*Important notice: This news reports on an unedited version of an accepted paper and is awaiting final editing. Therefore, the paper should not be regarded as conclusive or treated as established information.
In a paper published in Scientific Reports, researchers have introduced an automated electronic nose system that achieved 97.03% accuracy in distinguishing naturally ripened bananas from those treated with chemical ripening agents. The non-destructive system analyzes volatile organic compounds (VOCs) released during ripening using a low-cost gas sensor array.
Study: Identification of artificially ripened banana by chemical agent using gas sensor array. Image Credit: Aung Thurein Tun/Shutterstock.com
Can an Electronic Nose Spot Artificially Ripened Bananas?
Agricultural supply chains sometimes use low-cost chemical agents, such as ethephon and calcium carbide, to accelerate fruit ripening. However, concerns have been raised about the improper use of calcium carbide especially, due to its potential effect on fruit quality.
Traditional methods for detecting artificial ripening include gas chromatography, mass spectrometry, optical imaging, and wet-chemical assays. While these techniques provide detailed analyses, they often require expensive equipment and complex procedures, making them challenging for routine screening, especially in resource-limited markets.
Electronic noses offer a more accessible approach to assessing fruit quality.
Download a free copy of this page here!
These systems employ solid-state gas sensors based on metal-oxide semiconductors to detect VOCs emitted by fruit. When VOCs interact with heated metal-oxide sensing surfaces, reversible oxidation-reduction reactions alter the material's electrical resistance.
The sensor records these changes as an electrical pattern, producing a digital fingerprint of the gas mixture. The system uses the combined responses from multiple sensor networks to classify samples, making it suitable for portable, non-destructive screening of agricultural products.
Five Gas Sensors Track Banana Ripening Over 14 Days
To evaluate volatile emissions during fruit maturation, researchers developed a multi-channel gas-sensing system using Sagar Kola (BARI-1 hybrid banana), an AAA (Musa acuminata triploid hybrid) cultivar widely grown in South Asia.
They divided 20 unripe bananas into two groups of 10: one group ripened naturally, while the other was immersed in an aqueous ethephon solution (1.2 g/L) for five minutes to accelerate ripening.
The system used five commercial metal-oxide semiconductor sensors from Figaro Engineering Inc.: TGS 2602, TGS 2603, TGS 2610, TGS 2612, and TGS 2620. These sensors were selected for their distinct responses to various volatile compounds.
The sensors were installed in a custom dual-compartment metal chamber that separated the fruit samples from the sensing unit. Automated vacuum pumps moved volatile compounds from the sample compartment to the sensors.
Each measurement cycle included a 20-minute accumulation period, a five-minute sensing period, and a five-minute purge with fresh air. A National Instruments myRIO data acquisition card controlled the sampling process and recorded the sensor voltage outputs over 14 days.
The study generated 1009 time-series data points to track changes as the bananas moved from unripe to deteriorating. Four machine learning (ML) models were employed to analyze the data: logistic regression, support vector machines (SVMs), decision trees, and extreme gradient boosting (XGBoost).
The analysis addressed two tasks: distinguishing naturally ripened bananas from ethephon-treated ones and assigning samples to 28 classes based on their exact ripening days. Recursive feature elimination (RFE) and Shapley additive explanations (SHAP) were employed to assess feature importance and identify redundant sensor channels.
Machine Learning Detects Artificial Ripening With 97% Accuracy
The ML models demonstrated that nonlinear classifiers performed significantly better than linear models in distinguishing naturally ripened bananas from ethephon-treated samples. In the binary classification task, XGBoost, utilizing all five sensor channels, achieved a test accuracy of 97.03% and a 10-fold cross-validation accuracy of 94.91%.
The decision tree achieved an accuracy of 96.70%, while logistic regression reached only 59.08%, indicating that linear models struggled to differentiate the overlapping sensor response patterns.
Feature analysis using SHAP and RFE identified TGS 2603 and TGS 2620 as the most useful sensors for identifying artificial ripening. Removing TGS 2610 maintained XGBoost's test accuracy at 97.03% and increased its cross-validation accuracy to 95.05%, suggesting a four-sensor configuration could reduce system complexity without compromising accuracy.
For the 28-class task, the SVM achieved 99.95% test accuracy and 99.92% cross-validation accuracy using all five sensors. The decision tree achieved 99.94% test accuracy, while XGBoost reached 98.68%. Removing TGS 2612 maintained the SVM's test accuracy at 99.95%.
Overall, these results suggest that different sensor combinations are effective for identifying artificial ripening and tracking daily changes during fruit maturation.
Electronic Nose Could Bring Fruit Screening to Markets
This low-cost, non-destructive electronic nose system could, with further work, be used in fruit-packaging facilities, wholesale distribution hubs, and agricultural inspection checkpoints. Handheld or conveyor-mounted sensor units could screen fruit consignments and identify potentially chemically ripened batches before distribution.
Since the system employs commercially available sensors rather than specialized equipment, it could provide a more affordable option for routine inspection at packing facilities and markets.
The sensor architecture could also be integrated into cold-storage facilities and transport containers. Monitoring volatile compounds in ambient air could help track fruit ripening and estimate shelf life during storage and transit, thereby supporting storage management and reducing post-harvest losses.
The Possibility of Portable Fruit Testing
This study shows that multi-channel metal-oxide gas sensor arrays, combined with ML algorithms, provide a foundation for practical, non-destructive detection of artificial fruit ripening and maturity tracking. By analyzing VOC profiles, the system effectively distinguishes naturally ripened bananas from treated ones without damaging the fruit.
Future work should evaluate additional banana cultivars, other fruit varieties, and different ripening agents, and examine sensor performance under varying conditions. Miniaturizing the pneumatic system into battery-powered handheld devices with onboard data processing could support field deployment by food safety inspectors and agricultural distributors, making electronic nose systems more accessible for routine fruit quality monitoring.
Journal Reference
Jim, A.A.J., et al. (2026). Identification of artificially ripened banana by chemical agent using gas sensor array. Scientific Reports. DOI: 10.1038/s41598-026-76940-9, https://www.nature.com/articles/s41598-026-76940-9.
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.