Combined hyperspectral imaging technology with 2D convolutional neural network for near geographical origins identification of wolfberry

The composition of the functional components in wolfberry may be influenced by different geographical origins, and it is critical to identify the geographical origins of wolfberry. This paper proposes a near geographical origins identification model of Ningxia wolfberry using hyperspectral imaging technology (HSI) and 2D convolutional neural network (2D-CNN) algorithms. The model used methods including competitive adaptive reweighed sampling (CARS), uninformative variable elimination (UVE) and interval variable iterative space shrinking analysis (iVISSA) to extract different characteristic wavelengths. The 2D-CNN model was compared with linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), combined principal component analysis with linear discriminant analysis (PCA-LDA), combined principal component analysis with quadratic discriminant analysis (PCA-QDA), and partial least squares discriminant analysis (PLS-DA). For the original spectral dataset, the 2D-CNN model performed the best, with the calibration set and prediction set accuracies of 97.4% and 91.32%, respectively. For the characteristic wavelength dataset, the 2D-CNN-iVISSA-CNN model exhibited the best accuracy, with the calibration set and prediction set accuracies of 98.12% and 90.27%, respectively. All models established based on the characteristic wavelengths extracted by iVISSA method perform well in identifying the Zhongning wolfberries. Especially for the 2D-CNN-iVISSA-CNN model achieves the best classification effect with an accuracy of up to 99.5% and 97.7% on the training and prediction set for the Zhongning wolfberries. This study indicates that the combination of hyperspectral imaging technology and 2D-CNN algorithms can identify the near geographical origins of wolfberry efficiently.