A Smartphone Could Soon Tell When a Banana or Kiwifruit Is Spoiling
A small chemical indicator placed inside fruit packaging, combined with a smartphone camera and an artificial-intelligence model, could turn an invisible stage of ripening into an instantly readable signal. Researchers at Seoul Women’s University in South Korea have developed a system that detects the freshness of bananas and kiwifruit in real time by tracking ethylene, a plant hormone released as many fruits ripen. The approach links a potassium permanganate-based ethylene indicator to a mobile application powered by deep learning. In principle, a consumer, retailer or food distributor could photograph the indicator through the package and receive an automated freshness assessment without opening, touching or damaging the fruit. The study addresses a persistent problem in the food supply chain: external appearance is often an unreliable guide to internal quality, while laboratory measurements of gases, texture or chemical composition can be too slow or expensive for routine use. By converting fruit physiology into a visible color response and then allowing software to interpret that response, the researchers aim to create a low-cost bridge between intelligent packaging and everyday food monitoring.
The system is built around the biology of climacteric fruits. Unlike non-climacteric fruits, climacteric fruits undergo a characteristic ripening process associated with a rise in respiration and increased ethylene production. Ethylene binds to receptors in plant tissues and activates signaling pathways that alter gene expression, accelerating processes such as starch breakdown, softening, pigment changes and the development of aroma compounds. Bananas and kiwifruit are both strongly influenced by this hormone, although their visible and biochemical changes occur at different rates. Ethylene can accumulate inside a sealed or semi-sealed package, making it a potentially useful marker of ripening progression. It is not, however, a complete definition of spoilage: microbial contamination, bruising, water loss and temperature history can also determine whether fruit is safe or desirable to eat. The researchers therefore treat ethylene as a measurable indicator of freshness-related change rather than as a universal substitute for every quality test. Their objective was to determine whether a chemical sensor could respond consistently enough to support an automated image-based classification system.
Potassium permanganate, or KMnO₄, provides the chemical component of the detector. The compound is a powerful oxidizing agent that can react with ethylene, effectively removing the gas while undergoing a change that can be expressed through the indicator’s color. In an intelligent package, this chemistry transforms a gas concentration that cannot be seen by the naked eye into an optical signal. The researchers tested indicators containing 0.1, 0.5 and 1.0 percent KMnO₄ by weight per volume. The different concentrations were intended to reveal how sensor formulation affects sensitivity and how closely the visual response tracks the ethylene released by the fruit. Too little reactive material could produce a weak or delayed signal, while a higher concentration might alter the response range or timing. Among the formulations examined, the 0.5 percent indicator showed the highest correlation between ethylene concentration and color change. That result identifies a practical middle ground for the tested conditions, although it does not establish that the same concentration will be optimal for every fruit, package design, temperature or storage duration.
For the experiments, bananas and kiwifruit were stored at 25 degrees Celsius for 10 days in polypropylene pouches containing the indicator. The setup created a controlled package environment in which ethylene released during ripening could interact with the sensing material. Polypropylene is commonly used in food packaging because it is lightweight and provides a controllable barrier to moisture and gases, but the precise exchange of oxygen, carbon dioxide and ethylene depends on pouch thickness, sealing and design. Those variables matter because gas accumulation determines how quickly a sensor changes. Temperature also has a major influence on fruit metabolism and chemical reaction rates; storage at 25 degrees Celsius represents a warm, accelerated ripening condition rather than every situation encountered during refrigerated transport or household storage. Across the storage period, the researchers compared the indicator’s optical response with ethylene-related freshness changes. The strongest relationship occurred with the 0.5 percent formulation, suggesting that the indicator could encode ripening information in a form suitable for image analysis. The study did not present the indicator as a preservation treatment, and it should not be confused with a packet that extends shelf life.
The second half of the innovation is software. The team trained a ResNet50 deep-learning model to interpret images of the indicator and predict the freshness status of the fruit. ResNet50 is a convolutional neural network architecture designed for visual recognition. Its defining feature is the use of residual connections, which allow information and gradients to pass through many layers more effectively during training. Rather than relying only on manually selected measurements such as average hue or brightness, a deep neural network can learn complex visual patterns from labeled examples, including subtle combinations of color distribution, intensity and spatial variation. In this application, the model does not directly smell the fruit or measure ethylene with a conventional gas analyzer. It infers the fruit’s freshness category from the image of a chemical response that has already integrated information about the package atmosphere. That distinction is important: the model’s performance depends on the quality and consistency of the indicator, the lighting conditions, the camera and the training data. A visually impressive prediction is only as reliable as the chain of chemical, photographic and statistical measurements behind it.
According to the researchers, the ResNet50 model achieved high accuracy when predicting the freshness of both bananas and kiwifruit, and the trained system was incorporated into a mobile application for real-time analysis. A user can photograph the indicator in the fruit package, after which the application processes the image and returns a freshness assessment. Mobile imaging offers a potentially powerful advantage over laboratory instrumentation because smartphones are already widely available and can perform sophisticated computer-vision tasks. The application could also standardize interpretation, reducing dependence on a person’s ability to judge small color differences. For retailers, such a system might support inventory rotation by identifying packages approaching a ripening threshold. For households, it could make freshness information more visible before food is discarded. For researchers and manufacturers, the same platform could be adapted to other colorimetric sensors. Yet “high accuracy” in a controlled study is not equivalent to perfect performance in the real world. Lighting, reflections from plastic, condensation, camera differences and background colors can all shift the apparent signal. Robust deployment would require testing across phones, packaging formats, cultivars and storage environments.
The work is part of a broader movement toward intelligent food packaging, in which labels do more than display a sell-by date. Conventional dates are assigned using expected storage conditions and conservative estimates, but they do not necessarily reflect the actual history of an individual package. A sensor that responds to biological or chemical changes could provide more dynamic information. Similar research has explored colorimetric systems for meat and other foods, while potassium permanganate has also been studied as an ethylene scavenger because removing ethylene can slow ripening. Combining sensing and machine learning adds a layer of interpretation: instead of asking a person to compare a label with a color chart, an algorithm can evaluate the image against patterns learned from experimental data. The result could be a more flexible freshness label, but it also raises practical questions about cost, disposal, chemical containment and regulatory requirements. Potassium permanganate must remain isolated from direct food contact, and an indicator designed for packaging would need to be stable, safe and resistant to accidental leakage. The study demonstrates a detection concept, not a final commercial package.
The researchers’ findings are especially relevant because food waste often occurs at the boundary between uncertainty and caution. Fresh produce can be discarded because shoppers or retailers cannot determine how much useful life remains, even when the fruit is still edible. A rapid indicator could help distinguish ripening from more advanced deterioration, potentially improving decisions throughout distribution. But the distinction between freshness and safety remains essential. Ethylene accumulation is closely tied to ripening in climacteric fruit, whereas harmful microorganisms may grow without producing a matching indicator response. A package that receives a favorable AI assessment should not override basic food-safety practices, and a negative assessment would not by itself identify the cause of deterioration. The next steps for this technology will likely involve broader validation under fluctuating temperatures, varying humidity and realistic transportation conditions, as well as tests involving different ripeness stages and fruit varieties. Researchers will also need to report detailed model-performance measures and establish how the application behaves when images fall outside its training set. For now, the study shows how plant hormones, oxidation chemistry, computer vision and mobile software can be combined into a single window on the hidden life of packaged fruit.
Subject of Research: Real-time freshness and spoilage detection of bananas and kiwifruit using a potassium permanganate-based ethylene indicator and a deep-learning mobile application
Subject of Research: Biology
Article Title: Real-time spoilage detection of climacteric fruits using a potassium permanganate-based ethylene indicator and deep learning-based mobile application
Article References: Kim, B. Y., Moh, C.-M., & Min, S. C. (2026). Real-time spoilage detection of climacteric fruits using a potassium permanganate-based ethylene indicator and deep learning-based mobile application. Food Science and Biotechnology, 35(9), 2557-2570. https://doi.org/10.1007/s10068-026-02200-1
Image Credits: AI Generated
DOI: 10.1007/s10068-026-02200-1
Keywords: intelligent packaging, fruit spoilage, ethylene indicator, potassium permanganate, bananas, kiwifruit, ResNet50, deep learning, mobile application
Cite this news
APA MLA Chicago
SCIENMAG. (August 27, 2026). Real-time deep-learning app detects climacteric fruit spoilage via potassium-permanganate ethylene indicator. https://scienmag.com/real-time-deep-learning-app-detects-climacteric-fruit-spoilage-via-potassium-permanganate-ethylene-indicator/
SCIENMAG. “Real-time deep-learning app detects climacteric fruit spoilage via potassium-permanganate ethylene indicator.” Scienmag, 27 August 2026, https://scienmag.com/real-time-deep-learning-app-detects-climacteric-fruit-spoilage-via-potassium-permanganate-ethylene-indicator/. Accessed 27 August 2026.
SCIENMAG. “Real-time deep-learning app detects climacteric fruit spoilage via potassium-permanganate ethylene indicator.” Scienmag. August 27, 2026. https://scienmag.com/real-time-deep-learning-app-detects-climacteric-fruit-spoilage-via-potassium-permanganate-ethylene-indicator/
Copy citation Download RIS
Tags: AI-powered fruit quality assessmentAI-powered fruit ripeness analysisautomated fruit freshness assessmentchemical-based ethylene detection in food packagingclimacteric fruits like bananas and kiwifruit produce ethylene gas during ripeningdeep learning in food quality assessmentdeep learning-based mobile app for fruit spoilage detectionethylene gas sensingethylene indicator technology for real-time fruit spoilage monitoringfood supply chain spoilage preventionfruit ripeningfruit ripening detectionintelligent packaging for climacteric fruitsintelligent packaging solutions for food freshnesslow-cost food freshness sensing systemsnon-invasive fruit quality testingportable ethylene detection technologypotassium permanganate ethylene indicatorreal-time fruit spoilage monitoringreal-time monitoring of climacteric fruit spoilagesmartphone-based freshness detectionsmartphone-based fruit ripeness detectionwhich can be detected for freshness assessment
