A fruit and vegetable cold chain environment monitoring method based on multi-sensor fusion

Abstract

This invention discloses a method for monitoring the cold chain environment of fruits and vegetables based on multi-sensor fusion, belonging to the field of intelligent sensing and agricultural product logistics technology. It includes: using a flexible multimodal sensor array attached to the surface of fruits and vegetables to acquire the temperature, humidity, and original level signals of the target gas in the surface microenvironment; performing a cross-drift compensation step based on local temperature and humidity physical feedback, using a multivariate nonlinear regression model to use temperature and humidity as dynamic compensation factors, calibrating the baseline drift of the gas sensor in real time, and obtaining high-confidence gas concentration values; calculating the rate of change of the gas concentration values ​​in real time, and automatically triggering an edge interruption when the rate of change exceeds a preset jump threshold, dynamically increasing the sampling frequency and adjusting the radio frequency duty cycle. This invention solves the technical problems of large measurement errors caused by the inability of traditional rigid sensors to conformally fit, severe sensor baseline drift in high humidity environments, and the mutual constraint between high-frequency monitoring and node energy consumption.

Description

Technical Field

[0001] This invention relates to a method for monitoring the cold chain environment of fruits and vegetables based on multi-sensor fusion, belonging to the technical field of intelligent sensing and agricultural product logistics. Background Technology

[0002] In the cold chain logistics system for fresh agricultural products, fruits and vegetables, as rapidly deteriorating organisms, experience post-harvest quality deterioration rates that are highly dependent on dynamic parameters of the storage microenvironment, including temperature, relative humidity, oxygen, carbon dioxide, ethylene concentration, and mechanical vibration. Studies show that the global loss rate of fruits and vegetables in post-harvest distribution is as high as 30% to 50%, with a significant proportion of this loss occurring during cold chain transportation and storage. Although temperature control technology has been widely deployed, the lack of precise perception of microenvironmental fluctuations and the physiological state of fruits and vegetables means that the loss problem has not been fundamentally curbed.

[0003] Currently, cold chain environmental monitoring mainly relies on two technological approaches: one is based on macroscopic environmental sensors at the container or cold storage level, and the other is based on IoT monitoring nodes using multi-sensor fusion. However, existing technologies generally face the following technical challenges in practical deployment: First, there is a physical disconnect between the sensing method and the monitored object. Traditional sensors often use rigid packaging structures, which cannot conformally fit the varied shapes of fruit and vegetable surfaces. A physical gap exists between the sensor probe and the fruit peel, resulting in the collected data being the average concentration of macroscopic mixed gases inside the packaging, rather than the true metabolic state of the microenvironment on the fruit and vegetable surface. Especially in gas detection (such as ethylene and carbon dioxide), the path length error of rigid probes can be as high as 15% to 20%, severely weakening the representativeness and accuracy of the monitoring data.

[0004] Secondly, sensor drift is a significant issue in cold chain environments. Under conditions of high humidity, low temperature, and frequent temperature changes (such as pre-cooling and defrosting), electrochemical or semiconductor gas sensors are highly susceptible to baseline drift. The cross-sensitivity effect of the sensitive materials to moisture further exacerbates measurement deviations. Existing compensation methods largely rely on factory calibration curves, which cannot adapt to dynamically changing field environments. This leads to a significant decrease in the confidence level of long-term monitoring data, making it difficult to support subsequent quality interpretation and early warning decisions.

[0005] Third, the contradiction between monitoring accuracy and energy consumption is difficult to reconcile. To achieve real-time capture of key physiological events such as the respiratory pulsation and ethylene release in fruits and vegetables, sensor nodes need to maintain high-frequency sampling and edge computing capabilities. However, high-frequency operation significantly increases battery power consumption and shortens node lifespan. If a low-frequency sampling strategy is adopted, transient events such as "cold chain breakage" or respiratory peaks are easily missed, resulting in monitoring blind spots and data distortion.

[0006] Fourth, the challenge of communication link penetration in enclosed metal environments. Refrigerated shipping containers and vans are mostly made of metal, and the densely packed, water-containing agricultural products inside have a strong absorption and shielding effect on radio frequency signals. This makes it difficult for sensor nodes deployed deep within the container to establish a stable communication link with the external gateway, hindering real-time data transmission and creating information silos.

[0007] In summary, existing monitoring technologies for the cold chain environment of fruits and vegetables suffer from systemic bottlenecks in terms of sensor fit, environmental adaptability, energy efficiency balance, and data integrity, making it difficult to achieve high-confidence perception and dynamic response to the microenvironment of fruits and vegetables. Therefore, there is an urgent need for a monitoring method that can achieve accurate, stable, low-power consumption, and physiological feedback capabilities in complex cold chain environments to overcome the limitations of current technological approaches. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a method for monitoring the cold chain environment of fruits and vegetables based on multi-sensor fusion. It solves the technical problems of large measurement error caused by the inability of traditional rigid sensors to conformally fit, severe baseline drift of sensors under high humidity environment, and the mutual constraint between high frequency monitoring and node energy consumption.

[0009] The technical problem to be solved by this invention is achieved by the following technical solution: A method for monitoring the cold chain environment of fruits and vegetables based on multi-sensor fusion, characterized by the following steps: Step S1: A flexible multimodal sensor array attached to the surface of representative fruits and vegetables from the same batch is used to acquire real-time physicochemical electrical signals of the microenvironment of the fruit and vegetable skin. The signals include at least local temperature readings, local humidity readings, and target gas level outputs characterizing the respiration and ripening state of the fruits and vegetables. The flexible multimodal sensor array conformally conforms to the curvature of the fruit and vegetable surface. Step S2: Perform cross-drift compensation based on local temperature and humidity physical feedback. That is, use a microprocessor to execute a multivariate nonlinear regression model, use the local temperature reading and local humidity reading as dynamic compensation factors, calibrate the baseline drift of the target gas level output in real time, and obtain a high-confidence target gas concentration value. Step S3: Calculate the concentration change rate of the high-confidence target gas concentration value in real time. When the system detects that the concentration change rate exceeds the preset jump threshold, it automatically triggers an edge interruption and dynamically increases the sampling frequency of the flexible multimodal sensor array.

[0010] The present invention is further configured such that, in step S2, the multivariate nonlinear regression model is specifically a BP neural network model optimized by the genetic simulated annealing algorithm. This model performs segmented processing for different temperature ranges to achieve accurate error calibration under different harsh micro-environments.

[0011] The present invention is further configured such that: in step S3, the concentration change rate is a derivative; when the derivative exceeds the jump threshold or a drastic step change occurs in the local temperature, the microprocessor not only increases the sampling frequency of each sensor module, but also synchronously adjusts the duty cycle of the passive RFID tag used for data transmission.

[0012] The present invention is further configured such that, in step S1, in order to minimize thermal crosstalk and electromagnetic interference when acquiring signals, the flexible multimodal sensing array adopts a specific topological spatial isolation layout: the heat-generating microcontroller circuit is arranged on the outer edge of the substrate, and the high-precision micro thermometer, thin-film capacitive hygrometer, and gas-sensitive microarray coated with polyethyleneimine and single-walled carbon nanotubes are arranged in a ring at a specific interval in the central region.

[0013] The present invention is further configured such that the method also includes a biochemical visual indication monitoring step: the surface of the flexible multimodal sensor array is integrated with a hydrogel breathable window, the window containing natural anthocyanins that are sensitive to volatile alkaline gases or drastic changes in environmental pH. When fruits and vegetables rot, a color change is generated through the color development layer to provide an intuitive naked-eye visual interpretation signal.

[0014] The present invention is further configured such that: after obtaining the high-confidence target gas concentration value, data is transmitted through a system separation and anti-attenuation relay fusion network, specifically including: the flexible multimodal sensor array uses ultra-high frequency RFID technology to perform short-distance data exchange with edge aggregation nodes arranged inside the cold chain container; after the edge aggregation nodes complete data cleaning and feature-level compression, they use low-frequency magnetic coupling technology or physical wires penetrating the container to transmit the data across the sealed metal wall of the container to a communication gateway outside the container, and then the communication gateway transmits the data to a cloud server through a cellular network.

[0015] The invention is further configured such that when the logistics vehicle travels to an area with no cellular network signal, the edge aggregation node automatically switches to independent data logger mode and continuously stores multi-source sensor readings using a local clock; once the network signal is recaptured, the breakpoint resume mechanism and timestamp fusion algorithm are activated to ensure the integrity of the data on the timeline.

[0016] The present invention is further configured such that, during the acquisition of the real-time physical, chemical and electrical signals, the system utilizes the piezoelectric effect generated by the minute mechanical vibrations in the cold chain logistics environment, or the thermoelectric effect generated by the temperature difference inside and outside the cold storage, to continuously replenish the micro-electrical energy of the flexible multimodal sensing array through a multi-source energy harvesting microsystem.

[0017] The present invention is further configured such that the method also includes an active predictive quality control step: the compensated multimodal fusion data is uploaded to the cloud-based digital twin model of fruits and vegetables in real time, and combined with real-time machine learning predictive analysis and the enzyme-induced respiration rate calculation model characterized by the Michaelis equation, the water loss rate, hardness reduction degree and spoilage probability of the monitored fruits and vegetables within a set time in the future are predicted prospectively.

[0018] The present invention is further configured such that the method also includes a smart scheduling step based on the prediction results: when the cloud-based digital twin model of fruits and vegetables predicts that the probability of softening or spoilage of fruit and vegetable tissue exceeds a set safety threshold, the system autonomously triggers a smart contract deployed on the blockchain to generate dynamic decision instructions, and redirects the batch of goods in real time to the nearest retail terminal for price reduction and promotion, or directly transfers them to the processing plant.

[0019] The beneficial effects of this invention are: 1. By constructing a multimodal sensor array using a flexible polymer substrate, it can conformally adhere to the surface of fruits and vegetables, effectively eliminating the physical gap between traditional rigid sensors and the fruit peel, and significantly reducing path length measurement errors. This structural design enables the sensor to directly capture real metabolic signals in the microenvironment of the fruit and vegetable peel, including local temperature and humidity changes and real-time concentration output of target gases (such as ethylene and carbon dioxide), providing a high-confidence data foundation for subsequent quality assessment; 2. By introducing a cross-drift compensation mechanism based on local temperature and humidity physical feedback, and using a multivariate nonlinear regression model to apply real-time acquired temperature and humidity readings as dynamic compensation factors, the gas sensor output is calibrated point by point. This compensation strategy effectively suppresses baseline drift and cross-interference caused by moisture adsorption or temperature fluctuations under high humidity and variable temperature environments, significantly improving the sensor's measurement stability and data reliability throughout the cold chain process. 3. An adaptive sampling control mechanism is constructed by calculating the rate of change of the target gas concentration in real time and using this metabolic characteristic as a trigger condition. When the rate of change of concentration exceeds a preset threshold, the system automatically increases the sampling frequency and adjusts the duty cycle of the RFID tag to ensure that key physiological events (such as respiratory pulsations and ethylene release) are captured in a timely manner. In a steady-state environment, a low-power inspection mode is maintained, which significantly extends the working life of the sensor node and resolves the contradiction between "high power consumption" and "risk of missed detection" in traditional fixed sampling strategies. 4. By adopting a topological spatial isolation layout, the heating elements and sensitive sensing units are arranged in separate zones, reducing thermal crosstalk and electromagnetic interference at the physical structure level, and improving the purity and consistency of signal acquisition. This design provides high-quality raw input for subsequent data fusion and feature extraction, and enhances the system's robustness in complex cold chain environments; 5. The compensated multimodal fusion data can be accessed through a cloud-based digital twin model. Combined with mechanistic models and machine learning algorithms, it enables forward-looking predictions of future quality evolution trends in fruits and vegetables (such as water loss rate, decrease in firmness, and probability of spoilage). Furthermore, the system can trigger blockchain smart contracts based on the prediction results to automatically generate logistics redirection or promotional scheduling instructions, promoting a leap from passive monitoring to proactive control in cold chain management, resulting in significant economic and social benefits. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process architecture of the present invention. Detailed Implementation

[0021] To facilitate a clear understanding of the technical means, inventive features, objectives, and effects of this invention, the following description is provided in conjunction with... Figure 1 The present invention will be further described below.

[0022] The physical carrier for implementing the method described in this invention is a miniature hardware system capable of conformal fitting and long-term operation in the harsh environment of the cold chain. This invention employs a flexible multimodal sensor array as the edge sensing node.

[0023] To address the path length error caused by the inability of traditional rigid sensors to conform to the curved surfaces of fruits and vegetables, which can result in a dead volume error of up to 15%, this embodiment uses polyimide or polydimethylsiloxane as a flexible substrate. In a preferred embodiment, a polyimide film with a thickness of 50 micrometers is selected.

[0024] This embodiment employs strict thermal and electromagnetic isolation in its hardware layout design. The microcontroller (MCU) and wireless RF antenna generate Joule heat during operation. If the gas sensor is placed too close to the MCU, the temperature rise will severely interfere with the baseline of the miniature thermometer and the electrochemical gas sensor.

[0025] Therefore, on a flexible polyimide substrate, a microcontroller, an RF matching network, and a miniature flexible patch antenna are deployed in the outer edge region. Between the edge region and the center region, a 2-millimeter-wide annular heat-insulating trench is etched on the substrate using laser etching technology to block heat conduction by utilizing the low thermal conductivity of air. In the core sensing region, high-precision miniature thermometers, thin-film capacitive hygrometers, and chemical resistivity cross-arrays sensitive to ethylene and carbon dioxide are deployed in a ring or array.

[0026] To achieve accurate monitoring of metabolic gases in the microenvironment of fruits and vegetables, the sensor's sensitive material must possess high selectivity. For carbon dioxide sensors, polyethyleneimine-functionalized single-walled carbon nanotubes are used. Liquid precursors are deposited on a substrate with interdigitated electrodes via electrospinning or aerosol printing. When carbon dioxide is present in the environment, the primary and secondary amine groups in polyethyleneimine undergo a reversible acid-base neutralization reaction with carbon dioxide to form urethane esters, altering the local charge distribution of the single-walled carbon nanotubes and thus causing a change in macroscopic resistance.

[0027] For ethylene sensors, nano-metal oxides or carbon nanotubes doped with specific metal-organic frameworks are used. The π electrons in the ethylene molecule can coordinate with the unsaturated metal centers in the metal-organic framework, resulting in a measurable impedance drift.

[0028] The biochemical visual indicator monitoring step relies on specific smart hydrogel materials. A 100-micron-thick layer of polyvinyl alcohol / chitosan composite hydrogel is spin-coated or drop-coated onto a localized area of ​​the top layer of the flexible sensor array. Natural anthocyanins extracted from purple cabbage or black goji berries are incorporated as indicators during the hydrogel preparation process.

[0029] When fruits and vegetables rot, they release volatile alkaline nitrogen such as ammonia or trimethylamine. The alkaline gas permeates the hydrogel polymer network, causing the pH value of the microenvironment to rise. When the pH value changes from acidic to alkaline, the molecular structure of anthocyanins will rearrange and change the absorption spectrum, resulting in a significant naked-eye visible color change from purplish-red to blue-green or even yellow. This window provides terminal sorting personnel with zero-power visual interpretation.

[0030] Cold chain logistics have long cycles, and relying solely on miniature button batteries results in extremely short lifespans for high-frequency sampling and wireless transmission. This embodiment introduces a micro-energy harvesting unit. For piezoelectric energy harvesting, a piezoelectric material layer, such as a polyvinylidene fluoride piezoelectric film, is integrated into a specific area of ​​a flexible substrate. Mechanical vibrations from refrigerated trucks or ship engines cause film deformation, generating an open-circuit voltage that satisfies a specific equation. For thermoelectric energy harvesting, a miniature thermoelectric generator is integrated using the small temperature difference between the fruits / vegetables and the cold air in the cold storage environment, generating an electromotive force based on the Seebeck effect. An ultra-low-power energy harvesting power management chip with nanowatt-level quiescent current is used, capable of cold-starting and boosting input voltages as low as 100 millivolts, and outputting a stable 1.8V or 3.3V DC power at the back end to charge backup miniature supercapacitors or thin-film solid-state batteries, achieving energy self-sufficiency for the entire sensor node.

[0031] After acquiring local temperature readings, local humidity readings, and target gas level outputs, the multimodal sensing array must perform cross-drift compensation at the edge. The high humidity environment during cold chain pre-cooling and defrosting cycles causes water molecules to competitively adsorb onto the surface of the nano-sensitive material, resulting in severe nonlinear drift of the gas sensor baseline. This embodiment uses a BP neural network model optimized by a genetic simulated annealing algorithm as the edge fusion engine. The system constructs a three-layer BP neural network with 3 nodes in the input layer. The input vector consists of temperature, humidity, and the original gas level output. Before input, the microcontroller normalizes the data and maps it to the negative one to one range. The hidden layer has 8 nodes, preferably selected based on empirical formulas, and uses the Sigmoid activation function. The output layer has 1 node, and the output value is the high-confidence target gas concentration value after temperature and humidity physical feedback calibration.

[0032] Traditional backpropagation (BP) networks, employing gradient descent, are prone to getting trapped in local minima. This invention utilizes a genetic simulated annealing algorithm to globally optimize the initial weights and thresholds of the BP network during the device factory calibration phase or cloud training phase, and then distributes the optimal parameters to the edge microcontroller for permanent execution. This algorithm integrates the global search capability of genetic algorithms and the local escape capability of simulated annealing algorithms. Through encoding and population initialization, fitness function design, annealing selection operations, crossover and mutation, and the Metropolis acceptance criterion, the globally optimal network weights are output as constants and permanently stored in the microcontroller's flash memory after hundreds of iterations. In actual cold chain operation, the microprocessor only needs to perform forward multiply-add operations, resulting in extremely low computational complexity, fully meeting the low-power hardware constraints of the edge. Through this segmented processing and nonlinear multivariate regression, the system forcibly removes spurious sensor signal outputs caused by condensation, sudden increases in humidity, or temperature fluctuations, obtaining real, drift-free fruit and vegetable metabolic gas data.

[0033] In the design of IoT sensor nodes, the conflict between energy efficiency and accuracy is the core challenge. To capture instantaneous physiological changes without depleting the battery, this system abandons the traditional mechanism of fixed sampling frequency. The microprocessor maintains a sliding data window of a specific length in memory. To suppress the interference of high-frequency shot noise from the sensor on derivative calculations, a central difference method with Savitzky-Golay smoothing filtering characteristics is used to calculate the current concentration change rate. The true step rate of the target gas is then calculated using this formula. Postharvest physiology of fruits and vegetables shows that climacteric fruits and vegetables experience exponential respiratory jumps in ethylene release and respiration rate during late maturity or when subjected to mechanical damage or fungal infection.

[0034] The system incorporates state machine-based adaptive scheduling logic within the microcontroller. During the stable period of cold chain storage and transportation, the concentration change rate is at an extremely low level. The microcontroller controls the gas sensor array to enter a power-off or extremely low duty cycle state, with only the low-power thermometer waking up the microcontroller every 30 minutes. If the temperature fluctuation is less than 0.5 degrees Celsius, the system remains in sleep mode. In this state, the overall average current consumption of the system is less than 10 microamps. When the concentration change rate calculated by the microcontroller exceeds the preset jump threshold, or when the micro-thermometer detects a drastic temperature jump due to a refrigeration unit malfunction, the system determines that a drastic change in the microenvironment has occurred or that the fruits and vegetables have entered a respiration jump period. At this time, an edge hardware interrupt is automatically triggered, the microcontroller pulls the GPIO pin high, and the heater and bias circuit of the high-power gas sensor array are activated. The system sampling frequency is dynamically increased, for example, from once every 30 minutes to once every 1 minute.

[0035] Meanwhile, the surge in data output demands higher data throughput. If a node includes an active RF transmission module, the microcontroller will dynamically shorten the broadcast interval; if the node uses passive or semi-active UHF RFID, the microcontroller will adjust the sensor data buffer write frequency connected to the RFID tag chip, increasing the duty cycle of backscatter modulation. This closed-loop mechanism, where physical and biochemical characteristics directly drive the IoT's underlying hardware control registers, ensures that the system only consumes precious energy for high-frequency monitoring when critical changes occur in the vital signs of fruits and vegetables, achieving a perfect balance between energy consumption and data granularity.

[0036] Because ocean-going refrigerated containers or refrigerated trucks are mostly made of corrugated steel plates or aluminum alloys, and filled with polyurethane foam insulation, they form a perfect Faraday cage. External 5G or 4G base station signals cannot penetrate the metal container. At the same time, the densely stacked fruits and vegetables inside, with a moisture content of over 80%, will cause severe absorption and attenuation of high-frequency radio frequency signals of 2.4GHz and above. This is a well-known communication blind spot in the industry. This invention adopts a physical system separation architecture to establish a two-level relay communication network. An edge aggregation node, powered by a large-capacity battery or DC power from the vehicle or ship hull, is deployed at the top inside the container. Hundreds or thousands of flexible multimodal sensor arrays inside the container communicate with the aggregation node using UHF RFID technology. The diffraction penetration capability of the UHF band among water-containing goods is superior to that of 2.4GHz Bluetooth or Wi-Fi.

[0037] To address the air interface collision issue caused by simultaneous data reporting from numerous sensor nodes, the aggregation node employs the slotted ALOHA anti-collision algorithm from the EPCGen2 standard protocol. By sending query commands and dynamically adjusting the number of time slots, the throughput reaches its theoretical limit. Upon receiving data, the aggregation node does not directly transmit it but utilizes edge computing power for feature-level compression. For example, it uses principal component analysis or simple incremental coding to eliminate redundant time series, compressing a multi-megabyte data stream into a critical state vector of several hundred kilobytes. This invention provides two highly engineering-feasible cross-metal shielded transmission architectures. One is a physical wire relay mechanism, utilizing pre-reserved holes in the standard structure of refrigerated containers. An armored physical wire passes through these holes, connecting the aggregation node inside the container to a communication gateway installed outside the container via an RS-485 bus or CAN bus.

[0038] Another technology is low-frequency magnetic coupling. If the container cannot tolerate any physical perforation that could damage the insulation layer, the system uses near-field magnetic induction coupling at 125 kHz or lower. A planar helical coil is adsorbed at corresponding positions on the inner and outer walls of the container. The low-frequency magnetic field has extremely strong penetrating power against non-ferromagnetic wall materials and moisture. Data transmission and exchange between the inner and outer nodes are achieved by modulating the amplitude of the low-frequency carrier wave. When a container-carrying ocean-going freighter enters international waters or a truck enters a remote mountainous area, causing the external gateway to lose satellite or cellular network signals, the system triggers an abnormal interruption. The external gateway sends a command to the aggregation node inside the container to enter offline mode, and the aggregation node automatically converts into an independent data logger. The microcontroller inside the aggregation node uses an onboard high-precision real-time clock to accurately timestamp each received sensor data packet and sequentially appends it to a large-capacity non-volatile memory to form a circular buffer.

[0039] When the logistics carrier re-enters the port or an area with network coverage, the gateway reconnects to the cloud server and triggers the breakpoint resume handshake protocol. The gateway and the cloud server synchronize their network time protocol, and then query the breakpoint identifier of the last successful upload record in the aggregation node. The aggregation node packages and sends all historical data blocks cached during the offline period in chronological order according to the timestamps and performs CRC32 verification. After receiving the data, the cloud server reconstructs the data sequence using an interpolation algorithm based on the absolute timestamp attached to the data packet. This ensures the absolute continuity of the physical cold chain and the information cold chain over the multinational transportation timeline that lasts for months, as well as the integrity of legal traceability.

[0040] Traditional cold chain monitoring is passive, triggering alarms only when temperatures exceed limits, by which time irreversible quality degradation of fruits and vegetables may have already occurred. This invention aims to transform post-event alarms into pre-event predictions through data fusion. The system establishes a digital twin model of the fruit and vegetable cold chain on a cloud server, using high-confidence multimodal fusion data after front-end compensation as input. Fruit and vegetable quality degradation is primarily driven by respiration and its resulting enzymatic biochemical reactions. The cloud-based digital twin model uses the Michaelis-Menten equations under unsteady-state conditions to model the respiration rate of fruits and vegetables, calculating oxygen consumption and carbon dioxide generation rates. Temperature has a significant impact on the maximum respiration rate and the Michaelis constant; the system uses the Arrhenius equation for dynamic correction. Based on the above respiration model and combined with a water transpiration kinetic model, the digital twin can output real-time evolution curves of core state variables representing fruit and vegetable quality over time by solving a system of partial differential equations, such as a model of the degree of hardness decline. By projecting timelines into the future, digital twins can quantify the probability distribution of a batch of fruits and vegetables within a set timeframe, including a water loss rate exceeding 5%, a decrease in firmness leading to loss of commercial value, and fungal growth causing spoilage. This forms the data foundation for high-level decision-making.

[0041] After achieving predictive quality assessment, this system further transforms information into business actions within the supply chain. This implementation step is based on the wisdom and intent dimensions of the DIKWP-TRIZ framework. Cold chain smart scheduling contracts are deployed on the consortium blockchain, with participating nodes including exporters from the country of origin, ocean carriers, customs, primary distributors, and various end-retail supermarkets.

[0042] The smart contract has strict pre-set state machine transition conditions. When the cloud-based digital twin model calculates, based on continuous real-time data, that the predicted probability of tissue softening in this batch of fruits and vegetables due to refrigeration unit fluctuations or abnormal ethylene accumulation will exceed the set safety threshold upon arrival at the destination port, the system will automatically generate a quality anomaly warning message signed by the cloud-based private key and broadcast it to the blockchain network. Upon receiving the warning message, the smart contract automatically triggers a redirection arbitration logic based on a pre-set economic model. The contract retrieves the current geographical location information, remaining voyage time, and the real-time demand and purchase price of alternative distribution nodes entered in the system, and executes an optimization objective function to maximize the remaining commercial residual value.

[0043] If calculations indicate that continuing shipments to the original high-end supermarkets will result in complete rejection and zero value, while optimizing economic loss mitigation by selling directly to secondary market fresh produce vendors upon arrival at the next port of call or transshipping to juice or jam processing enterprises near the port, the smart contract will autonomously generate an immutable transaction updating the bill of lading ownership and destination address for that batch of goods. It will then issue a physical redirection instruction to the carrier's ERP system and container control gateway via API. This entire process eliminates the need for lengthy manual negotiations and multi-party email confirmations, maximizing global business value and minimizing global food resource waste by leveraging the intelligence of digital systems before physical spoilage occurs.

[0044] Through the detailed system architecture description, underlying circuit mechanism, formula derivation, model calculation, and complete analysis of commercial operation logic, it can be seen that this invention creatively integrates materials science, electronic engineering, computer science, communication engineering, and systems engineering. Based on this, by embedding computer program instructions that execute the above algorithms and models within specific edge computing microcontrollers and cloud servers, and in conjunction with physical sensor networks, a highly rigorous, industrially implementable, and disruptive modern microenvironment adaptive monitoring system for the fruit and vegetable cold chain is constructed.

[0045] Conventional hardware replacements and software parameter adjustments made by those skilled in the art based on the technical essence of the present invention and the detailed embodiments described above do not depart from the scope of protection defined in the claims of this patent.

Claims

1. A method for monitoring the cold chain environment of fruits and vegetables based on multi-sensor fusion, characterized in that, Includes the following steps: Step S1: A flexible multimodal sensor array attached to the surface of representative fruits and vegetables from the same batch is used to acquire real-time physicochemical electrical signals of the microenvironment of the fruit and vegetable skin. The signals include at least local temperature readings, local humidity readings, and target gas level outputs characterizing the respiration and ripening state of the fruits and vegetables. The flexible multimodal sensor array conformally conforms to the curvature of the fruit and vegetable surface. Step S2: Perform cross-drift compensation based on local temperature and humidity physical feedback. That is, use a microprocessor to execute a multivariate nonlinear regression model, use the local temperature reading and local humidity reading as dynamic compensation factors, calibrate the baseline drift of the target gas level output in real time, and obtain a high-confidence target gas concentration value. Step S3: Calculate the concentration change rate of the high-confidence target gas concentration value in real time. When the system detects that the concentration change rate exceeds the preset jump threshold, it automatically triggers an edge interruption and dynamically increases the sampling frequency of the flexible multimodal sensor array.

2. The method according to claim 1, characterized in that, In step S2, the multivariate nonlinear regression model is specifically a BP neural network model optimized by the genetic simulated annealing algorithm. This model performs segmented processing for different temperature ranges to achieve accurate error calibration under different harsh micro-environments.

3. The method according to claim 1, characterized in that, In step S3, the concentration change rate is the derivative. When the derivative exceeds the jump threshold or a drastic step occurs in the local temperature, the microprocessor not only increases the sampling frequency of each sensor module, but also synchronously adjusts the duty cycle of the passive RFID tag used for data transmission.

4. The method according to claim 1, characterized in that, In step S1, in order to minimize thermal crosstalk and electromagnetic interference when acquiring signals, the flexible multimodal sensing array adopts a specific topological spatial isolation layout: the heat-generating microcontroller circuit is placed on the outer edge of the substrate, and the high-precision micro thermometer, thin-film capacitive hygrometer, and gas-sensitive microarray coated with polyethyleneimine and single-walled carbon nanotubes are arranged in a ring at a specific interval in the central region.

5. The method according to claim 1, characterized in that, The method also includes a biochemical visual indicator monitoring step: the surface of the flexible multimodal sensor array is integrated with a hydrogel breathable window, which contains natural anthocyanins that are sensitive to volatile alkaline gases or sudden changes in environmental pH. When fruits and vegetables rot, they produce color changes through the color development layer to provide an intuitive naked-eye visual interpretation signal.

6. The method according to claim 1, characterized in that, After obtaining the high-confidence target gas concentration value, data is transmitted through a system separation and anti-attenuation relay fusion network. Specifically, the flexible multimodal sensor array uses ultra-high frequency RFID technology to exchange data over short distances with edge aggregation nodes arranged inside the cold chain container. After the edge aggregation nodes complete data cleaning and feature-level compression, they use low-frequency magnetic coupling technology or physical wires penetrating the container to transmit the data across the sealed metal wall of the container to a communication gateway outside the container. The communication gateway then transmits the data to a cloud server through a cellular network.

7. The method according to claim 6, characterized in that, When the logistics vehicle travels to an area with no cellular network signal, the edge aggregation node automatically switches to independent data logger mode, relying on the local clock to continuously store multi-source sensor readings; once the network signal is recaptured, the breakpoint resume mechanism and timestamp fusion algorithm are activated to ensure the integrity of the data on the timeline.

8. The method according to any one of claims 1 to 7, characterized in that, In the process of acquiring the real-time physical, chemical and electrical signals, the system utilizes the piezoelectric effect generated by the minute mechanical vibrations in the cold chain logistics environment, or the thermoelectric effect generated by the temperature difference inside and outside the cold storage, to continuously replenish the micro-electrical energy of the flexible multimodal sensing array through a multi-source energy harvesting microsystem.

9. The method according to claim 1, characterized in that, The method also includes a proactive predictive quality control step: the compensated multimodal fusion data is uploaded to the cloud-based digital twin model of fruits and vegetables in real time, and combined with real-time machine learning predictive analysis and the enzyme respiration rate calculation model characterized by the Michaelis equation, the water loss rate, hardness reduction and spoilage probability of the monitored fruits and vegetables within a set time in the future are predicted in a forward-looking manner.

10. The method according to claim 9, characterized in that, The method also includes a smart scheduling step based on the prediction results: when the cloud-based digital twin model of fruits and vegetables predicts that the probability of softening or spoilage of fruit and vegetable tissues exceeds the set safety threshold, the system automatically triggers a smart contract deployed on the blockchain to generate dynamic decision instructions, and redirects the batch of goods in real time to the nearest retail terminal for price reduction and promotion, or directly transfers them to the processing plant.

Images