A research team led by Prof. WANG Qi from Hefei Institutes of Physical Science, Chinese Academy of Sciences, together with Luzhou Laojiao Co., Ltd, have developed two data-fusion methods for rapid and accurate identification of brewing sorghum varieties.
The related studies have been published in Microwave and Optical Technology Letters and Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy.
Sorghum is the main raw material used to make Chinese Baijiu. However, different varieties can be difficult to distinguish quickly and accurately, making rapid quality screening a challenge.
The first method combines information on key components in sorghum with conventional near-infrared (NIR) spectral analysis. Tested on more than 20,000 sorghum grains, the method achieved an accuracy of 87.86% and an F1-score of 89.99%, outperforming conventional identification methods.
The second method combines NIR spectroscopy with machine vision. By using both spectral data and images of sorghum kernels, the method enables the two types of information to complement each other in identifying different varieties. Tests on six representative sorghum varieties showed that the method achieved 98.88% accuracy and a Macro-F1 score of 98.74%, outperforming methods based on either spectral or image information alone.
Both methods were tested with the team' s self-developed high-throughput detection device, offering a practical way to rapidly screen and sort brewing sorghum varieties. The findings could help improve raw-material quality control in the Baijiu industry.
High throughput variety discrimination for single sorghum grains based on Multimodal Attention Network (MAN) (Image by XU Zhuopin)