A recent study published in Food Research International demonstrates how visible and near-infrared spectroscopy (Vis-NIRS) combined with machine-learning algorithms can accurately authenticate meat and fat based on livestock feeding systems, offering a sustainable and reliable solution for traceability in the meat industry.
Traceability is important for consumers when selecting meat products for purchase. Consumers increasingly want to know where their food originated from, as well as the ingredients in it, to ensure they are purchasing high-quality products. To better ensure the authenticity and quality of meat products, researchers and manufacturers are conducting more quality control tests of meat products under different feeding systems.
A recent study led by Nuria Prieto from the Lacombe Research and Development Centre explored this topic. The study, published in Food Research International, highlights how visible and near-infrared spectroscopy (Vis-NIRS), combined with advanced machine-learning algorithms, offers a reliable, eco-friendly method to verify meat origins (1).
Meat production has quadrupled over the past 50 years (2). Asia is the largest producer, leapfrogging both Europe and North America (2). By livestock type, poultry and pigs continue to lead the industry in terms of production, with approximately 139 million tons of poultry and 122 million tons of pigs produced as of 2022 (2). Cattle came in third place in terms of overall production, and it was the main focus on this study.
In their study, the research team examined beef from 45 steers raised on three distinct diets: barley, corn, and grass-fed systems. To investigate the authenticity of meat and fat, the study collected Vis-NIRS spectra across wavelengths ranging from 380 to 2,500 nanometers (1). These samples included subcutaneous fat, intact longissimus thoracis (LT) muscle at 72 hours postmortem, and ground longissimus lumborum (LL) after freezing and thawing (1).
Fresh raw beef steak sirloin with rosemary | Image Credit: © Sławomir Fajer - stock.adobe.com
Having samples of beef from three distinct diets were important to the study. The researchers wanted to see whether they could classify meat and fat based on the animals’ diets (1). Two machine learning (ML) techniques were used in conjunction with spectral analysis: partial least squares-discriminant analysis (PLS-DA) and linear-support vector machine (L-SVM) (1).
The results revealed the potential of Vis-NIRS in discriminating between feeding systems. Subcutaneous fat samples yielded remarkable classification accuracy, with PLS-DA correctly classifying 100% of the test samples across all spectral regions (visible, near-infrared, and combined Vis-NIR) (1). In comparison, L-SVM achieved accuracy rates between 75% and 100% (1).
For intact meat samples, PLS-DA also reached 100% classification success in the Vis-NIR spectral range (1). Although L-SVM exhibited slightly lower performance in this category, it still achieved test accuracy rates of up to 100% in some spectral regions (1). In ground meat samples, both PLS-DA and L-SVM demonstrated good classification, with PLS-DA consistently delivering a good performance.
The study identified critical variables influencing the spectral data, including fat and meat pigments, fatty acids, protein, and moisture absorption. These components played a significant role in distinguishing samples from grass-fed, barley-fed, and corn-fed cattle (1).
Notably, the study emphasized the environmental and operational benefits of Vis-NIRS technology. The non-destructive method eliminates the need for chemical reagents and lengthy processing, aligning with the growing demand for green analytical tools in the food industry (1).
Although both PLS-DA and L-SVM proved effective, PLS-DA emerged as the better approach because of its higher classification accuracy and easier data interpretation. The study underscores PLS-DA's advantages in terms of efficiency and reliability, positioning it as the preferred machine-learning algorithm for Vis-NIRS-based meat authentication (1).
This advancement is particularly relevant as consumer awareness of production practices, such as grass-fed versus grain-fed livestock, continues to grow. Accurate labeling not only meets consumer expectations, but it also supports producers who adopt sustainable and ethical farming practices (1). Meat is an important source of nutrition, so this industry is expected to grow considerably in the future (2).
While the study successfully demonstrated Vis-NIRS's potential for meat authentication, the authors note that further research with larger sample sizes and diverse feeding systems is needed to confirm its broader applicability (1). Additionally, integrating Vis-NIRS technology into routine industrial processes could improve quality control in the meat supply chain.
NIR Spectroscopy Explored as Sustainable Approach to Detecting Bovine Mastitis
April 23rd 2025A new study published in Applied Food Research demonstrates that near-infrared spectroscopy (NIRS) can effectively detect subclinical bovine mastitis in milk, offering a fast, non-invasive method to guide targeted antibiotic treatment and support sustainable dairy practices.
Smarter Sensors, Cleaner Earth Using AI and IoT for Pollution Monitoring
April 22nd 2025A global research team has detailed how smart sensors, artificial intelligence (AI), machine learning, and Internet of Things (IoT) technologies are transforming the detection and management of environmental pollutants. Their comprehensive review highlights how spectroscopy and sensor networks are now key tools in real-time pollution tracking.
New AI Strategy for Mycotoxin Detection in Cereal Grains
April 21st 2025Researchers from Jiangsu University and Zhejiang University of Water Resources and Electric Power have developed a transfer learning approach that significantly enhances the accuracy and adaptability of NIR spectroscopy models for detecting mycotoxins in cereals.