models achieved the highest prediction accuracy [ 14 ]. Li et al (2022) explored the potential of using the short-wave infrared (SWIR) hyperspectral imaging (HSI) technique combined with a one-dimensional convolutional neural network (1D-CNN) for predicting the SSC of dried Hami jujube, and the results indicated that the prediction accuracy of 1D-CNN was satisfactory [ 11 ]. Mishra et al (2022) presented a strategy of 1D-CNN modeling for the multi-response prediction for spectral data of fresh fruit, they found that the prediction performance of 1D-CNN is better than PLS [ 15 ].…”