CN118333501A - Logistics and supply chain management system based on RFID - Google Patents

Abstract

The invention relates to the field of logistics management, in particular to an RFID-based logistics and supply chain management system, which comprises a data acquisition and integration module, a transportation and distribution tracking module, a data analysis and prediction module and a supply chain cooperation and visualization module, wherein the scheme adopts an advanced TADN technology, integrates an RFID tag technology and an intelligent video monitoring image sequence analysis means, captures and identifies unique characteristics and motion tracks of each cargo in real time, thereby effectively guaranteeing the safety and efficiency of logistics transportation links; according to the scheme, the LOA-LSTM model is adopted to predict the time required by cargo transportation and the occurrence probability of emergency, real-time data monitoring and historical data are combined to analyze, efficient optimization of a logistics transportation path and reasonable resource distribution are achieved, the logistics transportation efficiency and the supply chain reaction speed are improved, and the influence of emergency and complex conditions on cargo transportation is reduced.

Description

Logistics and supply chain management system based on RFID

Technical Field

The invention relates to the field of logistics management, in particular to an RFID-based logistics and supply chain management system.

Background

In the field of modern logistics and supply chain management, RFID technology has been widely used, and the advantages of automatic identification, real-time tracking and mass data processing are improved, so that the efficiency and accuracy of the whole process are improved. However, the conventional RFID technology cannot effectively track, match and position and track the entity and RFID label of cargo transportation, so that a supply chain manager is difficult to realize seamless tracing and fine management and control of the whole logistics process, and the overall efficiency and accuracy of logistics management are seriously affected; when the system faces an emergency or a complex condition, a supply chain manager cannot accurately predict the system, so that it is difficult to accurately and intelligently allocate logistics resources, the logistics efficiency is reduced, and the overall reaction speed and the service quality of the supply chain are affected.

Disclosure of Invention

Aiming at the problems that the prior RFID technology cannot effectively track, match and position and track the entity and RFID labels of cargo transportation, so that a supply chain manager is difficult to realize seamless tracing and refined management and control on the whole logistics flow, and the overall efficiency and accuracy of logistics management are seriously influenced; aiming at the problem that logistics resources are difficult to accurately allocate when a supply chain manager cannot accurately predict the emergency or complex conditions, so that logistics efficiency is low, the LOA-LSTM model is adopted to predict the time required by cargo transportation and the occurrence probability of the emergency, real-time data monitoring and historical data are combined to analyze, efficient optimization of a logistics transportation path and reasonable resource allocation are realized, logistics transportation efficiency and supply chain reaction speed are improved, and the influence of the emergency and complex conditions on cargo transportation is reduced.

The logistics and supply chain management system based on RFID comprises a data acquisition and integration module, a transportation and distribution tracking module, a data analysis and prediction module and a supply chain coordination and visualization module;

The data acquisition and integration module comprises the steps of generating, encoding and reading RFID labels, deploying RFID readers at each key node of a warehouse, acquiring inventory information and warehouse entry information of goods in the warehouse, forming a structured database record, and updating the database record in real time;

The transportation and distribution tracking module tracks and updates the in-transit state of goods in real time by scanning RFID labels and monitoring equipment on packages and using TADN technology, meanwhile, the transportation and distribution tracking module has an abnormal condition early warning function, and when abnormal conditions occur in the transportation process, the transportation and distribution tracking module rapidly sends out an alarm and transmits the alarm information to the data analysis and prediction module for analysis and prediction, and timely adjusts the transportation strategy;

The data analysis and prediction module performs learning training by using an LOA-LSTM model method, collects and integrates various factors and alarm information affecting the stability of a supply chain, and performs probability prediction on emergent conditions through multi-dimensional data input and learning to obtain a prediction result;

The supply chain coordination and visualization module builds an RFID-based supply chain coordination platform, so that physical distribution activity data can be acquired and shared in real time among partners, a coordination decision is promoted, the physical distribution activity data is converted into a visual and understandable chart and map form, and a whole-course visual supply chain monitoring interface is provided.

Further, in the transportation and distribution tracking module, the TADN technology is used for tracking and updating the in-transit state of the goods in real time, and the method specifically comprises the following steps:

Step S1: in the detection stage, a pre-trained target detector is used for analyzing video content and information recorded in a database in real time in monitoring equipment, detecting detection entities related to transportation information, and acquiring position information of each detection entity;

step S2: in the feature extraction stage, for each detection entity, extracting high-level features by using a pre-trained ResNet network, and extracting deep learning features of an image area where the high-level features are positioned to obtain feature information;

Step S3: the target representation construction stage integrates the position information and the characteristic information, constructs the representation of the current detection entity, namely a detection target, and updates the current state of the detection entity which is identified and tracked previously, namely the current active target for short;

step S4: TADN, processing the detection target and the existing active target by using TADN technology, calculating an appearance similarity matrix, and judging the matching relationship between the detection target and the existing active target;

Step S5: in the matching and target updating stage, target matching is carried out according to the result output by TADN, the motion trail and the appearance model of the detection target are updated, and repeated matching results are removed;

Further, in step S4, the detected target and the existing active target are processed by using TADN technology, which specifically includes the following steps:

Step S41: preprocessing input data, processing a detection target and an existing active target by using a linear embedding layer, adding a learnable embedding vector representing an empty target to the existing active target, and generating detection target data and active target data, wherein the formula is as follows:

；

；

Wherein, The detection target is indicated to be a target of the detection,In the form of a linear embedded layer,In order to detect the number of targets,As a dimension of the features,In order to detect the target data of the object,Which represents the characteristics of the object of detection,A learnable embedded vector representing an empty object,In order to detect the characteristic dimension of the object,Is active target data;

Step S42: TADN module processing, to AndTADN processing is carried out to obtain an output vector set;

further, in step S42, the TADN module process specifically includes the following steps:

step S421: encoder-decoder processing, using encoder pairs Processing and using decoder pairsThe process is performed using the following formula:

；

；

Wherein, AndProcessing operations of the encoder and decoder respectively,AndThe parameters of the encoder and decoder respectively,AndRespectively isAndIs provided;

Step S422: transformer model generation, using two independent Transformer models to generate respectively AndThe formula used is as follows:

；

；

Wherein, AndRespectively isAndThe parameters of the transducer model are respectively as followsAnd,AndRespectively isAndProcessing the output vector set through a transducer model;

step S43: calculating an allocation score matrix, calculating AndThe dot product is scaled between the two to obtain an allocation score matrix, and the formula is as follows:

；

Wherein, Assigning a scoring matrix;

step S44: determining a final allocation, performing argmax operation for each row of the allocation score matrix, finding the best allocation, using the following formula:

；

Wherein, The direction in which the dimension is 1 is indicated,As a set of integers,,For the final selected matching index set.

Furthermore, in the data analysis and prediction module, the LOA-LSTM model method is used for learning and training, and the method specifically comprises the following steps:

step P1: data preprocessing and input representation, and collecting cargo transportation related information, wherein the cargo transportation related information comprises weather conditions, holiday influences, traffic conditions and alarm information of a transportation and distribution tracking module, and the information is integrated and packaged into a picture form to form a two-dimensional tensor;

step P2: continuous feature extraction, namely processing the two-dimensional tensor by using a convolutional layer of CNN, fusing daily cargo transportation related information and extracting features;

Step P3: the feature abstraction and compression, further process the feature extracted by the convolution layer by using the pooling layer, generate complex and generalized features, and mark as high-level abstract features;

step P4: the LOA-LSTM model learning training is carried out, and high-level abstract features are input into LSTM units in the LOA-LSTM model to extract final feature vectors;

Step P5: optimizing and selecting parameters, namely optimizing the parameters of the LOA-LSTM model by using an optimized Lepski self-adaptive parameter selection method;

further, in step P5, the parameters of the LOA-LSTM model are optimized by using an optimized Lepski adaptive parameter selection method, which specifically comprises the following steps:

Step P51: the estimator representation defines a set of estimators and their corresponding regulars, using the following formulas:

；

Wherein, In order to generalize the regularization function,The feature matrix is represented by a matrix of features,The feature vector of the object is represented,Indicating that all probabilities do not exceed 1Is set at the maximum upper bound of (2),As a dual function;

Step P52: a loss function and a gradient, wherein the loss function and the loss function gradient are used for measuring the performance of the general estimator and adjusting the loss function gradient of the parameters of the estimator;

step P53: adjusting parameter set, setting an orderly adjusting parameter set WhereinAnd the values of the two are all in the range of (0, ++);

Step P54: constructing constraints based on gradient differences, on a given set of tuning parameters, for each Calculate the differenceAndGradient difference of the position estimator and find out that the constraint condition is satisfiedThe formula used is as follows:

；

Wherein, In order to loss-function gradients,AndRepresenting two different regularization parameter values,For regularization parameter ofIs used in the method of the present invention,For regularization parameter ofIs used in the method of the present invention,Representing a dependence onAndFor controlling the upper bound of the gradient difference,Representing an optimal regularized parameter estimate,Represents the value of the argument when minimizing the function,Expressed in a collectionMiddle greater than or equal toTaking the maximum difference in parameters of (a)、；

Step P55: selecting a function constant and obtaining parameter values to determine a sufficiently large function constantSo that the function constantSelect in step P54When the inequality constraint is satisfied, the specific constraint conditions are as follows:

；

；

Wherein, In order to adjust the parameters of the device,Indicating the selected adjustment parameter asIs an estimator of (a)，As a real parameter of the data set,Is a constant reflecting the relationship between gradient differences and tuning parameters.

By adopting the scheme, the beneficial effects obtained by the invention are as follows:

(1) Aiming at the problems that the prior RFID technology cannot effectively track, match and position and track the entity and RFID labels of cargo transportation, so that a supply chain manager is difficult to realize seamless tracing and fine management and control on the whole logistics process, and the overall efficiency and accuracy of logistics management are seriously influenced, the scheme adopts an advanced TADN technology, aims at realizing fine and accurate matching and tracking on the state of the cargo in the transportation process, can continuously acquire the position information and state change of the cargo by installing the RFID labels on the cargo, realizes seamless monitoring on the whole cargo flowing process, and simultaneously can effectively capture the visual characteristics of the cargo by carrying out deep learning and intelligent analysis on a high-definition video monitoring image sequence, accurately trace the motion path of the cargo in a complex transportation environment, thereby effectively guaranteeing the safety and efficiency of logistics transportation links and realizing effective management and control from cargo loading to unloading of each link;

(2) Aiming at the problem that logistics resources are difficult to accurately allocate when a supply chain manager cannot accurately predict the supply chain manager in the face of emergency or complex conditions, so that logistics efficiency is low, the LOA-LSTM model is adopted to predict the time required by cargo transportation of the supply chain and the occurrence probability of the emergency, real-time data monitoring and historical data are combined to analyze, an optimal transportation path and loading strategy are selected, efficient optimization of the logistics transportation path and reasonable distribution of resources are achieved, overall efficiency of logistics transportation is effectively improved, operation cost is reduced, and competitiveness and response speed of enterprises are enhanced.

Drawings

FIG. 1 is a schematic diagram of an RFID-based logistics and supply chain management system provided by the present invention;

FIG. 2 is a schematic flow diagram of TADN technology;

FIG. 3 is a flow chart of step S4;

FIG. 4 is a flow chart of the LOA-LSTM model method;

FIG. 5 is a flow chart of step P5;

the accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In the description of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," "outer," and the like indicate orientation or positional relationships based on those shown in the drawings, merely to facilitate description of the invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus should not be construed as limiting the invention.

Referring to fig. 1, the RFID-based logistics and supply chain management system provided by the present invention includes a data acquisition and integration module, a transportation and distribution tracking module, a data analysis and prediction module, and a supply chain coordination and visualization module;

The data acquisition and integration module comprises the steps of generating, encoding and reading RFID labels, deploying RFID readers at each key node of a warehouse, acquiring inventory information and warehouse in-out information of cargoes in the warehouse, sorting and gathering the inventory information and the warehouse in-out information to form a structured database record, wherein the database record comprises information related to transportation such as historical transportation time data, real-time transportation data and the like, and updating the database record in real time;

The generation and coding of the RFID tags comprise the creation, initialization and coding of the RFID tags, each tag is guaranteed to have a unique identifier, and the reading identification is that goods attached with the RFID tags are read in real time and in batches through a fixed or mobile RFID reader-writer, so that automatic data acquisition of links such as warehouse entry, warehouse exit, inventory, sorting and the like is realized;

The transportation and distribution tracking module is used for tracking and updating the in-transit state of goods in real time by scanning RFID labels and monitoring equipment on packages and using TADN technology, meanwhile, the system has an abnormal condition early warning function, when an abnormal condition occurs in the transportation process, the system can rapidly send an alarm according to the occurring abnormal condition and transmit alarm information to the data analysis and prediction module for analysis and prediction, and the transportation strategy is timely adjusted, wherein the abnormal condition comprises the conditions of delay, deviation from a preset path, overdrive, weather change, traffic accident and the like in the transportation process;

The data analysis and prediction module performs learning training by using an LOA-LSTM model method, collects and integrates various factors influencing the stability of a supply chain, predicts the probability of an emergency in the transportation process and performs route planning and distribution by multi-dimensional data input and learning to obtain a prediction result, wherein the prediction result not only comprises the probability of an abnormal condition in the transportation process, but also comprises cargo distribution timeliness, cargo quantity and time for reaching each logistics node under normal conditions;

The supply chain coordination and visualization module builds an RFID-based supply chain coordination platform, so that physical distribution activity data can be acquired and shared in real time among partners, a coordination decision is promoted, the physical distribution activity data is converted into a visual and understandable chart and map form, and a whole-course visual supply chain monitoring interface is provided.

In a second embodiment, referring to fig. 1 and 2, the method for tracking and updating the in-transit state of the cargo in real time by using TADN technology in the transportation and distribution tracking module specifically includes the following steps:

Step S1: in the detection stage, a pre-trained target detector is used for analyzing video content and information recorded in a database in real time in monitoring equipment, detecting detection entities related to transportation information, and acquiring position information of each detection entity;

step S2: in the feature extraction stage, for each detection entity, extracting high-level features by using a pre-trained ResNet network, and extracting deep learning features of an image area where the high-level features are positioned to obtain feature information;

Step S3: the target representation construction stage integrates the position information and the characteristic information, constructs the representation of the current detection entity, namely a detection target, and updates the current state of the detection entity which is identified and tracked previously, namely the current active target for short;

step S4: TADN, processing the detection target and the existing active target by using TADN technology, calculating an appearance similarity matrix, and judging the matching relationship between the detection target and the existing active target;

Step S5: and in the matching and target updating stage, target matching is carried out according to the result output by TADN, the motion trail and the appearance model of the detection target are updated, repeated matching results are removed, a new target trail segment is created for all unmatched detection targets, the motion model and the appearance model are initialized, and new target tracking is started.

An embodiment III, referring to FIG. 2 and FIG. 3, based on the above embodiment, in step S4, the detected target and the existing active target are processed by using TADN technology, specifically including the following steps:

Step S41: preprocessing input data, processing a detection target and an existing active target by using a linear embedding layer, using a real boundary box position for a normal detection target, and predicting a new boundary box position for an abnormal detection target by using a motion model. Adding a learnable embedded vector representing an empty target to the existing active target to generate detection target data and active target data, wherein the formula is as follows:

；

；

Wherein, The result of the detection target is indicated,In the form of a linear embedded layer,In order to detect the number of targets,As a dimension of the features,In order to detect the target data of the object,Which represents the characteristics of the object of detection,A learnable embedded vector representing an empty object,In order to detect the characteristic dimension of the object,Is active target data;

Step S42: TADN module processing, to AndTADN processing is carried out to obtain an output vector set;

In step S42, the TADN module process specifically includes the following steps:

step S421: encoder-decoder processing, using encoder pairs Processing and using decoder pairsThe process is performed using the following formula:

；

；

Wherein, AndProcessing operations of the encoder and decoder respectively,AndThe parameters of the encoder and decoder respectively,AndRespectively isAndIs provided;

Step S422: transformer model generation, using two independent Transformer models to generate respectively AndThe formula used is as follows:

；

；

Wherein, AndRespectively isAndThe parameters of the transducer model are respectively as followsAnd,AndRespectively isAndProcessing the output vector set through a transducer model;

step S43: calculating an allocation score matrix, calculating AndThe dot product is scaled between the two to obtain an allocation score matrix, and the formula is as follows:

；

Wherein, Assigning a scoring matrix;

step S44: determining final allocation, performing argmax operation on each row of an allocation score matrix, calculating cross entropy loss, normalizing a loss function value, and obtaining an optimal allocation mode, wherein the formula is as follows:

；

Wherein, The direction in which the dimension is 1 is indicated,As a set of integers,，For the final selected matching index set.

Aiming at the problems that the prior RFID technology cannot effectively track, match and position and track the entity and RFID labels for cargo transportation, so that a supply chain manager is difficult to realize seamless tracing and precise management and control on the whole logistics process, and the overall efficiency and the precision of logistics management are seriously influenced, the scheme adopts the advanced TADN technology, aims at realizing precise and precise matching and tracking on the state of the cargo in the transportation process, can continuously acquire the position information and the state change of the cargo by installing the RFID labels on the cargo, realizes seamless monitoring on the whole cargo flowing process, and simultaneously, can effectively capture the visual characteristics of the cargo by deep learning and intelligent analysis on a high-definition video monitoring image sequence, precisely trace the motion path of the cargo in a complex transportation environment, thereby effectively guaranteeing the safety and the efficiency of logistics transportation links and realizing the effective management and control from cargo loading to unloading of each link.

Referring to fig. 1 and 4, in the data analysis and prediction module, the embodiment uses the LOA-LSTM model method to perform learning training, which specifically includes the following steps:

step P1: data preprocessing and input representation, and collecting cargo transportation related information, wherein the cargo transportation related information comprises weather conditions, holiday influences, traffic conditions and alarm information of a transportation and distribution tracking module, and the information is integrated and packaged into a picture form to form a two-dimensional tensor as input of an LOA-LSTM method;

Step P2: continuous feature extraction, processing the two-dimensional tensor by using a 3 multiplied by 3 filter in the CNN convolution layer, fusing daily cargo transportation related information and extracting higher-level features to represent cargo transportation states of each day, and capturing a trend mode of time sequence data in cargo transportation;

Step P3: feature abstraction and compression, applying 64 filters in the second layer of the LOA-LSTM model, each filter continuously filtering data of three days of cargo transportation related information, performing 2 x2 max pooling operation in the third layer, constructing more complex features and aggregating information for longer time intervals, then adopting a convolution layer with 128 filters, and then constructing higher-level features using another pooling layer similar to the first pooling layer, convolution and pooling operation, aggregating available information in a certain period of time, and combining low-level input features into higher-level input features;

Step P4: the LOA-LSTM model learning training is carried out, the LSTM unit is used for processing high-level abstract features obtained through the convolution layer and the pooling layer, long-term dependency relationship in time sequence data is obtained, the features after LSTM processing are flattened into one-dimensional vectors and are input into the full-connection layer, the full-connection layer maps the features to a prediction result, and expected fluctuation conditions in future cargo transportation are output;

step P5: and optimizing and selecting parameters, and optimizing the parameters of the LSTM model by using an optimized Lepski self-adaptive parameter selection method.

Fifth embodiment, referring to fig. 4 and 5, based on the above embodiment, in step P5, the parameters of the LSTM model are optimized using an optimized Lepski adaptive parameter selection method, which specifically includes the following steps:

Step P51: the estimator representation defines a set of estimators and their corresponding regulars, using the following formulas:

；

Wherein, In order to generalize the regularization function,The feature matrix is represented by a matrix of features,The feature vector of the object is represented,Indicating that all probabilities do not exceed 1Is set at the maximum upper bound of (2),As a dual function;

Step P52: a loss function and a gradient, wherein the loss function and the loss function gradient are used for measuring the performance of the general estimator and adjusting the loss function gradient of the parameters of the estimator;

step P53: adjusting parameter set, setting an orderly adjusting parameter set WhereinAnd the values of the two are all in the range of (0, ++);

Step P54: constructing constraints based on gradient differences, on a given set of tuning parameters, for each Calculate the differenceAndGradient difference of the position estimator and find out that the constraint condition is satisfiedThe formula used is as follows:

；

Wherein, In order to loss-function gradients,AndRepresenting two different regularization parameter values,For regularization parameter ofIs used in the method of the present invention,For regularization parameter ofIs used in the method of the present invention,Representing a dependence onAndFor controlling the upper bound of the gradient difference,Representing an optimal regularized parameter estimate,Represents the value of the argument when minimizing the function,Expressed in a collectionMiddle greater than or equal toTaking the maximum difference in parameters of (a)、；

Step P55: selecting a function constant and obtaining parameter values to determine a sufficiently large function constantSo that the function constantSelect in step P54When the inequality constraint is satisfied, the specific constraint conditions are as follows:

；

；

Wherein, In order to adjust the parameters of the device,Indicating the selected adjustment parameter asIs an estimator of (a)，As a real parameter of the data set,Is a constant reflecting the relationship between gradient differences and tuning parameters.

Aiming at the problem that logistics resources are difficult to accurately allocate when a supply chain manager cannot accurately predict the supply chain manager in the face of emergency or complex conditions, so that logistics efficiency is low, the LOA-LSTM model is adopted to predict the time required by cargo transportation of the supply chain and the occurrence probability of the emergency, real-time data monitoring and historical data are combined to analyze, an optimal transportation path and loading strategy are selected, efficient optimization of the logistics transportation path and reasonable distribution of resources are achieved, overall efficiency of logistics transportation is effectively improved, operation cost is reduced, and competitiveness and response speed of enterprises are enhanced.

It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.

Claims (6)

1. RFID-based logistics and supply chain management system, characterized in that: the system comprises a data acquisition and integration module, a transportation and distribution tracking module, a data analysis and prediction module and a supply chain coordination and visualization module;

The data acquisition and integration module comprises the steps of generating, encoding and reading RFID labels, deploying RFID readers at each key node of a warehouse, acquiring inventory information and warehouse entry information of goods in the warehouse, forming a structured database record, and updating the database record in real time;

The transportation and distribution tracking module tracks and updates the in-transit state of goods in real time by scanning RFID labels and monitoring equipment on packages and using TADN technology, meanwhile, the transportation and distribution tracking module has an abnormal condition early warning function, and when abnormal conditions occur in the transportation process, the transportation and distribution tracking module rapidly sends out an alarm and transmits the alarm information to the data analysis and prediction module for analysis and prediction, and timely adjusts the transportation strategy;

The data analysis and prediction module performs learning training by using an LOA-LSTM model method, collects and integrates various factors and alarm information affecting the stability of a supply chain, and performs probability prediction on emergent conditions through multi-dimensional data input and learning to obtain a prediction result;

The supply chain coordination and visualization module builds an RFID-based supply chain coordination platform, so that physical distribution activity data can be acquired and shared in real time among partners, a coordination decision is promoted, the physical distribution activity data is converted into a visual and understandable chart and map form, and a whole-course visual supply chain monitoring interface is provided.

2. The RFID-based logistics and supply chain management system of claim 1, wherein: in the transportation and distribution tracking module, the TADN technology is used for tracking and updating the in-transit state of the goods in real time, and the method specifically comprises the following steps:

Step S1: in the detection stage, a pre-trained target detector is used for analyzing video content and information recorded in a database in real time in monitoring equipment, detecting detection entities related to transportation information, and acquiring position information of each detection entity;

step S2: in the feature extraction stage, for each detection entity, extracting high-level features by using a pre-trained ResNet network, and extracting deep learning features of an image area where the high-level features are positioned to obtain feature information;

Step S3: the target representation construction stage integrates the position information and the characteristic information, constructs the representation of the current detection entity, namely a detection target, and updates the current state of the detection entity which is identified and tracked previously, namely the current active target for short;

step S4: TADN, processing the detection target and the existing active target by using TADN technology, calculating an appearance similarity matrix, and judging the matching relationship between the detection target and the existing active target;

Step S5: and in the matching and target updating stage, target matching is carried out according to the result output by TADN, the motion trail and the appearance model of the detection target are updated, and repeated matching results are removed.

3. The RFID-based logistics and supply chain management system of claim 2, wherein: in step S4, the detected target and the existing active target are processed by using TADN technology, which specifically includes the following steps:

Step S41: preprocessing input data, processing a detection target and an existing active target by using a linear embedding layer, adding a learnable embedding vector representing an empty target to the existing active target, and generating detection target data and active target data, wherein the formula is as follows:

；

；

Wherein, The detection target is indicated to be a target of the detection,In the form of a linear embedded layer,In order to detect the number of targets,As a dimension of the features,In order to detect the target data of the object,Which represents the characteristics of the object of detection,A learnable embedded vector representing an empty object,In order to detect the characteristic dimension of the object,Is active target data;

Step S42: TADN module processing, to AndTADN processing is carried out to obtain an output vector set;

step S43: calculating an allocation score matrix, calculating AndThe dot product is scaled between the two to obtain an allocation score matrix, and the formula is as follows:

；

Wherein, Assigning a scoring matrix;

step S44: determining a final allocation, performing argmax operation for each row of the allocation score matrix, finding the best allocation, using the following formula:

；

Wherein, The direction in which the dimension is 1 is indicated,As a set of integers,,For the final selected matching index set.

4. The RFID-based logistics and supply chain management system of claim 3 wherein: in step S42, the TADN module process specifically includes the following steps:

step S421: encoder-decoder processing, using encoder pairs Processing and using decoder pairsThe process is performed using the following formula:

；

；

Wherein, AndProcessing operations of the encoder and decoder respectively,AndThe parameters of the encoder and decoder respectively,AndRespectively isAndIs provided;

Step S422: transformer model generation, using two independent Transformer models to generate respectively AndThe formula used is as follows:

；

；

Wherein, AndRespectively isAndThe parameters of the transducer model are respectively as followsAnd,AndRespectively isAndThe output vector set is processed by a transducer model.

5. The RFID-based logistics and supply chain management system of claim 1, wherein: in the data analysis and prediction module, the LOA-LSTM model method is used for learning and training, and the method specifically comprises the following steps:

step P1: data preprocessing and input representation, and collecting cargo transportation related information, wherein the cargo transportation related information comprises weather conditions, holiday influences, traffic conditions and alarm information of a transportation and distribution tracking module, and the information is integrated and packaged into a picture form to form a two-dimensional tensor;

step P2: continuous feature extraction, namely processing the two-dimensional tensor by using a convolutional layer of CNN, fusing daily cargo transportation related information and extracting features;

Step P3: the feature abstraction and compression, further process the feature extracted by the convolution layer by using the pooling layer, generate complex and generalized features, and mark as high-level abstract features;

step P4: the LOA-LSTM model learning training is carried out, and high-level abstract features are input into LSTM units in the LOA-LSTM model to extract final feature vectors;

Step P5: and optimizing and selecting parameters, and optimizing the parameters of the LOA-LSTM model by using an optimized Lepski self-adaptive parameter selection method.

6. The RFID-based logistics and supply chain management system of claim 5, wherein: in step P5, the parameters of the LOA-LSTM model are optimized by using an optimized Lepski self-adaptive parameter selection method, and the method specifically comprises the following steps:

Step P51: the estimator representation defines a set of estimators and their corresponding regulars, using the following formulas:

；

Wherein, In order to generalize the regularization function,The feature matrix is represented by a matrix of features,The feature vector of the object is represented,Indicating that all probabilities do not exceed 1Is set at the maximum upper bound of (2),As a dual function;

Step P52: a loss function and a gradient, wherein the loss function and the loss function gradient are used for measuring the performance of the general estimator and adjusting the loss function gradient of the parameters of the estimator;

step P53: adjusting parameter set, setting an orderly adjusting parameter set WhereinAnd the values of the two are all in the range of (0, ++);

Step P54: constructing constraints based on gradient differences, on a given set of tuning parameters, for each Calculate the differenceAndGradient difference of the position estimator and find out that the constraint condition is satisfiedThe formula used is as follows:

；

Wherein, In order to loss-function gradients,AndRepresenting two different regularization parameter values,For regularization parameter ofIs used in the method of the present invention,For regularization parameter ofIs used in the method of the present invention,Representing a dependence onAndIs used to determine the threshold function of (2),Representing an optimal regularized parameter estimate,Represents the value of the argument when minimizing the function,Expressed in a collectionMiddle greater than or equal toTaking the maximum difference in parameters of (a)、；

Step P55: selecting a function constant and obtaining parameter values to determine a sufficiently large function constantSo that the function constantSelect in step P54When the inequality constraint is satisfied, the specific constraint conditions are as follows:

；

；

Wherein, In order to adjust the parameters of the device,Indicating the selected adjustment parameter asIs an estimator of (a)，As a real parameter of the data set,Is a constant reflecting the relationship between gradient differences and tuning parameters.