Computer Vision-Based Fire–Ice Ion Algorithm for Rapid and Nondestructive Authentication of Ziziphi Spinosae Semen and Its Counterfeits


In order to verify the accuracy and reliability of the results, four machine learning algorithms, BP, SVM, DBN, and RF, were developed for further analysis. Data matrices were constructed using original data (chromaticity values and texture parameters) and fire–ice ion data, respectively, which were then imported into the MATLAB 2020b software (MathWorks, Natick, MA, USA) to create these models. Accuracy is an important metric for assessing the performance of the algorithmic models. The results in Figure 8 show that in the data matrices constructed with the original data (chromaticity values and texture parameters) from ZSS, ZMS, and HAS, the training accuracies of the BP, SVM, DBN, and RF classifiers were 90.35%, 83.04%, 85.96%, and 100.00%, respectively. And the testing accuracies are 88.36%, 77.40%, 84.25%, and 82.88%. In the data matrix constructed with fire–ice ion data from ZSS, ZMS, and HAS, the training accuracies of BP, SVM, DBN, and RF classifiers were 99.12%, 98.83%, 99.12%, and 100.00%. And the testing accuracies were 98.63%, 95.89%, 99.32%, and 99.32%, respectively. This result shows that the reduced dimensionality data can be better classified, and the learning effect of the model can be significantly improved. Meanwhile, the graphs based on the confusion matrix are shown in Figure S1. Obviously, in the data matrix constructed with the original data (chromaticity values and texture parameters) from ZSS, ZMS, and HAS, the BP model shows better discrimination accuracy, while in the data matrix constructed with the fire–ice ion data from ZSS, ZMS, and HAS, the RF model shows better discrimination accuracy. In addition, the convergence speed of the model is significantly improved due to the reduction in data volume after dimensionality reduction. In summary, using the fire–ice dimensionality reduction algorithm significantly enhances the classification accuracy and convergence efficiency of machine learning algorithms. This finding is particularly valuable for practical applications involving high-dimensional data.



Source link

Peng Chen www.mdpi.com