Rapid sorghum variety identification by hyperspectral imaging combined with super-depth-of-field microscopy

Abstract

Sorghum, as the primary raw material for brewing, has varieties that are crucial to the quality and yield of the brewing process. To accurately identify and classify different sorghum varieties, a Two-Dimensional Feature Adaptive Convolution Model (DD-FACM) based on data acquired by Hyperspectral imaging (HSI) and 3D Super-Depth-of-Field Microscopy was built. The experimental results demonstrated that the DD-FACM that was built using the combined spectral data and super-depth-of-field image data achieved 100 % accuracy in the identification of 5 varieties of sorghum grains, which was 8 %, 4.2 %, and 4.1 % higher than the classification accuracies of the support vector machine (SVM) model that was built based only the spectral data, the EfficientNet_B3 model built using only the depth-of-field image data, and the DD-FACM built using the combination of the HSI(spectral and RGB image) data, respectively. To verify the effectiveness of the DD-FACM's feature extraction, the extracted features were visualized using t-distributed stochastic neighbor embedding (t-SNE). The results indicated that the DD-FACM based on the spectral data and the image data could achieve the rapid, accurate, and non-destructive identification of different sorghum varieties. This study not only provides brewing enterprises with an efficient method for sorghum variety identification but also offers technical support for variety identification research in related fields.