A method, system, device, and medium for predicting risk of packaged food

Abstract

The application relates to the technical field of food safety monitoring. A packaged food risk prediction method, system, device and medium are provided, wherein the method comprises the following steps: before transmitting food state parameters, a pre-defined probe sequence is transmitted to an external receiving end to obtain a distorted probe sequence transmitted through a packaged-food medium; based on the distorted probe sequence transmitted through the packaged-food medium, machine learning inversion is implemented to generate time-varying channel fingerprint features; the time-varying channel fingerprint features are mapped to a preset sensor signal transmission channel model; based on the sensor signal transmission channel model, sensor signals containing food state parameters are dynamically compensated to obtain corrected food state data; the corrected food state data is subjected to risk assessment, and a packaged food risk prediction result is output, so that the technical effects of improving dynamic channel modeling accuracy, improving multipath-medium coupling interference separation effect and enhancing risk assessment data reliability are achieved.

Description

Technical Field

[0001] This invention relates to the field of food safety monitoring technology, and in particular to a method, system, equipment and medium for risk prediction of packaged food. Background Technology

[0002] With the widespread application of IoT technology in the food supply chain, wireless sensing monitoring inside packaging has become a core technology for ensuring food safety. This technology can collect status parameters such as temperature, gas concentration, and humidity in real time, enabling early warning of food spoilage processes.

[0003] Traditional techniques often employ fixed compensation parameters to build static channel models. This modeling approach cannot adapt to the dynamic changes in dielectric properties during food spoilage, and struggles to distinguish between the static influence of packaging materials and the detection interference caused by the dynamic changes in the food's own state. During signal distortion correction, traditional frequency domain filtering methods fail to effectively decouple multipath reflections and time-varying dielectric properties, easily leading to the mixing of various phase components and causing baseline shifts in core detection parameters. Furthermore, traditional techniques directly extract food spoilage discrimination features from raw sensor data, failing to eliminate inherent noise interference from channel transmission, easily causing distortion of spoilage judgment features, and thus significantly reducing the accuracy and practical effectiveness of food spoilage risk assessment. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, system, device and medium for risk prediction of packaged food to address the above-mentioned technical problems, so as to improve the accuracy of dynamic channel modeling, improve the separation effect of multipath-medium coupling interference and enhance the reliability of risk assessment data.

[0005] In a first aspect, this application provides a method for predicting the risks of packaged food, the method comprising:

[0006] Data is collected using wireless sensors integrated within the packaging to obtain food status parameters;

[0007] Before transmitting food status parameters, a predefined probe sequence is transmitted to an external receiver to obtain a distorted probe sequence after transmission through the packaging-food medium.

[0008] Based on the distorted probe sequence transmitted through the packaging-food medium, machine learning inversion is performed to generate time-varying channel fingerprint features;

[0009] The time-varying channel fingerprint features are mapped to a preset sensor signal transmission channel model; the sensor signals containing food state parameters are dynamically compensated based on the sensor signal transmission channel model to obtain the corrected food state data.

[0010] Perform a risk assessment on the corrected food condition data and output the risk prediction results for packaged foods.

[0011] In one embodiment, based on the distorted probe sequence transmitted through the packaging-food medium, machine learning inversion is performed to generate time-varying channel fingerprint features, including:

[0012] The distorted probe sequence after transmission through packaging-food media is preprocessed to extract multidimensional signal distortion features;

[0013] Multidimensional signal distortion characteristics are analyzed using convolutional neural networks to identify multipath propagation modes;

[0014] Based on radial basis function network analysis of multidimensional signal distortion characteristics, dynamic medium variation components are captured;

[0015] By integrating multipath propagation modes and dynamic medium variation components, time-varying channel fingerprint features are generated.

[0016] In one embodiment, the method of capturing dynamic medium variation components by analyzing multidimensional signal distortion characteristics based on radial basis function networks includes:

[0017] By mapping the kernel function of the radial basis function network, the multidimensional signal distortion characteristics are transformed into a nonlinear separable space;

[0018] Identifying nonlinear patterns characterizing changes in food state within a nonlinearly separable space;

[0019] Separate the time-varying medium response component and the static medium response component in the nonlinear mode;

[0020] The time-varying medium response component is taken as the dynamic medium change component.

[0021] In one embodiment, sensor signals containing food state parameters are dynamically compensated based on a sensor signal transmission channel model to obtain corrected food state data, including:

[0022] The amplitude of the sensor signal is compensated by using the signal attenuation factor in the sensor signal transmission channel model to obtain the amplitude-compensated signal.

[0023] The demodulated baseband signal is obtained by demodulating the amplitude-compensated signal.

[0024] Based on the phase distortion mode in the sensor signal transmission channel model, the demodulated baseband signal is phase-corrected to generate a phase-corrected signal.

[0025] Decode the phase-corrected signal to obtain the corrected food state data.

[0026] In one embodiment, a risk assessment is performed on the corrected food condition data, and the packaged food risk prediction result is output, including:

[0027] A multi-dimensional spoilage feature vector was extracted from the corrected food condition data. The multi-dimensional spoilage feature vector includes temperature change gradient, gas concentration slope and humidity accumulation.

[0028] By analyzing multi-dimensional corruption feature vectors using a pre-defined spatiotemporal correlation model, a corruption process prediction index is obtained.

[0029] The critical state of corruption process prediction indicators is assessed based on the dynamic risk threshold algorithm to determine the risk level parameters.

[0030] By mapping the risk level parameters to a preset food shelf-life decay curve, the risk prediction results for packaged foods are obtained.

[0031] In one embodiment, phase correction is performed on the demodulated baseband signal based on the phase distortion mode in the sensor signal transmission channel model to generate a phase-corrected signal, including:

[0032] Analyze the phase distortion mode and extract the phase shift component caused by multipath propagation and the phase jitter component caused by medium changes;

[0033] A joint compensation function is constructed based on the phase offset component and the phase jitter component;

[0034] The phase spectrum of the demodulated baseband signal is corrected by applying a joint compensation function to generate the corrected phase spectrum;

[0035] The time-domain signal corresponding to the corrected phase spectrum is reconstructed to obtain the phase-corrected signal.

[0036] In one embodiment, decoding the phase-corrected signal to obtain corrected food state data includes:

[0037] The phase-corrected signal is then subjected to soft-decision decoding to obtain the decoded data frame;

[0038] Verify the integrity of the decoded data frame and extract the payload data;

[0039] The sensor measurements in the payload data are analyzed to obtain the corrected food condition data.

[0040] Secondly, this application also provides a packaged food risk prediction system, the system comprising:

[0041] The data acquisition module is used to collect data through wireless sensors integrated within the packaging to obtain food status parameters;

[0042] The channel detection module is used to transmit a predefined probe sequence to an external receiver before transmitting food status parameters, and to obtain a distorted probe sequence after transmission through the packaging-food medium.

[0043] The feature inversion module is used to perform machine learning inversion based on the distorted probe sequence transmitted through the packaging-food medium to generate time-varying channel fingerprint features;

[0044] The dynamic compensation module is used to map time-varying channel fingerprint features to a preset sensor signal transmission channel model; based on the sensor signal transmission channel model, it dynamically compensates the sensor signals containing food state parameters to obtain corrected food state data.

[0045] The risk assessment module is used to assess the risk of corrected food status data and output the risk prediction results for packaged foods.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0047] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0048] This application provides a method, system, device, and medium for risk prediction of packaged food. The method involves collecting food state parameters using wireless sensors integrated within the packaging, and transmitting a predefined probe sequence to an external receiver before transmitting these parameters. This acquires a distorted probe sequence after transmission through the packaging-food medium. Based on this distorted probe sequence, machine learning inversion is performed to generate time-varying channel fingerprint features, enabling channel modeling to adapt to the dynamic changes in the medium and thus improving the accuracy of dynamic channel modeling. The time-varying channel fingerprint features are mapped to a preset sensor signal transmission channel model. This model is used to dynamically compensate sensor signals containing food state parameters, effectively distinguishing the interference effects caused by multipath propagation and medium changes, thereby improving the separation effect of multipath-medium coupling interference.

[0049] The corrected food status data obtained after dynamic compensation processing eliminates interference generated during channel transmission. Based on this data, risk assessment of packaged food is conducted and prediction results are output. This reduces the impact of invalid interference factors on the assessment process, helps enhance the reliability of risk assessment data, and provides strong support for predicting the risks of packaged food. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of a risk prediction method for packaged food according to one embodiment of the present invention;

[0052] Figure 2 A flowchart of a phase-corrected signal is generated by performing phase correction on the demodulated baseband signal based on the phase distortion mode in the sensor signal transmission channel model in one embodiment of the present invention.

[0053] Figure 3 This is a structural diagram of a packaged food risk prediction system according to one embodiment of the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0055] First, the application scenarios of the embodiments of this application are described. The embodiments of this application provide methods, systems, equipment, and media for risk prediction of packaged food applicable to, but not limited to, scenarios such as cold chain transportation, warehousing and inventory management, and offline supermarket display and sales.

[0056] In illustrative purposes, the packaged food risk prediction method, system, equipment and medium provided in this application embodiment can also be applied to application scenarios such as fresh food sealed storage, pre-made food turnover and distribution, and bulk storage of snack foods. This is only an example and does not limit the specific application scenarios.

[0057] like Figure 1 As shown, this application provides a method for predicting the risk of packaged food, the method comprising:

[0058] S101: Data is collected through wireless sensors integrated within the packaging to obtain food status parameters.

[0059] For example, the packaged food risk prediction terminal integrates wireless sensors into preset locations inside the food packaging, ensuring a reasonable distance between the wireless sensors and the food to meet monitoring needs and guarantee the validity of the sensed data. The packaged food risk prediction terminal sends a start command to the integrated wireless sensor, triggering the sensor's sensing function and putting it into a continuous and stable working state.

[0060] When operational, the wireless sensor continuously senses physical and chemical quantities in the surrounding environment that are directly related to the food's condition. The sensor converts these quantities into corresponding raw electrical signals, which are then preliminarily conditioned to filter out irrelevant interference. The signal processing unit of the packaged food risk prediction terminal receives the conditioned electrical signals, performs targeted analysis and format conversion, and outputs the processed results through a preset data transmission path to obtain the food's condition parameters.

[0061] S102: Before transmitting food status parameters, a predefined probe sequence is transmitted to an external receiver to obtain a distorted probe sequence after transmission through the packaging-food medium.

[0062] For example, before transmitting food status parameters, the packaged food risk prediction terminal performs a comprehensive self-check on its signal transmission system. After confirming there are no transmission conflicts and the equipment is operating stably, it retrieves a pre-stored predefined probe sequence and performs an integrity check on the sequence. Once it confirms that the data is complete, undamaged, and meets the transmission standards, it activates its own signal transmission unit. The packaged food risk prediction terminal then transmits the verified predefined probe sequence towards the external receiving end according to the preset transmission parameters, transmission protocol, and signal strength standards.

[0063] After being emitted from the packaged food risk prediction terminal, the predefined probe sequence sequentially penetrates the surface of the food packaging and passes through the interior of the food itself, continuously propagating along the transmission path formed by the packaging and food media. During propagation, the physical characteristics of the food packaging, the media properties of the food, and environmental interference in the transmission path work together to alter the original signal structure, phase characteristics, and amplitude parameters of the predefined probe sequence, resulting in natural distortion of the sequence signal.

[0064] The external receiver captures the distorted sequence signal transmitted through the packaging-food medium in real time. After preliminary demodulation processing, the signal is fed back to the packaged food risk prediction terminal through a preset feedback channel. Upon receiving the distorted sequence signal from the external receiver, the packaged food risk prediction terminal performs a matching verification of the signal source. After confirming that the signal uniquely corresponds to the predefined probe sequence transmitted and that there has been no transmission loss, the terminal completes signal morphology identification and calibration through its internal signal processing flow, obtaining the distorted probe sequence transmitted through the packaging-food medium.

[0065] S103: Based on the distorted probe sequence after transmission through the packaging-food medium, machine learning inversion is performed to generate time-varying channel fingerprint features.

[0066] For example, after obtaining the distorted probe sequence transmitted through the packaging-food medium, the packaged food risk prediction terminal performs signal preprocessing to remove environmental interference signals and transmission noise in the sequence, and standardize the signal format of the sequence itself so that the processed data conforms to the access specifications of machine learning inversion.

[0067] The packaged food risk prediction terminal retrieves the machine learning inversion model built into the device and feeds the normalized distortion probe sequence into the model. The model then analyzes the signal distortion trend and propagation state changes reflected in the sequence, extracting various multi-dimensional features that reflect the actual channel conditions, including signal distortion-related features, medium response-related features, and multipath propagation-related features.

[0068] All effective features have been extracted. The packaged food risk prediction terminal relies on the feature fusion mechanism built into the model to sort out the inherent relationship between different features and complete the integration process. Redundant feature content that has no reference value is filtered out, highlighting the core change law formed by the channel following the change of medium state, and generating time-varying channel fingerprint features.

[0069] S104: Map the time-varying channel fingerprint features to a preset sensor signal transmission channel model; dynamically compensate the sensor signals containing food state parameters based on the sensor signal transmission channel model to obtain corrected food state data.

[0070] For example, the packaged food risk prediction terminal retrieves the preset sensor signal transmission channel model stored inside the device, and matches the obtained time-varying channel fingerprint features with the model's baseline feature system to complete the mapping process between features and the channel model.

[0071] After completing the feature mapping operation, the packaged food risk prediction terminal extracts the sensor signals carrying the food state parameters and sends them to the corresponding channel model. Based on the transmission law presented by the channel model, it determines various deviation problems generated during the signal transmission stage. According to the established correction specifications of the model, it performs dynamic compensation on the sensor signals to eliminate signal distortion caused by medium propagation and multipath interference, and obtains the corrected food state data.

[0072] S105: Perform a risk assessment on the corrected food status data and output the risk prediction results for packaged foods.

[0073] For example, after obtaining the corrected food status data, the packaged food risk prediction terminal verifies the completeness and validity of the data, removes invalid abnormal information, and then retrieves the food risk assessment standards and corresponding quality requirements pre-stored in the device.

[0074] The packaged food risk prediction terminal compares the actual condition of the food with the established evaluation standards and quality requirements, analyzes the corrected food status data item by item, comprehensively judges the potential safety hazards and quality change trends of the food, summarizes all the judgment information, organizes it into standardized judgment content according to the preset format, and outputs the packaged food risk prediction results.

[0075] One embodiment of this application provides a method for risk prediction of packaged food. After collecting food state parameters through wireless sensors integrated within the packaging, a predefined probe sequence is transmitted to an external receiver before transmitting these parameters. This acquires a distorted probe sequence after transmission through the packaging-food medium. Based on this distorted probe sequence, machine learning inversion is performed to generate time-varying channel fingerprint features, enabling channel modeling to adapt to the dynamic changes in the medium and thus improving the accuracy of dynamic channel modeling. The time-varying channel fingerprint features are mapped to a preset sensor signal transmission channel model. This model is used to dynamically compensate sensor signals containing food state parameters, effectively distinguishing the interference effects caused by multipath propagation and medium changes, thereby improving the separation effect of multipath-medium coupling interference.

[0076] The corrected food status data obtained after dynamic compensation processing eliminates interference generated during channel transmission. Based on this data, risk assessment of packaged food is conducted and prediction results are output. This reduces the impact of invalid interference factors on the assessment process, helps enhance the reliability of risk assessment data, and provides strong support for predicting the risks of packaged food.

[0077] In one embodiment, based on the distorted probe sequence transmitted through the packaging-food medium, machine learning inversion is performed to generate time-varying channel fingerprint features, including:

[0078] (1) The distorted probe sequence after transmission through the packaging-food medium is preprocessed to extract multidimensional signal distortion features.

[0079] For example, the packaged food risk prediction terminal performs signal denoising processing on the distorted probe sequence after transmission through the packaging-food medium, filters out environmental noise and transmission interference signals mixed in the sequence, aligns the time axis of the denoised sequence, corrects the time delay deviation generated during the sequence propagation, and then performs normalization processing on the aligned sequence to unify the signal amplitude range of the sequence.

[0080] After preprocessing, the packaged food risk prediction terminal extracts multidimensional signal distortion features related to channel distortion from the processed sequence. The extraction process follows the mapping relationship between features and channel state to ensure that the features can characterize the distortion characteristics of the sequence.

[0081] The multidimensional signal distortion feature set includes four types of feature parameters: amplitude distortion features, phase shift features, frequency dispersion features, and time spread features.

[0082] (2) Analyze the multidimensional signal distortion characteristics through convolutional neural networks to identify multipath propagation modes.

[0083] For example, the packaged food risk prediction terminal inputs the extracted multidimensional signal distortion features into a preset convolutional neural network. The input layer of the convolutional neural network performs format conversion on the multidimensional signal distortion features to adapt them to the network's processing requirements. The convolutional layers of the convolutional neural network use convolutional kernels of preset size to perform sliding convolution operations on the feature matrix, extracting local correlation information from the features and generating a feature map with a hierarchical structure.

[0084] The pooling layers of a convolutional neural network downsample the feature maps output by the convolutional layers, reducing feature dimensionality and computational complexity while preserving key feature information. The fully connected layers integrate and classify the pooled features, identifying multipath propagation patterns formed during the packaging-food transport process based on the correspondence between features and multipath propagation modes. The convolution operation formula of the convolutional neural network is as follows:

[0085]

[0086] In the formula, For the first Convolutional layer coordinates Output feature values ​​at the location; The convolutional network layer number; These are the coordinates of the feature map position. The height of the convolution kernel. This refers to the width of the convolution kernel. For the first Layer convolution kernel weight parameters; The coordinates are inside the convolution kernel; Input feature values ​​for the corresponding positions in the previous level; These are the bias parameters for the convolutional layer.

[0087] Convolutional neural networks (CNNs) include four types of network layer structures: input layer, convolutional layer, pooling layer, and fully connected layer.

[0088] (3) Based on the radial basis function network, analyze the multidimensional signal distortion characteristics and capture the dynamic medium change components.

[0089] For example, the packaged food risk prediction terminal simultaneously inputs multidimensional signal distortion features into a preset radial basis function (RBF) network. The input layer of the RBF network receives the multidimensional signal distortion features and passes them to the hidden layer. Each neuron in the hidden layer uses a RBF as its activation function. Using a preset center vector as a reference, it calculates the Euclidean distance between the multidimensional signal distortion features and the center vector, and then maps this distance to the neuron's output value through the RBF, thus achieving a nonlinear transformation of the features.

[0090] The output layer of the radial basis function network performs a linear weighted summation of the outputs of the hidden layers. Based on the correlation between features and medium changes, the dynamic medium change component contained in the multidimensional signal distortion features is analyzed. This component can reflect the property changes of the packaging-food medium in real time during transmission. The activation function formula of the hidden layer of the radial basis function network is as follows:

[0091]

[0092] In the formula, For the hidden layer Each neuron outputs a numerical value; The sequence number of the hidden layer neurons; It is an exponential operation function; The Euclidean distance between the feature set and the center vector of the neuron; The center vector of the neuron; This is the neuron response width parameter.

[0093] The formula for the linear weighted summation of the output layer of a radial basis function network is as follows:

[0094]

[0095] In the formula, This represents the dynamic medium change component, reflecting the real-time changes in the medium's properties. This represents the total number of neurons in the hidden layer. These are the neuron connection weights; This represents the output value of a single neuron. These are the output layer bias parameters.

[0096] The radial basis function network consists of a three-layer basic network architecture: an input layer, a hidden layer, and an output layer.

[0097] (4) Integrate multipath propagation modes and dynamic medium change components to generate time-varying channel fingerprint features.

[0098] For example, the packaged food risk prediction terminal performs feature standardization processing on the multipath propagation mode and dynamic medium change component, adjusts the numerical range of the two to a unified range, and eliminates the fusion bias caused by the difference in dimensions; then, a weighted fusion strategy is adopted to assign corresponding weight coefficients to the two based on the importance of the multipath propagation mode and dynamic medium change component to the channel state representation.

[0099] The packaged food risk prediction terminal integrates the standardized multipath propagation mode and dynamic medium change components into a unified feature vector through weighted summation. This feature vector can comprehensively and in real time reflect the dynamic change state of the channel and generate time-varying channel fingerprint features.

[0100] In one embodiment, the method of capturing dynamic medium variation components by analyzing multidimensional signal distortion characteristics based on radial basis function networks includes:

[0101] (1) The multidimensional signal distortion features are transformed into a nonlinear separable space by mapping the kernel function of the radial basis function network.

[0102] For example, the packaged food risk prediction terminal retrieves a preset radial basis function network, determines the appropriate kernel function type based on the properties of the multidimensional signal distortion features, and performs a nonlinear transformation on each feature dimension of the multidimensional signal distortion features according to the mapping rules defined by the kernel function. This maps the originally linearly inseparable feature data to a higher-dimensional nonlinearly separable space, ensuring that the mapped features can fully reflect the potential correlation between changes in the packaging-food medium and signal distortion.

[0103] (2) Identify nonlinear patterns that characterize changes in the state of food in a nonlinearly separable space.

[0104] For example, relying on the high-dimensional feature data that has been mapped, the packaged food risk prediction terminal calculates the similarity between the mapped feature vectors, uses a predetermined clustering algorithm to group and classify all mapped features, and selects feature clusters that are closely related to changes in the state of the food. From the selected feature clusters, nonlinear patterns that can reflect changes in medium properties and fluctuations in signal transmission response are extracted. These patterns can reflect the dynamic relationship between the packaging-food medium and signal distortion.

[0105] Among them, the nonlinear mode includes three core types of information: the correlation law of signal characteristics caused by changes in media composition, the trend of signal response changes under the influence of packaging structure, and the characteristic distribution pattern corresponding to changes in food state.

[0106] (3) Separate the time-varying medium response component and the static medium response component in the nonlinear mode.

[0107] For example, referring to pre-stored media baseline response parameters, the packaged food risk prediction terminal delineates component distinction boundaries. Using a sliding window approach, it performs time-series analysis on nonlinear patterns, statistically analyzing the rate of change of characteristic responses within a single window. Characteristic responses with rates of change exceeding the component distinction boundaries are classified as time-varying media response components. These components change continuously over time and are directly related to changes in the food's state. Characteristic responses with rates of change not exceeding the component distinction boundaries are classified as static media response components. These components are determined by the inherent properties of the packaging-food media and do not change over time. The formula for separating time-varying and static media response components is as follows:

[0108]

[0109] In the formula, The response components of the time-varying medium; for Time and The amount of change in the time-mapping features; for The mapping feature vector at time step; for The mapping feature vector at time step; This is an indicator function that outputs 1 if the condition within the parentheses is met, and 0 otherwise. The L2 norm of the characteristic change; The component separation threshold; This refers to the static medium response component; is the mapping eigenvector in a nonlinear separable space.

[0110] Among them, the component separation threshold is a criterion used to distinguish between dynamic changes in characteristic response and static stable state, and is calibrated and set according to the fluctuation range of medium properties and signal noise level.

[0111] (4) The time-varying medium response component is taken as the dynamic medium change component.

[0112] For example, after completing the decomposition of the response components of the two types of media, the packaged food risk prediction terminal performs data verification and judgment. It compares the distribution pattern of the effective time-varying components retained in history, eliminates abnormal response data caused by environmental interference and signal noise, retains the effective time-varying components that are directly related to the changes in the state of the food, and defines the verified time-varying media response components as dynamic media change components, ensuring that the components can reflect the dynamic changes in the properties of the packaging-food media in real time and with relatively high accuracy.

[0113] In one embodiment, sensor signals containing food state parameters are dynamically compensated based on a sensor signal transmission channel model to obtain corrected food state data, including:

[0114] (1) The sensor signal is amplitude compensated by using the signal attenuation factor in the sensor signal transmission channel model to obtain the amplitude compensated signal.

[0115] For example, the packaged food risk prediction terminal retrieves the signal attenuation factor stored in the sensor signal transmission channel model, calculates the matching compensation gain value with reference to the amplitude loss law corresponding to the attenuation factor, and performs amplitude reverse correction processing sequentially along all sampling points of the signal waveform to offset the signal amplitude loss caused by the packaging-food medium and the channel path, stabilize the signal output amplitude level, and generate an amplitude-compensated signal.

[0116] Among them, the signal attenuation factor can reflect the degree of amplitude attenuation of the signal in the channel transmission path, which is consistent with the actual variation law of medium transmission loss.

[0117] (2) The demodulated baseband signal is obtained by demodulating the amplitude-compensated signal.

[0118] For example, the packaged food risk prediction terminal calls the built-in signal demodulation program to identify the carrier frequency band and signal modulation type of the signal after amplitude compensation. It strictly follows the established demodulation protocol to peel off the carrier waveform superimposed on the outside of the basic signal layer by layer, filters out the carrier noise interference remaining in the process, organizes the timing arrangement structure of the basic signal, and separates the basic waveform content that has not been carrier modulated, thereby obtaining the demodulated baseband signal.

[0119] (3) Based on the phase distortion mode in the sensor signal transmission channel model, the demodulated baseband signal is phase corrected to generate a phase corrected signal.

[0120] For example, based on the demodulated baseband signal content, the packaged food risk prediction terminal extracts the phase distortion pattern recorded by the sensor signal transmission channel model, compares it with the phase distribution reference curve of the distortion-free standard baseband signal, calculates the actual phase offset difference of the demodulated baseband signal at each sampling position, applies the corresponding phase compensation amount according to the deviation correction logic corresponding to the phase distortion pattern, gradually eliminates various phase distortion problems caused by the channel transmission link, normalizes the signal phase state, and generates a phase-corrected signal.

[0121] Among them, the phase distortion mode records the characteristics and distortion patterns of signal phase offset during channel transmission.

[0122] (4) Decode the phase-corrected signal to obtain the corrected food state data.

[0123] For example, the packaged food risk prediction terminal calls the signal decoding specification that matches the signal encoding format, splits the phase-corrected signal into independent encoded segments according to the data segmentation boundaries defined by the specification, performs content parsing on each of the split encoded segments, extracts the various food state-related parameter information carried within the segments, and systematically summarizes all the effective parameter content obtained from the parsing to form a complete data set, thus obtaining the corrected food state data.

[0124] Among them, the signal decoding specification defines the encoding and splitting rules and the basis for analyzing and judging food state parameters.

[0125] In one embodiment, a risk assessment is performed on the corrected food condition data, and the packaged food risk prediction result is output, including:

[0126] (1) Extract multi-dimensional spoilage feature vectors from the corrected food state data. The multi-dimensional spoilage feature vectors include temperature change gradient, gas concentration slope and humidity accumulation.

[0127] For example, the packaged food risk prediction terminal retrieves all corrected food status data and, according to the calculation rules corresponding to food spoilage characteristics, solves for three types of feature parameters for the data sequence: temperature change gradient, gas concentration slope, and humidity accumulation. The temperature change gradient is obtained by calculating the temperature difference between consecutive time nodes, the gas concentration slope is solved based on the temporal change trend of the concentration of spoilage-related gas components, and the humidity accumulation is obtained by summing the humidity data within a specific time period. The three types of parameters are integrated and grouped in a fixed order to construct a complete and standardized multi-dimensional spoilage feature vector.

[0128] Among them, the multi-dimensional corruption feature vector relies on environmental temperature and humidity and internal gas composition data to intuitively reflect the basic changes in the evolution of food spoilage.

[0129] (2) By analyzing multi-dimensional corruption feature vectors through a pre-set spatiotemporal correlation model, a corruption process prediction index is obtained.

[0130] For example, after completing the feature vector construction, the packaged food risk prediction terminal inputs the obtained multi-dimensional spoilage feature vector into a preset spatiotemporal correlation model. This model combines the regional distribution characteristics of food storage space to analyze the impact of temperature, humidity, and gas concentration distribution differences in different spatial locations on the spoilage process. Then, based on the evolution law over time, it captures the fluctuation characteristics of feature parameters in different time periods. Through the temporal correlation algorithm, it mines the linkage relationship between temperature change gradient, gas concentration slope, and humidity accumulation. It comprehensively calculates the coordinated fluctuation trend of the three types of feature parameters generated by spatiotemporal conditions, and deduces the development trajectory and subsequent evolution direction of food spoilage from the initial stage to the current state, and calculates the spoilage process prediction index.

[0131] Among them, the pre-set spatiotemporal correlation model relies on the dual dimensions of spatial location and temporal sequence to depict the linkage and evolution of food spoilage characteristics.

[0132] (3) Evaluate the critical state of corruption process prediction indicators based on dynamic risk threshold algorithm and determine risk level parameters.

[0133] For example, a basic risk threshold range is preset based on the characteristics of food categories (such as fresh food and processed food). Then, the basic threshold range is dynamically corrected by combining real-time storage environment parameters (temperature, humidity, ventilation conditions) to determine the dynamic threshold range that is suitable for the current scenario. After that, the upper and lower limits of the dynamic threshold range are continuously compared with the spoilage process prediction indicators to determine whether the indicators exceed the safety threshold, are in a critical state, or are within the safety range. Based on the degree of deviation between the indicators and the thresholds, high, medium, and low levels are divided. The risk weight corresponding to each level is calculated in combination with the food shelf life benchmark value to determine the corresponding risk level parameters.

[0134] Among them, the dynamic risk threshold algorithm has a built-in environmental parameter response mechanism and food category adaptation unit, which can autonomously adjust the judgment boundary according to the real-time changes of the external storage environment. It can dynamically optimize the threshold range through the threshold correction coefficient, and adapt to the spoilage characteristics and safety standards of different food categories, so as to ensure that a relatively accurate and reasonable risk assessment can be achieved in a variety of scenarios.

[0135] (4) Map the risk level parameters to the preset food shelf life decay curve to obtain the risk prediction results of packaged food.

[0136] For example, after determining the risk level parameter value, the packaged food risk prediction terminal retrieves the preset food shelf-life decay curve stored in the system. This curve is stored according to food category. The curve for each category is generated by fitting a large number of spoilage acceleration experiments and actual storage monitoring data. Then, it retrieves the horizontal coordinate point in the curve coordinate system that matches the current risk level parameter, reads the remaining shelf life data of the vertical coordinate corresponding to the point, and completes the parameter mapping conversion based on the shelf-life decay rate and decay ratio corresponding to the point. It integrates the decay information, remaining shelf life information and risk level information obtained by conversion to form a comprehensive judgment result that includes risk level, remaining shelf life and spoilage development trend, and obtains the packaged food risk prediction result.

[0137] The preset food shelf-life decay curve uses the risk level parameter as the horizontal axis and the remaining shelf life as the vertical axis. It includes exclusive curve branches for different food categories. Each curve is marked with the critical node of shelf-life decay corresponding to different risk levels, which can map the quantitative relationship between risk level and shelf-life decay, and provide scientific and reliable data support for risk prediction results.

[0138] like Figure 2 As shown, based on the phase distortion mode in the sensor signal transmission channel model, phase correction is performed on the demodulated baseband signal to generate a phase-corrected signal, including:

[0139] S201: Analyze the phase distortion mode and extract the phase shift component caused by multipath propagation and the phase jitter component caused by medium changes.

[0140] For example, the packaged food risk prediction terminal retrieves the phase distortion pattern file stored inside the sensor signal transmission channel model, disassembles the phase distortion evolution law recorded in the pattern layer by layer, divides the signal transmission influencing factors and the medium state influencing factors according to the root cause of distortion, counts the phase deviation caused by different propagation paths and the phase jitter caused by medium property fluctuations, and separates and extracts the phase offset component caused by multipath propagation and the phase jitter component caused by medium changes.

[0141] S202: Construct a joint compensation function based on the phase offset component and the phase jitter component.

[0142] For example, based on the two types of phase distortion components that have been separated, the packaged food risk prediction terminal combines the mechanism of phase distortion superposition, matches a linear compensation adjustment coefficient for the fixed characteristics of the phase offset component, matches an adaptive smoothing compensation coefficient for the random characteristics of the phase jitter component, incorporates the linear operation logic of fixed offset correction and the nonlinear filtering criterion of random jitter smoothing, and constructs a joint compensation function that can simultaneously cancel the dual phase distortion problem by weighted correlation of the compensation operation relationship of the two types of components.

[0143] The joint compensation function adopts a frequency domain adaptive structure, which has the dual function of synchronously correcting fixed phase deviation and random phase jitter. The compensation coefficient can be dynamically adjusted according to the signal frequency characteristics. The linear unit is used to offset the fixed phase offset caused by multipath propagation, and the nonlinear smoothing unit is used to suppress random phase jitter caused by medium changes, ensuring that the phase distortion of different frequency components can be corrected.

[0144] S203: Apply the joint compensation function to correct the phase spectrum of the demodulated baseband signal and generate the corrected phase spectrum.

[0145] For example, the completed joint compensation function is applied to the frequency domain data system of the demodulated baseband signal. The packaged food risk prediction terminal traverses all frequency points in the phase spectrum in order of signal frequency from low to high, extracts the original phase value corresponding to each frequency point, substitutes it into the joint compensation function to complete the targeted correction calculation, eliminates the residual phase shift and phase jitter defects in each frequency domain one by one, ensures the continuous phase transition between frequency points through smoothing processing, summarizes the corrected phase values ​​of all frequency points, and combines them to generate a complete and smooth corrected phase spectrum.

[0146] Among them, the phase spectrum is the core representation of the frequency domain characteristics of the signal, with the horizontal axis representing the signal angular frequency and the vertical axis representing the phase value of the corresponding frequency component.

[0147] S204: Reconstruct the time-domain signal corresponding to the phase spectrum after correction to obtain the phase-corrected signal.

[0148] For example, using standard algorithms for frequency-domain and time-domain conversion, the packaged food risk prediction terminal relies on the generated corrected phase spectrum as its core. It retrieves the amplitude spectrum information inherent in the demodulated baseband signal, performs zero-padding expansion on the combined complex spectrum data to improve conversion accuracy, applies a window function to suppress spectral leakage, performs a discrete inverse Fourier transform, and rearranges the transformed discretized data according to time sequence to reconstruct a coherent, complete, and phase-distortion-free time-domain waveform, obtaining the phase-corrected signal. The time-domain signal reconstruction formula is as follows:

[0149]

[0150] In the formula, This is the time-domain signal after phase correction; Represents the inverse Fourier transform operation; This refers to the amplitude spectrum of the baseband signal. The imaginary unit; The corrected phase spectrum; It is a time variable.

[0151] Among them, the frequency domain to time domain conversion algorithm specifically refers to the Discrete Inverse Fourier Transform (IDFT) or Fast Inverse Fourier Transform (IFFT) algorithm. Its core function is to realize the switching and restoration of the signal between the complex spectrum in the frequency domain and the waveform in the time domain. During the conversion process, operations such as spectrum zero-padding and window function optimization are used to ensure that the timing characteristics and amplitude ratio of the time domain signal are consistent with the original undistorted signal, avoid introducing new signal distortion during the conversion process, and ensure the integrity and accuracy of the signal after phase correction.

[0152] In one embodiment, decoding the phase-corrected signal to obtain corrected food state data includes:

[0153] (1) Perform soft decision decoding on the phase-corrected signal to obtain the decoded data frame.

[0154] For example, the packaged food risk prediction terminal samples the phase-corrected signal at equal intervals, converting the continuous time-domain waveform into a discrete digital signal sequence. Based on a preset modulation and coding scheme, the packaged food risk prediction terminal calculates the posterior probability of the symbol represented by each sampling point, and uses a soft-decision decoding algorithm to iteratively calculate the path metric and backtrack the probability sequence to recover the most probable original encoded bitstream using the maximum likelihood criterion. Finally, it assembles the decoded data frame according to the communication frame format.

[0155] Among them, the soft decision decoding algorithm is a method of decoding using the probability information of the received signal amplitude. Its core lies in calculating the log-likelihood ratio of each bit to optimize the decoding performance.

[0156] (2) Verify the integrity of the decoded data frame and extract the payload data.

[0157] For example, the packaged food prediction terminal invokes a pre-set cyclic redundancy check algorithm to calculate the integrity verification field of the decoded data frame, and compares the calculation result with the check value carried in the frame to confirm that no errors occurred during data transmission and processing. After successful verification, the packaged food risk prediction terminal, according to the frame structure definition, strips the frame header, frame trailer, and check field to extract the core payload data carrying sensor information.

[0158] Cyclic Redundancy Check (CRC) is an error detection method that uses generator polynomial division to detect whether errors occur in data during transmission or storage.

[0159] (3) Analyze the sensor measurements in the payload data to obtain the corrected food status data.

[0160] For example, the packaged food risk prediction terminal, based on the data communication protocol agreed upon by the wireless sensor nodes, parses the extracted payload data, identifies and separates the binary representations of sensor measurements such as temperature, gas concentration, and humidity, converts them into physically meaningful floating-point values, and summarizes them to obtain corrected food status data. The calculation of the log-likelihood ratio in soft-decision decoding can be expressed as:

[0161]

[0162] In the formula, For the first bits The log-likelihood ratio; For a given transmitted bit of 1, the received signal sample The conditional probability; For a given transmitted bit of 0, the received signal sample The conditional probability; The pending judgment One information bit; For corresponding bits The received signal sample value.

[0163] The data communication protocol defines the frame format, encoding rules, and physical quantity conversion relationships for data exchange between wireless sensor nodes and receiving terminals.

[0164] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0165] In one embodiment, such as Figure 3 As shown, this application also provides a packaged food risk prediction system 300, which includes:

[0166] The data acquisition module 301 is used to acquire data through the wireless sensor integrated in the packaging to obtain food status parameters;

[0167] The channel detection module 302 is used to transmit a predefined probe sequence to an external receiver before transmitting food status parameters, and to obtain a distorted probe sequence after transmission through the packaging-food medium.

[0168] The feature inversion module 303 is used to perform machine learning inversion based on the distorted probe sequence after transmission through the packaging-food medium to generate time-varying channel fingerprint features;

[0169] The dynamic compensation module 304 is used to map the time-varying channel fingerprint features to a preset sensor signal transmission channel model; and to dynamically compensate the sensor signals containing food state parameters based on the sensor signal transmission channel model to obtain the corrected food state data.

[0170] The risk assessment module 305 is used to perform risk assessment on the corrected food status data and output the risk prediction results of packaged food.

[0171] Specifically, the packaged food risk prediction terminal includes a data acquisition module 301, a channel detection module 302, a feature inversion module 303, a dynamic compensation module 304, and a risk assessment module 305.

[0172] The data acquisition module activates and controls the wireless sensor integrated within the food packaging, enabling the sensor to operate continuously and sense physical and chemical quantities in the surrounding environment, such as temperature, gas concentration, and humidity. The wireless sensor converts the sensed physical and chemical quantities into corresponding raw electrical signals and performs preliminary signal conditioning to filter out irrelevant interference. The signal processing unit of the data acquisition module receives the conditioned electrical signal, analyzes and converts its format to obtain the food's state parameters.

[0173] Before initiating the food status parameter transmission process, the channel detection module performs a self-check on the signal transmission system to confirm the absence of transmission conflicts and stable equipment operation. The channel detection module retrieves and verifies a pre-defined probe sequence stored internally, confirming its compliance with transmission standards. Then, according to preset transmission parameters, transmission protocol, and signal strength standards, the channel detection module directionally transmits the verified predefined probe sequence towards the external receiver. After being emitted from the transmitter, the probe sequence sequentially penetrates the surface of the food packaging and passes through the interior of the food itself, continuously propagating along the transmission path formed by the packaging-food medium. Its signal structure, phase characteristics, and amplitude parameters undergo natural distortion due to medium properties and environmental interference. The external receiver captures this distorted sequence signal in real time, performs preliminary demodulation, and then feeds it back to the channel detection module through a preset feedback channel. The channel detection module receives the feedback signal, performs matching verification, and completes signal morphology identification and calibration through its internal signal processing flow, obtaining the distorted probe sequence after transmission through the packaging-food medium.

[0174] The feature inversion module performs signal preprocessing on the acquired distorted probe sequences, removing environmental interference and transmission noise, and standardizing the signal format to ensure the processed data conforms to the access specifications for machine learning inversion. The module then invokes a built-in machine learning inversion model, feeding the standardized distorted probe sequences into it. This model sequentially analyzes the signal distortion trends and propagation state changes reflected in the sequences, extracting signal distortion-related features, medium response-related features, and multipath propagation-related features. After extracting all effective features, the feature inversion module utilizes the model's built-in feature fusion mechanism to identify and integrate the inherent relationships between different features, removing redundant features and highlighting the core change patterns of the channel following changes in the medium state, generating time-varying channel fingerprint features.

[0175] The dynamic compensation module retrieves a preset sensor signal transmission channel model stored internally in the device. It then matches the obtained time-varying channel fingerprint features with the model's baseline feature system, completing the mapping process between features and the channel model. After feature mapping, the dynamic compensation module extracts the sensor signals carrying food state parameters and sends them to the corresponding channel model. The dynamic compensation module, based on the transmission patterns presented by the channel model, determines various deviations occurring during signal transmission and dynamically compensates the sensor signals according to the model's predetermined correction specifications. This eliminates signal distortion caused by medium propagation and multipath interference, resulting in corrected food state data.

[0176] After acquiring the corrected food condition data, the risk assessment module verifies the completeness and validity of the data and removes invalid or abnormal information. The module then retrieves pre-stored food risk assessment standards and corresponding quality requirements. It analyzes the actual food condition reflected in the corrected food condition data against these standards and requirements, comprehensively assessing potential safety hazards and quality change trends. Finally, the module summarizes all assessment information, organizes it into standardized judgment content according to a preset format, and outputs the packaged food risk prediction results.

[0177] Feature inversion module 303 is also used for:

[0178] The distorted probe sequence after transmission through packaging-food media is preprocessed to extract multidimensional signal distortion features;

[0179] Multidimensional signal distortion characteristics are analyzed using convolutional neural networks to identify multipath propagation modes;

[0180] Based on radial basis function network analysis of multidimensional signal distortion characteristics, dynamic medium variation components are captured;

[0181] By integrating multipath propagation modes and dynamic medium variation components, time-varying channel fingerprint features are generated.

[0182] Feature inversion module 303 is also used for:

[0183] By mapping the kernel function of the radial basis function network, the multidimensional signal distortion characteristics are transformed into a nonlinear separable space;

[0184] Identifying nonlinear patterns characterizing changes in food state within a nonlinearly separable space;

[0185] Separate the time-varying medium response component and the static medium response component in the nonlinear mode;

[0186] The time-varying medium response component is taken as the dynamic medium change component.

[0187] The dynamic compensation module 304 is also used for:

[0188] The amplitude of the sensor signal is compensated by using the signal attenuation factor in the sensor signal transmission channel model to obtain the amplitude-compensated signal.

[0189] The demodulated baseband signal is obtained by demodulating the amplitude-compensated signal.

[0190] Based on the phase distortion mode in the sensor signal transmission channel model, the demodulated baseband signal is phase-corrected to generate a phase-corrected signal.

[0191] Decode the phase-corrected signal to obtain the corrected food state data.

[0192] Risk assessment module 305 is also used for:

[0193] A multi-dimensional spoilage feature vector was extracted from the corrected food condition data. The multi-dimensional spoilage feature vector includes temperature change gradient, gas concentration slope and humidity accumulation.

[0194] By analyzing multi-dimensional corruption feature vectors using a pre-defined spatiotemporal correlation model, a corruption process prediction index is obtained.

[0195] The critical state of corruption process prediction indicators is assessed based on the dynamic risk threshold algorithm to determine the risk level parameters.

[0196] By mapping the risk level parameters to a preset food shelf-life decay curve, the risk prediction results for packaged foods are obtained.

[0197] The dynamic compensation module 304 is also used for:

[0198] Analyze the phase distortion mode and extract the phase shift component caused by multipath propagation and the phase jitter component caused by medium changes;

[0199] A joint compensation function is constructed based on the phase offset component and the phase jitter component;

[0200] The phase spectrum of the demodulated baseband signal is corrected by applying a joint compensation function to generate the corrected phase spectrum;

[0201] The time-domain signal corresponding to the corrected phase spectrum is reconstructed to obtain the phase-corrected signal.

[0202] The dynamic compensation module 304 is also used for:

[0203] The phase-corrected signal is then subjected to soft-decision decoding to obtain the decoded data frame;

[0204] Verify the integrity of the decoded data frame and extract the payload data;

[0205] The sensor measurements in the payload data are analyzed to obtain the corrected food condition data.

[0206] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0207] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0208] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0209] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method of predicting the risk of a packaged food, characterized by, The method includes: Data is collected using wireless sensors integrated within the packaging to obtain food status parameters; Before transmitting the food state parameters, a predefined probe sequence is transmitted to an external receiving end to obtain a distorted probe sequence after transmission through the packaging-food medium. Based on the distorted probe sequence transmitted through the packaging-food medium, machine learning inversion is performed to generate time-varying channel fingerprint features; The time-varying channel fingerprint features are mapped to a preset sensor signal transmission channel model; the sensor signals containing the food state parameters are dynamically compensated based on the sensor signal transmission channel model to obtain corrected food state data; A risk assessment is performed on the corrected food status data, and the risk prediction results for packaged food are output.

2. The method for predicting the risk of packaged food according to claim 1, characterized in that, The process of performing machine learning inversion based on the distorted probe sequence transmitted through the packaging-food medium to generate time-varying channel fingerprint features includes: The distorted probe sequence transmitted through the packaging-food medium is preprocessed to extract multidimensional signal distortion features; The multidimensional signal distortion characteristics are analyzed using a convolutional neural network to identify multipath propagation modes. The multidimensional signal distortion characteristics are analyzed based on radial basis function networks to capture dynamic medium variation components. The time-varying channel fingerprint feature is generated by fusing the multipath propagation mode and the dynamic medium variation component.

3. The method for predicting the risk of packaged food according to claim 2, characterized in that, The method of analyzing the multidimensional signal distortion features based on radial basis function networks and capturing dynamic medium variation components includes: The multidimensional signal distortion features are transformed into a nonlinear separable space by using the kernel function mapping of the radial basis function network; Identify nonlinear patterns characterizing changes in food state within the nonlinearly separable space; Separate the time-varying medium response component and the static medium response component in the nonlinear mode; The time-varying medium response component is used as the dynamic medium change component.

4. The method for predicting the risk of packaged food according to claim 1, characterized in that, The process of dynamically compensating the sensor signal containing the food state parameters based on the sensor signal transmission channel model to obtain corrected food state data includes: The sensor signal is amplitude compensated by using the signal attenuation factor in the sensor signal transmission channel model to obtain the amplitude-compensated signal. Demodulate the amplitude-compensated signal to obtain the demodulated baseband signal; Based on the phase distortion mode in the sensor signal transmission channel model, the demodulated baseband signal is phase-corrected to generate a phase-corrected signal. Decode the phase-corrected signal to obtain the corrected food state data.

5. The method for predicting the risk of packaged food according to claim 1, characterized in that, The step of performing a risk assessment on the corrected food condition data and outputting a risk prediction result for packaged food includes: A multi-dimensional spoilage feature vector is extracted from the corrected food state data. The multi-dimensional spoilage feature vector includes temperature change gradient, gas concentration slope and humidity accumulation. By analyzing the multi-dimensional corruption feature vectors using a pre-defined spatiotemporal correlation model, a corruption process prediction index is obtained. The critical state of the corruption process prediction indicators is evaluated based on the dynamic risk threshold algorithm to determine the risk level parameters; The risk level parameters are mapped to a preset food shelf-life decay curve to obtain the risk prediction results for the packaged food.

6. The method for predicting the risk of packaged food according to claim 4, characterized in that, The step of performing phase correction on the demodulated baseband signal based on the phase distortion mode in the sensor signal transmission channel model to generate a phase-corrected signal includes: The phase distortion mode is analyzed to extract the phase shift component caused by multipath propagation and the phase jitter component caused by medium changes; Based on the phase offset component and the phase jitter component, a joint compensation function is constructed; The phase spectrum of the demodulated baseband signal is corrected using the joint compensation function to generate the corrected phase spectrum; The time-domain signal corresponding to the corrected phase spectrum is reconstructed to obtain the phase-corrected signal.

7. The method for predicting the risk of packaged food according to claim 4, characterized in that, Decoding the phase-corrected signal to obtain the corrected food state data includes: The phase-corrected signal is subjected to soft-decision decoding to obtain a decoded data frame; Verify the integrity of the decoded data frame and extract the payload data; The sensor measurements in the payload data are analyzed to obtain the corrected food state data.

8. A risk prediction system for packaged food, characterized in that, The system includes: The data acquisition module is used to collect data through wireless sensors integrated within the packaging to obtain food status parameters; The channel detection module is used to transmit a predefined probe sequence to an external receiver before transmitting the food state parameters, and to obtain a distorted probe sequence after transmission through the packaging-food medium. The feature inversion module is used to perform machine learning inversion based on the distorted probe sequence transmitted through the packaging-food medium to generate time-varying channel fingerprint features; The dynamic compensation module is used to map the time-varying channel fingerprint features to a preset sensor signal transmission channel model; and to dynamically compensate the sensor signals containing the food state parameters based on the sensor signal transmission channel model to obtain corrected food state data. The risk assessment module is used to perform risk assessment on the corrected food status data and output the risk prediction results for packaged food.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the packaged food risk prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the packaged food risk prediction method according to any one of claims 1 to 7.

Images