Researchers at Shanxi Agricultural University have pioneered the use of visible and near-infrared (VIS-NIR) hyperspectral imaging combined with machine learning to non-destructively detect the nutrient contents of sorghum. This innovative approach targets the quantification of crude protein, tannin, and crude fat, offering significant improvements over traditional chemical methods, which are often costly, destructive, and time-intensive.
Sorghum, known for being one of the world's major cereal crops, is cultivated largely in arid and semi-arid regions. It offers numerous nutritional benefits, including protein, starch, and amino acids. The diversity among sorghum varieties leads to variable nutrient compositions, making accurate nutrient content detection critically important for its application across food production, animal feed, and brewing industries.
Traditional methods for quantifying nutrient compositions, like chemical assays, can only analyze single components at once, leading to wasted time and resources. Such methods also typically require extensive sample preparation, specialized equipment, and can destroy tissue samples, rendering them unusable for subsequent analyses. This study seeks to address such limitations through rapid, efficient, and non-destructive techniques.
The research involved the collection of data from 93 different sorghum varieties, amounting to 279 samples, which were scanned using VIS-NIR hyperspectral imaging systems. The technology captures detailed spectral images across multiple wavelengths, allowing researchers to assess individual components using machine learning algorithms.
To identify the key wavelengths associated with nutritional components, advanced algorithms such as Competitive Adaptive Reweighted Sampling (CARS) and Bootstrapping Soft Shrinkage (BOSS) were implemented. These techniques were instrumental for feature extraction and selecting the most relevant wavelengths for predicting nutrient contents. The study reported successful outcomes whereby specific wavelengths related to crude protein, tannin, and fat were identified: 41, 38, and 22 respective wavelengths, showcasing the effectiveness of this innovative detection approach.
The detection models constructed from the spectral data exhibited strong coefficients of determination, with models achieving Rp2 values demonstrating satisfactory predictive capabilities for protein, fat, and tannin contents. Notably, the accuracy of the models was supported by comparison with traditional chemical analysis methods.
The results unveiled the potential of this non-destructive technology to revolutionize how nutrient content is assessed within agricultural applications. Instead of traditional, labor-intensive methods, farmers and producers can leverage VIS-NIR hyperspectral imaging, augmented by machine learning analytics, to quickly and accurately assess nutrient quality.
According to the authors of the article, "these detection models effectively achieved real-time and nondestructive detection of crude protein, tannin, and crude fat contents in sorghum grains." They also highlighted how this technology assists grain quality assurance, significantly benefitting agricultural practices and product optimization.
For future developments, the researchers indicate plans to integrate multiple predictive models with big data analytics and artificial intelligence technologies. This effort aims to create real-time nutritional assessment systems for grains, enabling simultaneous detection of various components. Such advancements promise to streamline agricultural processes, ensuring the quality and safety of food sources across the board.