How Electronic Noses Detect Freshness Compounds?
Machine Learning Turns Sensor Signals into Decisions
Applications in Meat and Fish
Coffee, Dairy, and Other Products
Limitations Outside the Laboratory
Conclusion
References and Further Reading
Food spoilage starts before the food looks or smells bad. Bacteria grow on meat and fish, fats slowly oxidize in milk and oils, and each process releases small amounts of volatile organic compounds into the air around the product. Usually, quality checks depend on trained testers or lab equipment. This method can be slow, expensive, and hard to deploy in busy supply chains.
Image Credit: Natali _ Mis/Shutterstock
In response to these challenges, electronic noses have emerged as a promising option. This idea began in the early 1980s when researchers first combined gas sensor arrays with pattern recognition software to mimic the human sense of smell. This article covers the fundamentals of electronic noses, the role of machine learning in reading the signals, their applications in testing meat, fish, coffee, and dairy, and their limitations.1
How Electronic Noses Detect Freshness Compounds?
An electronic nose has three working parts. A sensor array samples the air above the food, a signal-processing unit converts the sensor responses into digital data, and a pattern-recognition system compares that data with odor profiles it has learned. Many low-cost devices also include humidity and temperature sensors, as these factors affect the sensitivity of gas sensors.1
The sensors themselves are available in different types. Metal-oxide-semiconductor sensors, the most commonly used, change their electrical resistance when volatile molecules land on a heated surface. Other types include conducting polymers, quartz crystal microbalances, electrochemical cells, and optical sensors, each operating through distinct physical principles. Each sensor generates broad responses to various gases, and the overlapping responses across the array create a combined fingerprint of the entire sample.1
As food ages, the fingerprints produced by these sensors change. In meat, enzymes and bacteria break down proteins, fats, and carbohydrates into ammonia, hydrogen sulfide, acetaldehyde, and biogenic amines, which accumulate steadily during storage. Since the sensors can detect rising concentrations early, an electronic nose can flag declining freshness even when the product still looks acceptable, which is the central promise of the technology.2
Machine Learning Turns Sensor Signals into Decisions
Multiple sensors produce overlapping curves that change over time, making it difficult to interpret the raw data stream. Machine learning algorithms sort these patterns into meaningful categories that a quality manager can act on. Principal component analysis (PCA) compresses the data. It reveals clusters, while linear discriminant analysis, support vector machines, k-nearest neighbors, and artificial neural networks classify samples as fresh, semi-fresh, or spoiled.1
Researchers have reported high accuracy with these tools. Across a decade of food studies, most portable, low-cost electronic noses paired with pattern recognition exceeded 90% accuracy within well-defined food categories. A beef study reported in MDPI Sensors demonstrated complete accuracy in separating fresh from spoiled samples using a simple three-sensor setup with PCA and a probabilistic neural network.1
However, these headline figures need careful examination. In a recent study on chicken freshness, a random forest model achieved 100% accuracy when readings were randomly divided between training and testing sets. But when samples from different storage days were kept separate, accuracy dropped to 69%. The best model reached only 78.5%. This happens because random splits can cause a model to memorize near-identical readings, inflating its apparent skill.3
Applications in Meat and Fish
Meat is the most studied application due to its cost, perishability, and susceptibility to microbial contamination. Electronic noses have been used to separate chilled from frozen meat, track the effects of processing temperature and time, detect species adulteration, and monitor shelf life. Some researchers are also testing whether specific odor signatures can reveal pathogenic bacteria, which would expand the use of this technology from quality checks to food safety.2
Fish spoils even faster, and its decay produces trimethylamine, dimethylamine, and ammonia. These compounds constitute total volatile basic nitrogen, a standard measure of freshness. Electronic noses directly target these compounds, and some sensors can detect them in just seconds. Recently, an eight-sensor system correctly distinguished fresh Tuna from contaminated samples 99% of the time. Similar systems have also successfully graded crabs and prawns.1,4
Coffee, Dairy, and Other Products
Coffee shows how electronic noses can evaluate quality factors such as roast level. The roasting process generates most of the volatile compounds that contribute to coffee's aroma. Traditionally, roasters judge the result by eye and smell, which makes consistency hard to guarantee between batches. This is where the electronic nose comes in. It uses a gas sensor array and a smart neural network to analyze roast levels. Recent studies show that the electronic nose can accurately identify five roast levels with 95.9%-98.8% accuracy.5
In contrast, dairy and fat-rich foods involve different chemistry. Lipid oxidation produces aldehydes, ketones, alcohols, and hydrocarbons that cause rancid off-flavors in milk, edible oils, meat, and cocoa. Electronic noses track these oxidation volatiles as response patterns, which allows rapid shelf-life checks. Studies have also used the technology to trace milk to its source and to authenticate tea, spices, and alcoholic beverages.1,6
Limitations Outside the Laboratory
Sensor drift is the most common issue with electronic noses. Gas sensors slowly change their response to the same stimulus because of material aging and surface contamination. A model trained on new sensors gradually misreads samples as the hardware ages. To address this, drift compensation algorithms and adaptive machine learning can help, but they also introduce complexity and still require regular reference measurements.1,2
Another related issue is calibration. Even sensors of the same model can behave differently when exposed to the same gas because of manufacturing inconsistencies. This means that replacing one sensor might require recalibrating the entire device. Researchers are developing calibration transfer methods to move a trained model between instruments, along with standardized data-collection protocols. However, these methods have not yet been widely validated in industrial settings.6
Saving this for later? Download a PDF here.
Adding to the complexity, real-world conditions such as fluctuations in humidity and temperature can affect sensor responses. Moreover, complex food matrices release many overlapping compounds that blur the signals. Most research studies focus on individual foods in controlled settings, and the lack of field validation is a major gap between promising prototypes and reliable commercial tools.1
Conclusion
Electronic noses can detect chemical signs of declining freshness before visible changes appear. Their sensor arrays respond to the amines, sulfur compounds, and oxidation products that build up during storage. Machine learning then converts these responses into practical grades for meat, fish, coffee, dairy, and many other foods, often with high reported accuracy under controlled conditions.
The technology's next step depends on reliability in everyday use. For it to transition from research to everyday use, stable sensors, transferable calibrations, and rigorous testing under actual storage conditions are required. Closing these gaps could make early spoilage detection a common practice in keeping food safe and reducing waste.
References and Further Reading
- Sanislav, T. et al. (2025). A Comprehensive Review on Sensor-Based Electronic Nose for Food Quality and Safety. Sensors, 25(14), 4437. DOI:10.3390/s25144437. https://www.mdpi.com/1424-8220/25/14/4437
- Munekata, P. E. et al. (2022). Applications of Electronic Nose, Electronic Eye and Electronic Tongue in Quality, Safety and Shelf Life of Meat and Meat Products: A Review. Sensors, 23(2), 672. DOI:10.3390/s23020672. https://www.mdpi.com/1424-8220/23/2/672
- Anwar, H., & Anwar, T. (2025). Quality assessment of chicken using machine learning and electronic nose. Sensing and Bio-Sensing Research, 47, 100739. DOI:10.1016/j.sbsr.2025.100739. https://www.sciencedirect.com/science/article/pii/S2214180425000054
- Tonezzer, M. et al. (2021). Quantitative Assessment of Trout Fish Spoilage with a Single Nanowire Gas Sensor in a Thermal Gradient. Nanomaterials, 11(6), 1604. DOI:10.3390/nano11061604. https://www.mdpi.com/2079-4991/11/6/1604
- Astuti, S. D. et al. (2024). Electronic nose coupled with artificial neural network for classifying of coffee roasting profile. Sensing and Bio-Sensing Research, 43, 100632. DOI:10.1016/j.sbsr.2024.100632. https://www.sciencedirect.com/science/article/pii/S221418042400014X
- Floch, M., Gunenc, A., & Hosseinian, F. (2026). Miniaturized Electronic Noses for Food Quality and Safety: Applications in Lipid Oxidation Monitoring. Molecules, 31(19), 3424. DOI:10.3390/molecules31193424. https://www.mdpi.com/1420-3049/31/19/3424
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.