Grain logistics supervision method and system considering sensitive information

Abstract

The invention discloses a grain logistics supervision method considering sensitive information, and the method comprises the steps: presetting a plurality of grain logistics transportation features and influence factors thereof, and presetting the score of each influence factor in different states; constructing a judgment matrix according to the influence factors to calculate the weight of each influence factor; obtaining the current state of each influence factor in real time, calculating the comprehensive score of each influence factor, and selecting m influence factors as key influence factors; various abnormal behaviors are preset, and the posterior probability of each abnormal behavior is calculated through a Bayesian model according to the conditional probability of the key influence factor; and after the transportation is finished, encrypting the vehicle track based on a differential privacy protection technology. According to the method, the abnormal track behavior of the grain transport vehicle can be identified through the analytic hierarchy process, and meanwhile, the track of the grain transport vehicle is encrypted through the differential privacy technology to guarantee the grain logistics safety.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent logistics monitoring, and in particular to a method and system for monitoring grain logistics that takes sensitive information into account. Background Art

[0002] In recent years, my country has produced over 1.3 trillion jin (approximately 1.3 trillion kilograms) of grain annually, 70% of which enters the consumer market through grain circulation, totaling over 900 billion kilograms. Against this backdrop, grain transportation safety has become a crucial component of national food security oversight, placing higher demands on information-based oversight of grain purchase and sales.

[0003] Current research on grain logistics supervision primarily focuses on regulating grain purchases and sales within grain depots. For example, patent "CN119359213A - A Smart Grain Depot Operation Platform and System" relies primarily on video surveillance within the depot to collect data and intelligently analyze perimeter security and personnel behavior. Patents such as "CN119048545A - A Grain Depot Grain Transport Vehicle Operation Supervision Method" disclose technical content that regulates grain transport vehicle operations, but their oversight of grain logistics remains insufficient, primarily due to the following issues:

[0004] (1) There is a lack of algorithms for big data intelligent analysis of the spatiotemporal location information and vehicle status information obtained by on-board equipment, which makes it impossible to timely and effectively judge abnormal behavior in grain transportation and give early warning prompts.

[0005] (2) Traditional grain logistics supervision methods lack the protection of logistics information. If the spatiotemporal location information and vehicle status information of grain logistics vehicles are leaked, sensitive grain information such as grain warehouse location information and grain warehouse storage information can be analyzed and extracted through technical means, endangering the country's food security.

[0006] In summary, there is an urgent need for an intelligent food logistics full-process supervision method that can capture abnormal behaviors during food logistics in real time and protect the privacy of transportation trajectory data after the food transportation is completed. Summary of the Invention

[0007] In order to overcome the problems in the prior art, the purpose of the present invention is to provide a food logistics supervision method and system that takes into account sensitive information, which can identify abnormal behavior of the trajectory of grain trucks through hierarchical analysis, and at the same time encrypt the trajectory of grain trucks through differential privacy technology to ensure the safety of food logistics.

[0008] To achieve the above objectives, the present invention provides a method for supervising food logistics that takes into account sensitive information, comprising the following steps:

[0009] S1: Preset multiple grain logistics transportation characteristics, each of which includes multiple influencing factors, preset scores of each influencing factor in different states, and store an influencing factor score table;

[0010] S2: Construct a judgment matrix based on the impact factors and calculate the weight of each impact factor through the hierarchical analysis method;

[0011] S3: Obtain the current status of each impact factor in real time, calculate the comprehensive score of each impact factor according to the impact factor score table and impact factor weight, sort the impact factors according to the comprehensive score, and select m impact factors as key impact factors;

[0012] S4: Preset multiple abnormal behaviors and the probability table of each influencing factor in abnormal behaviors. Calculate the posterior probability of each abnormal behavior through the Bayesian model based on the conditional probability of key influencing factors, and preset the abnormal probability threshold. When the posterior probability of the abnormal behavior is greater than the abnormal probability threshold, a real-time alarm is triggered;

[0013] S5: After the transportation is completed, the vehicle trajectory is encrypted based on differential privacy protection technology.

[0014] Furthermore, the grain logistics transportation characteristics include four categories: driver identity characteristics, spatiotemporal characteristics, grain characteristics, and driver behavior characteristics, among which:

[0015] The factors influencing driver identity characteristics include gender, age, number of transports, education level, and historical penalty records;

[0016] The factors affecting spatiotemporal characteristics include operating hours, weather conditions, road conditions, deviation routes, and abnormal station points;

[0017] Factors influencing grain characteristics include grain type and weight;

[0018] The influencing factors of driver behavior characteristics include driver expression.

[0019] Furthermore, the method of constructing the judgment matrix in step S2 and calculating the weight of each influencing factor by the hierarchical analysis method includes the following steps:

[0020] S21: Construct an n×n positive and negative judgment matrix C, the formula is expressed as: C=(C ij ) n×n ;

[0021] Among them, the judgment matrix element C ij It represents the importance of factor i relative to factor j, and n represents the order of the matrix;

[0022] S22: Get the product M of each row i , the formula is as follows:

[0023]

[0024] M i Doing the nth root operation gives The formula is as follows:

[0025]

[0026] right Normalize and get the weight w of the i-th influencing factor i , the formula is as follows:

[0027]

[0028] S23: Calculate the consistency ratio If CR<0.1, the logic of the judgment matrix C is reasonable. If CR≥0.1, the judgment matrix is readjusted;

[0029] Among them, the consistency index

[0030] Maximum eigenvalue of a matrix

[0031] The average random consistency index RI is obtained by looking up the reference table.

[0032] Further, in step S3,

[0033] S31: Get the current status of each impact factor in real time;

[0034] S32: Calculate the comprehensive score Si of each impact factor based on the impact factor score table and the impact factor weight. The calculation formula is:

[0035] S i =f i ×w i , where i is the impact factor, fi is the impact factor score, and w i is the weight of factor i;

[0036] S33: Sort the comprehensive score Si of each impact factor from large to small, and select the first m impact factors as key impact factors, namely D0, D1, D2, ..., D m-1 .

[0037] Furthermore, the specific method for obtaining the current status of each influencing factor in real time is:

[0038] The driver characteristics and food characteristics are directly obtained based on manually set information;

[0039] The operating time is obtained by recording the timestamp of the vehicle's onboard positioning system in UTC time;

[0040] The weather conditions are obtained by accessing the weather information API;

[0041] The road condition is obtained by collecting the latitude and longitude coordinates and timestamp of the current vehicle location and accessing the road condition query service API;

[0042] The offset route generates a buffer zone on both sides of the preset route, periodically obtains the latitude and longitude coordinates of the current vehicle, and if the latitude and longitude coordinates of the vehicle fall outside the buffer zone, it is an offset route;

[0043] The abnormal station point is based on a preset residence time threshold and collects vehicle trajectory points. If the vehicle's residence time is greater than the set residence time threshold and the vehicle trajectory point is not in the service area, it is considered an abnormal station point;

[0044] The driver's facial expression is obtained by a monitoring camera in the vehicle cockpit, and the driver's facial expression is recognized by a convolutional neural network and a recognition result is output.

[0045] Furthermore, in step S31, the driver's facial expression recognition includes:

[0046] S311: Build and train the ResNet model;

[0047] S312: Obtain the cockpit camera video stream through the RTSP protocol and sample frame images at 5-second intervals;

[0048] S313: Use the MTCNN model to grayscale the driver's cockpit video stream, detect faces, and locate key points, extract facial region images, and detect facial regions;

[0049] S314: The driver's facial expression image is obtained in real time through the monitoring in the cockpit, and the facial expression image is input into the ResNet model for recognition. The probability value A of the seven facial expression states of anger, disgust, fear, happiness, sadness, surprise, and neutral is obtained. The facial expression state with the maximum probability value A is output as the classification result, and the classification result is associated with the timestamp and stored, and a query interface is provided through the REST API.

[0050] Furthermore, the method of step S4 specifically includes the following steps:

[0051] S41: Preset k types of abnormal behaviors, with E representing abnormal behaviors of grain transport vehicles; expressed as: E=0, E=1, ..., E=(k-1);

[0052] S42: Establish a probability table of each influencing factor in abnormal behavior, calculate the posterior probability of each abnormal behavior through the Bayesian model according to the conditional probability of the key influencing factors, and take the abnormal behavior with the maximum posterior probability as the prediction result of the current time; S43: Preset the abnormal probability threshold as If the maximum posterior probability does not exceed the preset abnormal probability threshold When the maximum posterior probability is greater than the preset abnormal probability threshold, it is judged to be normal. A real-time alarm is triggered when an abnormality occurs. The alarm is in the form of a pop-up window on the system interactive interface. The alarm content includes abnormal behavior, posterior probability value, driver and vehicle information.

[0053] Furthermore, the step S42 specifically includes the following steps:

[0054] S421: Based on historical data or expert experience, a probability table of each influencing factor occurring in different abnormal behaviors is preset;

[0055] S422: Calculate the joint conditional probability of all key factors occurring simultaneously in different abnormal behavior events according to the probability table described in the query step S421, P(D0, D1, D2, ..., D m-1 |E), the formula is expressed as:

[0056]

[0057] Calculate the probability P(D0, D1, D2, ..., D m-1 ), the historical events include abnormal behaviors and non-abnormal behaviors, and the formula is:

[0058]

[0059] Through the Bayesian model, the posterior probability of each abnormal behavior is calculated as:

[0060]

[0061] S423: Obtain the posterior probability P of k types of abnormal behaviors calculated at the current moment post (E=0), P post (E=1),…,P post (E=(k-1)), take the abnormal behavior corresponding to the maximum posterior probability as the prediction result at the current moment.

[0062] Furthermore, in step S5, the specific steps of protecting the trajectory privacy include:

[0063] S51: Preset definition of privacy budget ε and sensitivity Δ;

[0064] S52: Acquire trajectory data recorded by the onboard positioning system of the transport vehicle;

[0065] S53: For trajectory point (x i ,y i ) adds noise, and the specific calculation method is:

[0066]

[0067] Among them, (x i ,y i ) is the original trajectory coordinate, The trajectory coordinates after perturbation, Laplce(0, Δ / ε) is the probability density function of the Laplace distribution, and the parameter '0' indicates that the mean of the noise distribution is 0.

[0068] The present invention also provides a food logistics supervision system, comprising:

[0069] Multi-source data acquisition module: including Beidou / GPS locator, environmental API interface, and cockpit camera, used to collect the current status of influencing factors in real time;

[0070] Behavior recognition unit: processes surveillance video and driver expressions based on OpenCV and CNN technology;

[0071] Real-time analysis module: used to perform AHP weight calculation and Bayesian probability inference;

[0072] Alarm output module: Generates an alarm pop-up window instruction containing abnormal behavior, posterior probability value, driver and vehicle information.

[0073] Privacy processing module: used to calculate the noise of vehicle trajectory points and generate the disturbed trajectory coordinates.

[0074] This paper uses the Analytic Hierarchy Process (AHP) to identify factors influencing abnormal behavior in grain logistics and employs Bayesian probabilistic analysis to identify abnormal behavior. This effectively captures abnormal behavior during grain transportation and generates alarm feedback. Furthermore, a trajectory protection algorithm based on differential privacy introduces controllable noise during data processing, effectively protecting the privacy of trajectory data while maintaining data availability, providing a viable solution for users and service providers. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A flow chart of a method for supervising food logistics that takes sensitive information into account, provided in Example 1 of the present invention; DETAILED DESCRIPTION

[0076] Example 1

[0077] This embodiment provides a method for supervising food logistics that takes into account sensitive information, such as Figure 1 As shown, the specific steps include:

[0078] S1: Preset a plurality of grain logistics transportation characteristics, each of which includes a plurality of influencing factors, preset scores of each influencing factor in different states, and store an influencing factor score table.

[0079] Specifically, the grain logistics transportation characteristics include four categories: driver expression characteristics, driver identity characteristics, spatiotemporal characteristics, grain characteristics, and behavioral characteristics, as shown in the following table:

[0080]

[0081] Each influencing factor includes multiple influencing factor values, such as the gender includes male and female, the age includes 20-30 years old, 30-40 years old, and 40-50 years old, and the weather conditions include sunny, cloudy, rainy, snowy, foggy, and other;

[0082] Preset the scores of each influencing factor in different states and store the influencing factor score table. Take the weather condition score table as an example:

[0083] Weather conditions clear Negative rain Snow fog other Off-route 0.05 0.05 0.15 0.5 0.1 0.15 Multiple warehousing 0.1 0.05 0.15 0.3 0.3 0.1 traffic accidents 0.7 0.16 0.02 0.01 0.01 0.02 Stealing food 0.2 0.25 0.2 0.15 0.15 0.05

[0084] The impact factor values and scores included in the impact factors can be set according to one's own situation, and the 13 impact factors provided in this embodiment are not exhaustively listed here.

[0085] S2: Construct a judgment matrix and calculate the weight of each influencing factor through the analytic hierarchy process. The specific steps include:

[0086] S21: Construct an n×n positive and negative judgment matrix C, the formula is expressed as: C=(C ij ) n×n ;

[0087] Among them, the judgment matrix element C ij It indicates the importance of factor i relative to factor j, for example:

[0088] C ij =1: factors i and j are equally important,

[0089] C ij =3: Factor i is slightly more important than j,

[0090] C ij =5: Factor i is significantly more important than factor j.

[0091] and

[0092] n represents the order of the matrix, which is the same as the number of influencing factors. Since 13 influencing factors are listed in this embodiment, n=13 in this embodiment.

[0093] S22: Get the product M of each row i , the formula is as follows:

[0094]

[0095] M i Doing the nth root operation gives The formula is as follows:

[0096]

[0097] right Normalize and get the weight w of the i-th influencing factor i , the formula is as follows:

[0098]

[0099] in, is the weight of the i-th impact factor;

[0100] S23: In order to verify whether the logic of the judgment matrix C is reasonable and avoid contradictions in the expert scores, it is necessary to perform a consistency check on the judgment matrix. The specific process is as follows:

[0101] Calculating the consistency ratio If CR<0.1, the logic of the judgment matrix C is reasonable. If CR≥0.1, the judgment matrix is readjusted.

[0102] Among them, the consistency index Maximum eigenvalue of a matrix The average random consistency index RI is obtained by looking up the reference table.

[0103] The reference table of the average random consistency index RI is shown in the figure below:

[0104]

[0105] In this embodiment, since this embodiment lists 13 influencing factors, the RI=1.56.

[0106] S3: Obtain the current status of each impact factor in real time, calculate the comprehensive score of each impact factor according to the impact factor score table and impact factor weight, sort the impact factors according to the comprehensive score, and select m impact factors as key impact factors;

[0107] S31: Obtain the current status of each impact factor in real time; the specific method is:

[0108] The influencing factors of the driver characteristics include gender, age, number of transports, education level and historical penalty records. The influencing factors of the driver characteristics are manually filled in and stored in the system database. They can be directly read from the database during calculation.

[0109] and grain characteristics are directly obtained based on artificially set information;

[0110] The operating time is obtained by recording the timestamp of the vehicle positioning system in UTC time;

[0111] The weather condition is obtained by accessing the weather information API to obtain six state codes: sunny (0), cloudy (1), rainy (2), snowy (3), foggy (4), and other (5).

[0112] The road condition is obtained by collecting the latitude and longitude coordinates and timestamp of the current vehicle location and accessing the road condition query service API; four status codes are returned: smooth travel (0), slow travel (1), congested (2), and extremely congested (3).

[0113] The offset route is based on generating a buffer zone on both sides of the preset route, periodically obtaining the latitude and longitude coordinates of the current vehicle. If the latitude and longitude coordinates of the vehicle are outside the buffer zone, it is an offset route. The specific method is as follows:

[0114] Generate a width W on both sides of the planned route road =d road ×e GPS The buffer, where d road is the road width, e GPS To account for positioning errors, the vehicle trajectory points are obtained every 3 minutes. If three consecutive trajectory points are outside the buffer zone, it is determined to be off-route.

[0115] The abnormal stationary point is determined according to the preset residence time threshold T stay , periodically collect vehicle trajectory points, if the vehicle stay time is greater than the set stay time threshold T stay , the current vehicle is in a stopped state. The system obtains the longitude and latitude coordinates of the stop point, obtains the POI type of the current stop point through the location retrieval service API, and determines whether the current vehicle track point is a service area. If so, it is a normal stop point; if not, it is an abnormal stop point.

[0116] The driver's facial expression is captured by the monitoring camera in the vehicle cockpit, and the driver's facial expression is recognized by a convolutional neural network and the recognition result is output. Specifically, the following steps are included:

[0117] S311: Build and train the ResNet model;

[0118] S312: Obtain the cockpit camera video stream through the RTSP protocol and sample frame images at 5-second intervals;

[0119] S313: Use the MTCNN model to grayscale the driver's cockpit video stream, detect faces, and locate key points, extract facial region images, and detect facial regions;

[0120] S314: The driver's facial expression image is obtained in real time through the monitoring in the cockpit, and the facial expression image is input into the CNN model for recognition, and the probability values A of seven types of facial expression states, namely anger (0), disgust (1), fear (2), happiness (3), sadness (4), surprise (5), and neutral (6) are obtained. The facial expression state with the maximum probability value A is output as the result, and the classification result is associated with the timestamp and stored, and a query interface is provided through the REST API.

[0121] S32: Combine the impact factor score table and the impact factor weight to calculate the comprehensive score Si of each impact factor. The calculation formula is:

[0122] S i =f i ×w i ,

[0123] Where i is the impact factor, fi is the impact factor score, and w i is the weight of factor i; (e.g., when the driver’s gender is “male”, fi = 0.7; when the driver’s gender is “female”, fi = 0.3).

[0124] S33: Sort the comprehensive score Si of each influencing factor from large to small, and select the top 7 influencing factors as key influencing factors D1, D2, ..., D7. In this embodiment, the top 7 influencing factors are: abnormal stop (D0), deviation from the established route (D1), grain weight (D2), whether there is a historical penalty record (D3), road conditions (D4), weather conditions (D5), and driver's expression (D6).

[0125] S4: Preset a probability table of multiple abnormal behaviors and the occurrence of each influencing factor in the abnormal behavior. Calculate the posterior probability of each abnormal behavior through the Bayesian model based on the conditional probability of the key influencing factors, and preset the abnormal probability threshold. When the posterior probability of the abnormal behavior is greater than the abnormal probability, trigger a real-time alarm;

[0126] S41: Preset k types of abnormal behaviors, with E representing the type of abnormal behavior of the grain transport vehicle, expressed as: E=0, E=1, ..., E=(k-1);

[0127] Specifically, in this embodiment, k=4, so E=0, 1, 2, 3 correspond to four abnormal behaviors, namely, route deviation, repeated entry and exit, traffic accidents, and grain theft.

[0128] S42: Establish a probability table for each influencing factor in abnormal behavior, calculate the posterior probability of each abnormal behavior through the Bayesian model based on the conditional probability of the key influencing factors, and take the abnormal behavior with the maximum posterior probability as the prediction result for the current time;

[0129] The step S42 specifically includes the following steps:

[0130] S421: Based on historical data or expert experience, a probability table of each influencing factor occurring in different abnormal behaviors is preset;

[0131] The probability table pre-stores a variety of conditional probabilities P (influence factor | abnormal behavior). For example, P(D2 | E = 3) = 0.6 indicates that the probability that the driver has a historical penalty record when grain theft occurs is 0.6, and P(E = 0) = P(E = 1) = P(E = 2) = P(E = 3) = 0.25, indicating that the probability of the four abnormal behaviors occurring is 0.25 regardless of any conditions. The probability table is adjusted and set based on historical data or expert experience, and will not be exhaustively described here.

[0132] S422: Based on the probability table described in query step S421, calculate the joint conditional probability of all key factors occurring simultaneously in different abnormal behavior events, P(D0, D1, D2, D3, D4, D5, D6|E), using the formula:

[0133]

[0134] Calculate the probability P(D0, D1, D2, D3, D4, D5, D6) of the key factors co-occurring in historical events, including abnormal and non-abnormal behaviors, using the formula:

[0135]

[0136] Through the Bayesian model, the posterior probability of each abnormal behavior is calculated as:

[0137]

[0138] S423: Obtain the posterior probabilities of the four abnormal behaviors calculated at the current moment, which are expressed as follows:

[0139] P post (E=0),P post (E=1),P post (E=2),P post (E=3), take the abnormal behavior corresponding to the maximum posterior probability as the prediction result at the current moment.

[0140] S43: The preset abnormal probability threshold is In this embodiment If the maximum posterior probability does not exceed the preset abnormal probability threshold, it is judged to be in a normal state. If the maximum posterior probability is greater than the preset abnormal probability threshold, a real-time alarm is triggered. The alarm is in the form of a pop-up window on the system interaction interface. The content of the alarm includes abnormal behavior, posterior probability value, driver and vehicle information.

[0141] S5: After the transportation is completed, the grain truck trajectory is encrypted using the Laplace noise addition method based on differential privacy protection technology. The specific steps include:

[0142] S51: Preset definition of privacy budget ε = 0.2 and sensitivity Δ = 0.001;

[0143] S52: Obtaining vehicle trajectory data based on vehicle positioning system records;

[0144] S53: For trajectory point (x i ,y i ) adds noise, and the specific calculation method is:

[0145]

[0146] Among them, (x i ,y i ) is the original coordinate of the trajectory, The trajectory coordinates after perturbation, Laplace(0, Δ / ε) is the probability density function of the Laplace distribution, and the parameter '0' indicates that the mean of the noise distribution is 0.

[0147] This example uses the Analytic Hierarchy Process (AHP) to identify factors influencing abnormal behavior in grain logistics and employs Bayesian probabilistic analysis to identify abnormal behavior. This effectively captures abnormal behavior during grain transportation and generates alarm feedback. Furthermore, a trajectory protection algorithm based on differential privacy introduces controllable noise during data processing, effectively protecting the privacy of trajectory data while maintaining data availability.

[0148] Example 2

[0149] This embodiment further provides a food logistics supervision system that takes sensitive information into consideration and is used to implement the method described in Example 1, including:

[0150] Multi-source data acquisition module: including Beidou / GPS locator, environmental API interface, and cockpit camera, used to collect the current status of influencing factors in real time;

[0151] Behavior recognition unit: processes surveillance video and driver expressions based on OpenCV and CNN technology;

[0152] Real-time analysis module: used to perform AHP weight calculation and Bayesian probability inference;

[0153] Alarm output module: Generates an alarm pop-up window instruction containing abnormal behavior, posterior probability value, driver and vehicle information.

[0154] Privacy processing module: used to calculate the noise of vehicle trajectory points and generate the disturbed trajectory coordinates;

[0155] The two embodiments described above are only preferred specific implementation methods of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the protection scope of the present invention.

Claims

1. A method for supervising food logistics that takes into account sensitive information, characterized in that: The following steps are involved: S1: Preset multiple grain logistics transportation characteristics, each of which includes multiple influencing factors, preset scores of each influencing factor in different states, and store them as an influencing factor score table; S2: Construct a judgment matrix based on the impact factors and calculate the weight of each impact factor through the hierarchical analysis method; S3: Obtain the current status of each impact factor in real time, combine the impact factor score table and the impact factor weight, calculate the comprehensive score of each impact factor, sort the impact factors according to the comprehensive score, and select m impact factors as key impact factors; S4: Preset a table of the probability of occurrence of multiple abnormal behaviors and various influencing factors in abnormal behaviors. Calculate the posterior probability of each abnormal behavior through the Bayesian model based on the conditional probability of key influencing factors, and preset an abnormal probability threshold. When the posterior probability of the abnormal behavior is greater than the abnormal probability threshold, a real-time alarm is triggered. S5: After the transportation is completed, the vehicle trajectory is encrypted based on differential privacy protection technology.

2. A method for supervising food logistics taking into account sensitive information as described in claim 1, characterized in that: The multiple grain logistics transportation characteristics in step S1 include four categories: driver identity characteristics, spatiotemporal characteristics, grain characteristics, and driver behavior characteristics, among which: The factors influencing driver identity characteristics include gender, age, number of transports, education level, and historical penalty records; The factors affecting spatiotemporal characteristics include operating hours, weather conditions, road conditions, deviation routes, and abnormal station points; Factors influencing grain characteristics include grain type and weight; The influencing factors of driver behavior characteristics include driver expression.

3. A method for supervising food logistics taking into account sensitive information as described in claim 1, characterized in that: In step S2, the method of constructing a judgment matrix and calculating the weight of each influencing factor by using the analytic hierarchy process includes the following steps: S21: Construct an n×n positive and negative judgment matrix C, the formula is expressed as: C=(C ij ) n×n ; Among them, the judgment matrix element C ij It represents the importance of factor i relative to factor j, and n represents the order of the matrix; S22: Get the product M of each row i , the formula is as follows: M i Doing the nth root operation gives The formula is as follows: right Normalize and get the weight w of the i-th influencing factor i , the formula is as follows: in, is the weight of the i-th impact factor; S23: Calculate the consistency ratio If CR<0.1, the logic of the judgment matrix C is reasonable. If CR≥0.1, the judgment matrix is readjusted, where the consistency index Maximum eigenvalue of a matrix The average random consistency index RI is obtained by looking up the reference table.

4. A method for supervising food logistics taking into account sensitive information as described in claim 2, characterized in that: The method of step S3 comprises the following steps: S31: Get the current status of each impact factor in real time; S32: Calculate the comprehensive score Si of each impact factor according to the impact factor score table and impact factor weight. The calculation formula is: S i =f i ×w i , Where i is the impact factor, fi is the impact factor score, and w i is the weight of factor i; S33: Sort the comprehensive score Si of each impact factor from large to small, and select the first m impact factors as key impact factors, which are represented as D0, D1, D2, ..., D m-1 .

5. A method for supervising food logistics taking into account sensitive information as described in claim 4, characterized in that: In step S21, the method for obtaining the current status of each influencing factor in real time is: The driver characteristics and food characteristics are directly obtained based on manually set information; The operating time is obtained by recording the timestamp of the vehicle's onboard positioning system in UTC time; The weather conditions are obtained by accessing the weather information API; The road condition is obtained by collecting the latitude and longitude coordinates and timestamp of the current vehicle location and accessing the road condition query service API; The offset route generates a buffer zone on both sides of the preset route, periodically obtains the latitude and longitude coordinates of the current vehicle, and if the latitude and longitude coordinates of the vehicle are outside the buffer zone, it is an offset route; The abnormal station point is based on a preset residence time threshold and collects vehicle trajectory points. If the vehicle's residence time is greater than the set residence time threshold and the vehicle trajectory point is not in the service area, it is considered an abnormal station point; The driver's facial expression is obtained by a monitoring camera in the vehicle cockpit, and the driver's facial expression is recognized by a convolutional neural network and a recognition result is output.

6. A method for supervising food logistics taking into account sensitive information as described in claim 5, characterized in that: In step S31, the driver's facial expression recognition includes: S311: Build and train the ResNet model; S312: Obtain the cockpit camera video stream through the RTSP protocol and sample frame images at 5-second intervals; S313: Use the MTCNN model to grayscale the driver's cockpit video stream, detect faces, and locate key points, extract facial region images, and detect facial regions; S314: The driver's facial expression image is obtained in real time through the monitoring in the cockpit, and the facial expression image is input into the ResNet model for recognition. The probability value A of the seven facial expression states of anger, disgust, fear, happiness, sadness, surprise, and neutral is obtained. The facial expression state with the maximum probability value A is output as the classification result, and the classification result is associated with the timestamp and stored, and a query interface is provided through the REST API.

7. A method for supervising food logistics taking into account sensitive information as described in claim 1, characterized in that: The step S4 specifically includes the following steps: S41: Preset k types of abnormal behaviors, with E representing abnormal behaviors of grain transport vehicles, expressed as: E=0, E=1, ..., E=(k-1); S42: Establish a probability table for each influencing factor in abnormal behavior, calculate the posterior probability of each abnormal behavior through the Bayesian model based on the conditional probability of the key influencing factors, and take the abnormal behavior with the maximum posterior probability as the prediction result for the current time; S43: The preset abnormal probability threshold is If the maximum posterior probability does not exceed the preset abnormal probability threshold When the maximum posterior probability is greater than the preset abnormal probability threshold, it is judged to be normal. A real-time alarm is triggered when an abnormal behavior occurs. The alarm is in the form of a pop-up window on the system interactive interface. The alarm content includes the abnormal behavior type, posterior probability value, driver and vehicle information.

8. A method for supervising food logistics taking into account sensitive information according to claim 7, characterized in that: The step S42 specifically includes the following steps: S421: Based on historical data or expert experience, a probability table of each influencing factor occurring in different abnormal behaviors is preset; S422: Based on the probability table in query step S421, calculate the joint conditional probability of all key factors occurring simultaneously in different abnormal behavior events. The formula is: Calculate the probability P(D0, D1, D2, ..., D m-1 ), the historical events include abnormal behaviors and non-abnormal behaviors, and the formula is: Through the Bayesian model, the posterior probability of each abnormal behavior is calculated as: S423: Obtain the posterior probability P of k types of abnormal behaviors calculated at the current moment post (E=0), P post (E=1),…,P post (E=(k-1)), take the abnormal behavior corresponding to the maximum posterior probability as the prediction result at the current moment.

9. A method for supervising food logistics taking into account sensitive information as described in claim 1, characterized in that: In step S5, the specific steps of trajectory privacy protection include: S51: Preset definition of privacy budget ε and sensitivity Δ; S52: Obtaining vehicle trajectory data based on vehicle positioning system records; S53: For trajectory point (x i ,y i ) adds noise, and the specific calculation method is: Among them, (x i ,y i ) is the original coordinate of the trajectory, The trajectory coordinates after perturbation, Laplace(0,Δ / ε) is the probability density function of the Laplace distribution, and the parameter '0' indicates that the mean of the noise distribution is 0.

10. A food logistics supervision system taking into account sensitive information, used to implement a food logistics supervision method taking into account sensitive information as described in any one of claims 1 to 9, characterized in that: The invention is characterized by comprising: Multi-source data acquisition module: including Beidou / GPS locator, environmental API interface, and cockpit camera, used to collect the current status of influencing factors in real time; Behavior recognition unit: processes surveillance video and driver expressions based on OpenCV and CNN technology; Real-time analysis module: used to perform AHP weight calculation and Bayesian probability inference; Alarm output module: Generates an alarm pop-up window containing abnormal behavior, posterior probability value, driver and vehicle information; Privacy processing module: used to calculate the noise of vehicle trajectory points and generate the disturbed trajectory coordinates.

Images