A freshness intelligent label integrating vision and gas sensing and a dynamic pricing method
By integrating intelligent tags with visual and gas sensing capabilities and a dynamic pricing system, the problems of real-time freshness perception and rigid pricing for perishable goods have been solved, enabling precise management and maximizing profits, reducing losses and improving supply chain efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHAANXI SCI TECH UNIV
- Filing Date
- 2026-03-01
- Publication Date
- 2026-06-02
Smart Images
Figure CN122133692A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of Internet of Things, intelligent sensing and retail management technology, and specifically to an intelligent label, system and dynamic pricing method based on monitoring results for real-time monitoring of the freshness of perishable goods. Background Technology
[0002] Perishable goods such as fresh produce, chilled meat, and prepared meals suffer significant losses in the supply chain. Statistics show that approximately 1.3 billion tons of food are wasted globally each year in the supply chain, with fresh perishable foods comprising the majority, causing substantial economic and environmental damage. Currently, retailers primarily rely on printed "best before" dates to manage goods. This method, based on static and uniform time calculations, fails to reflect the individualized spoilage process caused by variations in transportation and storage conditions (such as temperature fluctuations), making it neither scientific nor accurate. This often leads to two consequences: first, safe-to-eat goods are discarded prematurely, resulting in waste; second, expired goods are not detected in time, causing food safety issues.
[0003] In existing technologies, some research has emerged on smart packaging or sensor tags to address the aforementioned problems. For example:
[0004] 1. Single-gas response labels: These labels use indicators or sensors sensitive to specific gases (such as ammonia or hydrogen sulfide). When the gas concentration exceeds a threshold, the label color changes. While these labels provide some information, food spoilage is a complex biochemical process involving multiple volatile substances. Single indicators are easily interfered with, have poor specificity, and cannot quantify the degree of spoilage, let alone distinguish between different types of spoilage (such as bacterial spoilage and fungal mold).
[0005] 2. Time-Temperature Indicator (TTI): This indicator simulates food quality decay through the cumulative effect of temperature. However, TTI only reflects temperature history and cannot account for other key spoilage factors such as initial colony differences and packaging gas environment, resulting in limited predictive accuracy.
[0006] 3. External vision-based monitoring systems: These systems use shelf cameras to capture images of product appearance for AI recognition. However, this method is greatly affected by lighting conditions, product placement angles, and packaging obstructions (especially non-transparent packaging). Furthermore, it is completely unable to detect key biochemical changes inside the packaging, limiting its application scenarios.
[0007] 4. Laboratory instrument testing methods: such as measuring volatile basic nitrogen (TVB-N) and total bacterial count. These methods are accurate, but they are destructive, offline testing methods that are time-consuming and costly, and cannot be used for real-time, online, and non-destructive monitoring in the supply chain.
[0008] In terms of retail pricing, fresh produce is traditionally priced based on experience, with uniform discounts applied as the expiration date approaches. This method is slow to react, lacks flexibility, and the "near-expiration discount" label itself may raise consumer concerns about product quality. Dynamic pricing (revenue management) is already well-established in the airline and hotel industries, but in the fresh produce retail sector, the core obstacle lies in the lack of real-time, objective, and quantifiable data on the individual quality status of goods as a basis for pricing.
[0009] Therefore, there is an urgent need for an intelligent, closed-loop solution that can perceive the freshness of perishable goods in a multi-dimensional, real-time, in-situ, and non-destructive manner, and can directly and automatically transform the perceived data into business decisions (such as precise pricing). Summary of the Invention
[0010] (a) Purpose of the invention
[0011] The primary objective of this invention is to provide a smart label and system for freshness that integrates vision and gas sensing, in order to solve the problem that existing technologies cannot achieve real-time, accurate, and comprehensive freshness perception of perishable goods (especially packaged goods).
[0012] Another objective of this invention is to provide a dynamic pricing method and system based on the aforementioned intelligent tag sensing data, in order to solve the problems of rigid pricing and high loss rates in fresh food retail, and to achieve automated and intelligent revenue management based on the real-time quality status of goods.
[0013] (II) Technical Solution
[0014] To achieve the above objectives, the present invention adopts the following technical solution:
[0015] Option 1: A smart freshness tag integrating vision and gas sensing
[0016] This smart tag is an electronic device attached to product packaging or directly integrated with the product. Its core function is to collect multimodal spoilage feature data in situ and in real time and perform edge intelligent processing.
[0017] The smart tag includes the following modules:
[0018] 1. Main Control Unit: Serving as the "brain" of the system, this is typically an ultra-low-power microcontroller (MCU) integrating an artificial intelligence processing unit (NPU), such as a chip based on the ARM Cortex-M series architecture. It is responsible for coordinating the work of various modules, running lightweight AI algorithm models, processing data, and making local decisions.
[0019] 2. Multi-channel gas sensing module: Connected to the main control unit. This module contains a gas sensor array consisting of at least two miniature sensors selectively responding to different target gases, preferably metal-oxide-semiconductor (MOX) gas sensors. Typical sensor combinations are used to detect: ammonia (NH3) and trimethylamine (TMA) indicating protein spoilage, hydrogen sulfide (H2S) indicating microbial activity, ethylene (C2H4) indicating fruit ripening, and the total amount of volatile organic compounds (VOCs) or carbon dioxide (CO2) comprehensively reflecting the degree of spoilage. The module also integrates temperature and humidity sensors for real-time environmental compensation of gas sensor readings, improving data accuracy.
[0020] 3. Miniature Visual Sensing Module: Connected to the main control unit. This module includes a low-resolution, low-power CMOS image sensor and one or more LED auxiliary light sources of specific wavelengths (e.g., white LEDs for color recognition, near-infrared LEDs for detecting deeper features such as moisture and mold). This module is configured to acquire images of the product surface or a specific area at regular intervals (e.g., every 2-6 hours) or triggered by gas concentration changes.
[0021] 4. Low-power display screen: Connected to the main control unit, preferably an electronic ink screen (E-Ink). Its characteristic is that it consumes power only when refreshing the screen, and the displayed content remains after a power outage. It is used to intuitively display the current dynamic price of the product, freshness level (such as icons or text for "Fresh," "Good," or "Consume as soon as possible"), traceability QR code, and other information.
[0022] 5. Power Supply Module: Powers the entire tag and can use a miniature rechargeable lithium battery or a flexible printed battery. For easy reuse, a wireless charging receiver coil can be integrated, supporting contactless charging via RFID / NFC readers or a dedicated charging dock.
[0023] 6. Communication Module: Connected to the main control unit, used for data interaction with external networks. Low-power wide-area network (LPWAN) technology, such as LoRa or NB-IoT, is preferably used to achieve stable communication with low data volume, long distance, and low power consumption.
[0024] 7. Packaging Structure: All electronic components are encapsulated within a shell made of food-grade safe materials. The shell is designed with micropores for gas diffusion, allowing for sufficient exchange of gas between the packaging and the gas sensor, while also providing some dust and condensation protection. Labels can be adhesive, hanging, or embedded to suit different packaging formats.
[0025] Option 2: A Freshness Assessment Method Based on Multimodal Fusion
[0026] This method operates on the main control unit of the smart tag or a cloud server communicating with it, and is used to process raw sensor data to generate accurate freshness metrics. At its core is a lightweight multimodal information fusion artificial intelligence model.
[0027] The method includes the following steps:
[0028] S201: Data Preprocessing. Filtering (e.g., moving average filtering) and normalization are performed on the gas sensor time-series signals. White balance correction, region of interest (ROI) cropping, and image enhancement are performed on the acquired images.
[0029] S202: Feature extraction.
[0030] Gas Feature Extraction: From the preprocessed multi-channel gas concentration data, time-domain and frequency-domain features are extracted to construct a gas feature vector. .For example:
[0031]
[0032] in, For the first The average concentration of the gases gas With gas The concentration ratio (which can be used as a specific spoilage indicator). For the first The rate of change in the concentration of a gas.
[0033] Visual Feature Extraction: Using lightweight convolutional neural networks (such as a simplified version of MobileNetV2) or traditional image processing algorithms, extract visual feature vectors related to corruption from images. Key features include:
[0034] Color characteristics: Calculate color histograms and statistical moments in HSV or CIELAB color spaces to monitor browning and fading.
[0035] Texture features: Contrast, entropy, etc. are calculated using Local Binary Mode (LBP) and Gray-Level Co-occurrence Matrix (GLCM) to monitor surface stickiness, dryness, or mold formation.
[0036] Morphological characteristics: By edge detection and contour analysis, the area shrinkage rate and roundness changes are calculated to monitor shrinkage and deformation.
[0037] S203: Multimodal Fusion and Decision Making. A feature-level fusion strategy is employed to integrate gas feature vectors. and visual feature vectors Concatenate into a fused feature vector :
[0038]
[0039] Will Input a lightweight multilayer perceptron (MLP) or a small neural network that outputs a continuous freshness coefficient. Its range is ,in Represents the freshest state. This represents complete putrefaction. The model can simultaneously output the probability of putrefaction type (e.g., bacterial putrefaction probability). Probability of fungal mold growth .
[0040] The training data for the model was obtained by conducting accelerated spoilage experiments on target commodities (such as pork, fish, and leafy green vegetables) in a controlled laboratory environment. Sensor data and high-definition images were collected synchronously and frequently during the experiment, and corresponding standard physicochemical indicators (such as total bacterial count, TVB-N value, and pH value) were measured as ground truth for model training.
[0041] Option 3: A dynamic pricing method and system based on real-time freshness.
[0042] This method establishes an automated business loop from perception to pricing execution.
[0043] The system includes: a group of smart tags deployed on each product, wireless gateways deployed in the retail area, and a cloud management platform.
[0044] The method includes the following steps:
[0045] S301: Data Upload and Aggregation. The smart tag periodically uploads the calculated freshness coefficient. Raw sensor data summary (or only) (Upload detailed data in case of anomalies) The data is sent to the local gateway via the LoRa / NB-IoT network, and then forwarded to the cloud platform by the gateway.
[0046] S302: Dynamic pricing decision. The pricing engine on the cloud platform receives product pricing data. In time Freshness coefficient Then, combining the following factors, the optimal selling price is calculated using a pricing optimization model. :
[0047] Base price The initial price when the product is listed.
[0048] Real-time freshness index Core pricing basis.
[0049] Inventory levels : The remaining quantity of this product category / batch.
[0050] Demand forecasting model Predicting prices and freshness The immediate demand.
[0051] External market factors Examples include time of day (morning / evening market), holidays, and competitor prices.
[0052] The pricing optimization model aims to maximize the expected total revenue, and is constructed as follows constrained optimization problem:
[0053]
[0054] (Inventory constraints)
[0055] (Price boundary constraints)
[0056] (Safety threshold constraint)
[0057] in, Let be the total number of goods to be priced. This problem can be solved online using reinforcement learning algorithms (such as Q-learning, Deep Deterministic Policy Gradient (DDPG)) or real-time optimization solvers.
[0058] S303: Price command issuance and display. The cloud will calculate the optimal price. The corresponding marketing messages (such as "freshness guaranteed" and "limited-time offer") are sent to the relevant smart tags. The tag's main control unit drives the e-ink screen to refresh, displaying the new price and related information.
[0059] S304: Security Monitoring and Early Warning. The cloud platform continuously monitors all products. When any product Below the preset safety threshold When this happens, the platform automatically sends a delisting warning to the store's inventory management system and can display a "Stop Sales" message on the label screen.
[0060] (III) Beneficial Effects
[0061] Compared with the prior art, the present invention has the following significant advantages:
[0062] 1. Comprehensive Perception Dimensions, Precise and Reliable Assessment: For the first time, in-situ vision and multi-channel gas sensing depth are integrated into a single tag. Visual information compensates for the insensitivity of gas sensing to surface spoilage (such as mold), while gas information reveals internal biochemical spoilage that cannot be observed visually. This multimodal fusion fundamentally improves the accuracy and robustness of freshness assessment and can distinguish spoilage types, providing a basis for precise treatment.
[0063] 2. Achieve true "state management": Transform the management model of perishable goods from "shelf life management" based on fixed time periods to "state management" based on real-time sensing data. Each product has its own personalized "life cycle," and management decisions are based on objective conditions rather than subjective speculation.
[0064] 3. Building an automated business closed loop to improve operational efficiency: A complete automated closed loop has been formed, encompassing "perception (sensing) - cognition (AI analysis) - decision-making (pricing optimization) - execution (screen display update)". This significantly reduces the workload of manual inspections and manual price adjustments, and transforms dynamic pricing from theory into a large-scale, practical application.
[0065] 4. Significantly reduced losses and increased efficiency, with clear economic benefits: Accurate freshness assessment avoids the accidental discarding of marketable goods; dynamic pricing based on condition enables "pricing according to quality" and maximizes profits. The combination of these two factors can be expected to reduce fresh produce losses by 20%-40% while simultaneously improving overall gross profit margin.
[0066] 5. Enhance consumer transparency and trust: Linking price changes with intuitive freshness indicators makes the reasons for discounts transparent, eliminates consumers' doubts about "near-expiry products," and improves the shopping experience and brand trust.
[0067] 6. Empowering end-to-end digitalization: The high-granularity, time-series quality data generated by this invention is a valuable asset for optimizing cold chain logistics, inventory forecasting, and supplier evaluation, and can drive the digital upgrade of the entire fresh food supply chain. Attached Figure Description
[0068] Figure 1 This is a block diagram of the hardware system structure of the smart tag in an embodiment of the present invention.
[0069] Figure 2 This is a schematic diagram of the workflow of the freshness assessment fusion model in this embodiment of the invention.
[0070] Figure 3 This is a schematic diagram of the overall architecture and data flow of the dynamic pricing system in an embodiment of the present invention.
[0071] Figure 4 This is a schematic diagram illustrating the principle of the pricing optimization model in an embodiment of the present invention.
[0072] Figure 5 This is a schematic diagram showing the relationship between the freshness coefficient and the concentration of key gases and the change in meat color in Embodiment 1 (chilled pork) of the present invention. Detailed Implementation
[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and do not constitute a limitation on the scope of protection of this invention.
[0074] Example 1: Smart label system for tray packaging of chilled pork
[0075] This embodiment uses supermarket chilled pork packaging as an application scenario.
[0076] 1. Hardware implementation: such as Figure 1 As shown, the smart tag is designed as a thin sheet, with a size of approximately The casing is made of food-grade ABS plastic with a strong adhesive sticker on the bottom. The gas sensing module contains three MOX sensors, sensitive to ammonia / amines, hydrogen sulfide, and VOCs respectively. The vision module's camera lens is aimed at the boundary between the pork fat and lean meat, equipped with an 850nm near-infrared LED and a white LED. The main control uses an STM32H7 series MCU (with a Cortex-M7 core and Chrom-ART accelerator). The screen is a 3.7-inch three-color e-ink display. Communication uses a LoRa module, operating at a frequency of 470MHz.
[0077] 2. Model Training and Deployment:
[0078] Data Acquisition: Twenty pork samples from the same batch were monitored for 7 days under refrigeration conditions of 4±0.5℃. Sensor data was recorded hourly, and a set of high-resolution calibration images was captured every 6 hours. Simultaneously, three samples were randomly selected daily for destructive testing to measure their TVB-N value (national standard method) and total bacterial count.
[0079] Label creation: Using the TVB-N value as the primary quality indicator, it is normalized to... The interval, as the true value of freshness The training dataset is constructed by combining the sensor time-series data with the corresponding images and their ground truth values.
[0080] Model Training: A fusion network was constructed. The gas feature extraction part used a one-dimensional convolutional layer to process the temporal data; the vision part used a pruned MobileNetV2 backbone network to extract features; after fusion, the output FS was fed into two fully connected layers. The mean squared error (MSE) was used as the loss function, and training and quantization were performed in the TensorFlow Lite Micro framework.
[0081] Model deployment: Burn the quantized .tflite model file into the MCU of the smart tag.
[0082] 3. Workflow:
[0083] After the pork packaging is sealed, the smart tag is affixed to the transparent film on the top of the packaging box.
[0084] The tag starts a work cycle every 4 hours: wake up the sensor, collect 30 seconds of stable gas data, and take an image.
[0085] The MCU runs the local fusion model and calculates the current... .like The label remains silent, and the display shows a "fresh" icon and the original price (e.g., 39.8 yuan / kg).
[0086] like The model will provide a suggested discount rate based on its internal weights. A simplified local pricing rule could be:
[0087]
[0088] in This is the minimum guarantee factor (e.g., 0.6).
[0089] At the same time, the label will and Uploaded to the store's gateway via LoRa.
[0090] Cloud-based pricing engines aggregate all data and run more complex optimization models, potentially leading to... Make fine adjustments to form the final instructions. The label is then sent to the e-ink screen, which refreshes to display the new price (e.g., "¥31.8 7.9% off").
[0091] like The label screen displays a red "Please remove from shelf" sign and sends an emergency alert to the cloud, which then notifies the store staff to handle the situation.
[0092] 4. Results: After a week of in-store pilot testing, pork shelves using this smart tag saw an average loss rate reduced by 35%, and sales during the night clearance period (after 8 pm) increased by 50%, as dynamic pricing attracted price-sensitive customers.
[0093] Example 2: Smart Tag for Strawberry Plastic Box Packaging
[0094] Strawberry spoilage is mainly caused by fungal mold and fermentation, and it happens very quickly.
[0095] 1. Hardware Customization: The label is designed as a pad, placed at the bottom of the plastic box. The visual sensor lens faces upwards, capturing images of the bottom and sides of multiple strawberries (the initial area of mold growth). The gas sensor is specifically configured with elements sensitive to ethylene and ethanol (fermentation products). To address rapid spoilage, the sampling cycle is shortened to 1 hour.
[0096] 2. Model and Strategy Characteristics: Among visual features, mold detection has a high weight. The pricing strategy is more aggressive. A decrease to 0.8 may trigger the first discount, maximizing sales opportunities through frequent price adjustments.
[0097] 3. System linkage: When multiple strawberry labels in a certain batch... When the values drop rapidly across the board within a short period of time, the cloud system can identify it as a cold chain disruption and immediately alert logistics management personnel.
[0098] The parts of this invention not described in detail are well-known in the art. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A smart freshness label integrating visual and gas sensing, characterized in that, include: Main control unit; multi-channel gas sensing module, connected to the main control unit, used to detect the concentration of various specific spoilage-related gases in the environment where the goods are located; A miniature visual sensing module, connected to the main control unit, is used to acquire image information of the product surface; a low-power display screen, connected to the main control unit, is used to display information. A communication module, connected to the main control unit, is used for data interaction; wherein, the main control unit is configured to run a freshness assessment model, which calculates the freshness coefficient of the product based on fused data input from the multi-channel gas sensing module and the miniature visual sensing module. .
2. The freshness smart label according to claim 1, characterized in that, The multi-channel gas sensing module includes at least two metal oxide semiconductor gas sensors that are sensitive to ammonia (NH3), trimethylamine (TMA), hydrogen sulfide (H2S), ethylene (C2H4), volatile organic compounds (VOCs), or carbon dioxide (CO2), respectively.
3. The freshness smart label according to claim 1, characterized in that, The miniature vision sensing module includes a low-power CMOS image sensor and an LED auxiliary light source; the low-power display screen is an e-ink screen.
4. A dynamic pricing method based on the smart tag according to any one of claims 1 to 3, characterized in that, Includes the following steps: S1: Collect gas concentration data and surface image data of the target product through the smart tag, and calculate the real-time freshness coefficient. ; S2: Based on the real-time freshness coefficient Basic commodity price Inventory levels Taking into account market demand factors, the optimal selling price is calculated using a pricing optimization model. ; S3: Set the optimal selling price The message is sent to the smart tag and drives its display screen to update.
5. The dynamic pricing method according to claim 4, characterized in that, In step S2, the pricing optimization model aims to maximize the expected total revenue, and its mathematical expression is: And subject to the following constraints: (Inventory constraints) (Price boundary constraints) (Safety threshold constraint) in, The total number of goods to be priced. This is the demand forecasting function.
6. The dynamic pricing method according to claim 4, characterized in that, In step S1, the freshness coefficient This is obtained through a multimodal information fusion model, which integrates gas feature vectors. and visual feature vectors The gas feature vectors are fused together. It includes the average concentration, concentration ratio, and concentration change rate characteristics of multiple gases.
7. The dynamic pricing method according to claim 6, characterized in that, The multimodal information fusion model employs a feature-level fusion strategy to integrate gas feature vectors. and visual feature vectors Concatenate into a fused feature vector The freshness coefficient is obtained by inputting it into a neural network. .
8. The dynamic pricing method according to claim 4, characterized in that, When the real-time freshness coefficient Below the preset safety threshold At that time, the system will automatically generate a removal warning message.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the dynamic pricing method as described in any one of claims 4 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the dynamic pricing method as described in any one of claims 4 to 8.