CN114596102B - Block chain-based anti-counterfeiting traceability federated learning training method and device - Google Patents

Abstract

The invention discloses an anti-counterfeiting traceability federal learning training method and device based on a block chain, which comprises the steps of building a block chain network, chaining commodity certified product traceability data, manufacturing a commodity target detection data set, training a commodity target detection model, judging the identification accuracy and recall rate of the commodity target detection model, training a global federal learning model after label name alignment and data access are carried out based on a transverse federal learning target detection algorithm through a federal learning network platform, chaining target commodity description to the block chain network for disclosure, effectively carrying out commodity fake traceability through block chain federal learning and artificial intelligence technology, improving the accuracy of the model, and reducing the privacy problem of mixed data.

Description

Block chain-based anti-counterfeiting traceability federated learning training method and device

Technical Field

The invention relates to the technical field of block chain traceability, in particular to an anti-counterfeiting traceability federal learning training method and device based on a block chain.

Background

In the digital economic development planning, the technical application of artificial intelligence, block chains and the like and the fields of intelligent logistics, intelligent cities, intelligent government affairs and the like are mentioned to realize innovative data element development.

The counterfeit and shoddy commodities are forbidden frequently, and the harm to the country and the society is very serious.

Most of the existing methods for checking counterfeit goods or tracing the source of goods store information of goods in a block chain through two-dimensional codes, and scan the two-dimensional codes to obtain information of corresponding goods, so as to judge whether the goods are counterfeit or not, but the two-dimensional codes are easy to copy, and if each goods for sale is circulated to the market, whether the goods are counterfeit or not can be judged through manual scanning of the two-dimensional codes, so that certain trouble is brought to merchants and consumers.

Disclosure of Invention

Aiming at least one technical problem, the invention aims to provide an anti-counterfeiting traceability federated learning training method and device based on a block chain.

In one aspect, an embodiment of the present invention includes an anti-counterfeiting traceability federated learning training method based on a blockchain, including:

building a block chain network, and chaining the commodity certified product traceability data;

making a commodity target detection data set, combining a target detection framework based on the commodity target detection data set, and training a commodity target detection model;

acquiring a full-network authorized camera, traversing the video stream read by the full-network authorized camera through the commodity target detection model, intercepting a picture according to a preset frame frequency, carrying out target identification, and identifying whether a target commodity exists in the picture; if the target commodity exists, judging whether the target commodity circulated to the position where the camera is located exists in the commodity certified source data which is linked in the block chain network, and if not, judging that the target commodity circulated to the position where the camera is located is a suspected counterfeit commodity;

acquiring the position of the suspected counterfeit commodity corresponding to the camera, and judging whether the target detection result of the suspected counterfeit commodity is correct or not through manual inspection or server background inquiry whether a counterfeit record exists in the position or not;

judging the identification accuracy and recall rate of the commodity target detection model, inputting the description, efficacy and appearance of a target commodity when the recall rate is lower than a first threshold value, searching commodities with similarity higher than a second threshold value with the target commodity as first similar commodities on the Internet based on a text and image similarity algorithm, acquiring the name and contact way of a first similar commodity manufacturer, and sending a federal learning request to the first similar commodity manufacturer;

through a federated learning network platform, on the basis of a horizontal federated learning target detection algorithm, after label name alignment and data access are carried out, a global federated learning model is trained;

inquiring whether a pseudo-inferior product record exists in the position through the server background, and calculating the region weight of the pseudo-inferior product to obtain a region weight table of the pseudo-inferior product;

acquiring the whole-network authorized camera, traversing the video stream read by the whole-network authorized camera based on the global federated learning model, intercepting a picture according to a preset frame frequency, carrying out target identification, identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture, and really predicting whether the commodity manufacturer corresponding to the result identifies the truth;

counting the accuracy of the global federated learning model based on the prediction result of the global federated learning model in the whole network authorized camera, and adjusting the methods of label name alignment preprocessing and data access preprocessing when the accuracy is lower than a fifth threshold preset by a manufacturer;

retraining the global federal learning model, counting the accuracy of the global federal learning model until the accuracy of the global federal learning model is higher than a fifth threshold preset by a manufacturer, and linking a data access preprocessing method, a label name alignment preprocessing method, training parameters and a training network architecture to the block chain network for disclosure.

Further, the building of the block chain network links the good commodity traceability data, and includes: building a block chain network based on a Baidu super chain SuperChain, and chaining commodity certified product traceability data, wherein the commodity certified product traceability data comprises: the method comprises the steps that flow direction information of commodity distribution from production and processing to warehouse-in and warehouse-out is obtained, the commodity is currently sold in a store and the geographic position of the commodity, and the commodity certified product traceability data is identified by scanning a two-dimensional code label on the commodity to read data on a chain.

Furthermore, the full-network authorized camera is a camera authorized by the full network to a manufacturer for target detection, and after authorization, the manufacturer legally obtains video stream data of the camera, wherein the video stream data comprises a shopping mall street camera, a shopping mall internal camera, a large-scale shopping mall super camera and a shopping square camera.

Further, the obtaining of the position of the suspected counterfeit commodity corresponding to the camera, and determining whether the target detection result of the suspected counterfeit commodity is correct by manually checking or querying whether a counterfeit record exists in the position by a server background includes:

the manual checking of whether a counterfeit record exists includes: the method comprises the steps of detecting a position of a camera corresponding to a suspected counterfeit commodity, and determining whether a counterfeit commodity exists or not, or determining whether a target commodity is included in a picture corresponding to the suspected counterfeit commodity or not, if no counterfeit commodity exists, determining that a target detection result of the suspected counterfeit commodity is incorrect, and if a counterfeit commodity exists, determining that a target detection result of the suspected counterfeit commodity is correct;

the server background inquires whether a counterfeit record exists in the position, and the method specifically comprises the following steps: storing the positions of the various counterfeit inferior products through a server background, and judging that the target detection is correct when the positions of the suspected counterfeit commodities corresponding to the cameras exist at the positions of the various counterfeit inferior products;

the server background stores the positions of the false inferior products in various places, namely the positions of the false inferior products verified after receiving the reported information or the positions of the false inferior products verified and judged by other manual work.

Further, after the label name alignment and data access are performed through the federated learning network platform based on a horizontal federated learning target detection algorithm, a global federated learning model is trained, which includes:

the federated learning network platform includes: a client participant, a server and a federal learning framework;

the client participation in the federated learning network platform is two or more than two: respectively a target commodity manufacturer, the first similar commodity manufacturer and other similar commodity manufacturers;

a server side in the federated learning network platform is a cloud server, and a federated learning framework is a micro-controller;

after a client in the federated learning network platform deploys a federated learning framework, the following steps are executed, including:

a. preprocessing a target detection data set stored locally at a client;

preprocessing a target detection data set stored locally at a client, wherein the preprocessing comprises the preprocessing of label name alignment;

the aligning of the label name comprises: unifying label names of the target commodity manufacturer and a first similar commodity manufacturer in a target detection data set stored in a client into a fused label commodity;

the aligning of the label name of the label further comprises: keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in a target detection data set stored in a client, wherein the label names are respectively a target commodity and a first target commodity;

the preprocessing of the target detection data set stored locally at the client further comprises: data access preprocessing:

the data access preprocessing comprises the following steps:

when the label names aligned with the label names are not fusion label commodities, traversing the background similarity, the scene similarity and the light similarity of the target detection data set of the target commodities and the target detection data set of the first similar commodity manufacturer, and when the similarity is higher than a third threshold value, removing the data corresponding to the target detection data set of the first similar commodity manufacturer and not listing the data in a data set of federal learning training;

b. initiating a federal learning training task to other clients;

c. participating in a federal learning task, and training a federal learning model together with other clients;

d. deploying a global federated learning model to predict and infer locally;

the server side in the federal learning network platform is provided by a cloud server, and executes the following steps of: monitoring the connection condition of a client side participant in the federated learning network platform in real time; aggregating the federated learning model uploaded by the client in the federated learning network platform; selecting a client in the federated learning network platform to participate in federated learning training; and uploading and broadcasting a global federal learning model to a client in the federal learning network platform.

Further, the querying, by the server background, whether a pseudo-inferior product record exists at the position, and calculating the region weight where the pseudo-inferior product is located to obtain a region weight table where the pseudo-inferior product is located, includes:

inquiring the positions of the pseudo inferior products of all regions through the server background, and counting the number of the pseudo inferior products in each region, wherein the regions take regions, cities or streets as units, and a region weight table where the pseudo inferior products are located is obtained according to the number of the pseudo inferior products in combination with a normalization algorithm;

the region weight table where the pseudo inferior products are located also comprises the region weight table where the pseudo inferior products of each manufacturer are located;

further, the acquiring the full-network authorized camera, traversing the video stream read by the full-network authorized camera based on the global federated learning model, capturing a picture according to a preset frame frequency, performing target identification, identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture, and really predicting the authenticity of the commodity manufacturer corresponding to the result, includes:

the identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture further includes:

a1: identifying label text information in the picture by ocr technology, determining commodity names, judging whether the commodity names corresponding to the global federated learning model prediction result are consistent or not, and if not, judging that the prediction result is not trusted; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

a2: when the global federal learning model prediction result is the target commodity or the first similar commodity, combining the region weight table of the false inferior commodity, obtaining a prediction result value according to the confidence value of the global federal learning model prediction result and the weight value of a commodity manufacturer corresponding to the prediction result in the region weight table of the false inferior commodity, and judging that the prediction result is not reliable when the prediction result value is lower than a preset fourth threshold value; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

a3: when the global federal learning model prediction result is the fusion tag commodity, determining a commodity with higher regional weight as a predicted commodity in the target commodity and the first target commodity according to a regional weight table where the fake inferior goods are located and the location of a camera corresponding to the picture, and distributing a manufacturer corresponding to the predicted commodity to carry out home-going authentication;

the method for identifying the truth of the commodities distributed and predicted by the corresponding manufacturers comprises the following steps:

b1: the manufacturer sends reminding information to a sales merchant according to the location of the camera corresponding to the picture, the sales merchant scans the two-dimensional code of the predicted commodity to determine whether the predicted commodity is a genuine commodity or not and feeds back the information, and if the predicted commodity is not a genuine commodity, the sales merchant checks the goods and obtains evidence at home;

further, the method for conducting statistics on the accuracy of the global federal learning model based on the prediction result of the global federal learning model in the full-network authorized cameras includes the steps that when the accuracy is lower than a fifth threshold preset by a manufacturer, the method for conducting alignment preprocessing on the name of the label and conducting data access preprocessing on the label is adjusted, and the method includes the following steps:

c1: the method for adjusting data access preprocessing comprises the following steps: when the label name of the global federal learning model is not a fusion label commodity, traversing the background similarity, the scene similarity and the light similarity of the target commodity target detection data set and the first similar commodity manufacturer target detection data set, and when the similarity is higher than a third threshold value, removing the data corresponding to the first similar commodity manufacturer target detection data set and not listing the data in a federal learning training data set;

d1: the preprocessing method for adjusting the alignment of the name of the label comprises the following steps: judging the overlapping rate of the target commodity and the first similar commodity in each region based on the region weight table where the pseudo inferior product is located, and when the overlapping rate is higher than a sixth threshold value, keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in a commodity target detection data set stored in a client, wherein the label names are respectively the target commodity and the first target commodity;

and when the overlapping rate is lower than a sixth threshold value, unifying the label names of the target commodity manufacturer and the first similar commodity manufacturer in the commodity target detection data set stored in the client into a fused label commodity.

In another aspect, an embodiment of the present invention further includes a computer apparatus, including a memory and a processor, where the memory is used to store at least one program, and the processor is used to load the at least one program to perform the method of the embodiment.

The invention has the beneficial effects that:

1. when the recall rate is lower than a first threshold value, inputting description, efficacy and appearance of a target commodity, searching commodities with similarity higher than a second threshold value with the target commodity on the Internet as first similar commodities based on a text and image similarity algorithm, acquiring names and contact ways of manufacturers of the first similar commodities, sending a federal learning request to the manufacturers of the first similar commodities, and judging whether the manufacturers of the first similar commodities have a first similar commodity target detection data set according to request return information; similar commodities can be obtained for federal learning only when the commodity target detection model recall rate is lower than a first threshold value, so that a certain calculated amount is reduced, and calculation resources are saved.

2. Through a federated learning network platform, on the basis of a horizontal federated learning target detection algorithm, after label name alignment and data access are carried out, a global federated learning model is trained, the data sample sources are expanded, and the generalization capability of the global federated learning model is improved.

3. According to the region weight table where the fake inferior products are located and the location of the camera corresponding to the picture, determining the commodity with higher region weight as the predicted commodity in the target commodity and the first target commodity, and allocating the manufacturer corresponding to the predicted commodity to carry out home-entry authentication, so that the data privacy safety is improved, the commercial privacy problem caused by confusion of the fake inferior products of the commodity is eliminated, and the privacy exposure condition and the wrong allocation condition of the manufacturer are reduced.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments of the present application will be briefly described below. It is obvious that the drawings described below are only some embodiments of the application, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.

Fig. 1 is a schematic flowchart of an anti-counterfeiting traceability federated learning training method based on a block chain according to an embodiment of the present application;

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In the description of the present application, it is to be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate orientations or positional relationships based on those shown in the drawings, merely for convenience of description and simplicity of description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be considered limiting of the present application. Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or to implicitly indicate the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of the present application, "a plurality" means two or more unless specifically limited otherwise.

In this application, the word "exemplary" is used to mean "serving as an example, instance, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is presented to enable any person skilled in the art to make and use the application. In the following description, details are set forth for the purpose of explanation. It will be apparent to one of ordinary skill in the art that the present application may be practiced without these specific details. In other instances, well-known structures and processes are not set forth in detail in order to avoid obscuring the description of the present application with unnecessary detail. Thus, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

It should be noted that, since the method in the embodiment of the present application is executed in the electronic device, the processing objects of each electronic device all exist in the form of data or information, for example, time, which is substantially time information, and it is understood that, if the size, the number, the position, and the like are mentioned in the following embodiments, all corresponding data exist so as to be processed by the electronic device, and details are not described herein.

S1: building a block chain network based on a Baidu super chain SuperChain, and chaining commodity certified product traceability data, wherein the commodity certified product traceability data comprises: the commodity is distributed and distributed to flow information from production and processing to warehouse-in and warehouse-out, the commodity is sold in stores and the geographical position at present, and the commodity authentic product traceability data can be identified by scanning the two-dimensional code label on the commodity to read the data on the chain.

S2: and manufacturing a commodity target detection data set, combining an mmdetection target detection framework based on the commodity target detection data set, and training a commodity target detection model.

The commodity object detection data set includes: the manufacturer marks a picture with a target commodity in the picture based on a Lableimg marking tool according to the picture collected by the existing camera to form a commodity target detection data set;

s3: acquiring a full-network authorized camera, traversing a video stream read by the full-network authorized camera through the commodity target detection model, intercepting a picture according to a preset frame frequency, carrying out target identification, identifying whether a target commodity exists in the picture, judging whether the target commodity exists in the commodity certified source data which is linked in the block chain network and is distributed to the position of the camera if the target commodity exists, and judging that the target commodity distributed in the position of the camera is a suspected counterfeit commodity if the target commodity does not exist in the commodity certified source data which is linked in the block chain network;

the whole network authorized camera is a camera authorized by the whole network to a manufacturer for target detection, after authorization, the manufacturer can legally obtain video stream data of the camera, and the whole network authorized camera mainly comprises a shopping mall street camera, a shopping mall internal camera, a large-scale shopping mall camera, a shopping mall camera and the like.

S4: acquiring the position of the suspected counterfeit commodity corresponding to the camera, and judging whether the target detection result of the suspected counterfeit commodity is correct or not by manual inspection or server background inquiry on whether a counterfeit record exists in the position or not;

the manual checking for checking whether a false inferior record exists specifically comprises the following steps: and manually arriving at a store near the position of the camera corresponding to the suspected counterfeit commodity to investigate and judge whether a counterfeit commodity exists or not, or manually judging whether the picture corresponding to the suspected counterfeit commodity comprises the target commodity or not, if no counterfeit commodity exists, judging that the target detection result of the suspected counterfeit commodity is incorrect, and if the counterfeit commodity exists, judging that the target detection result of the suspected counterfeit commodity is correct.

The server background inquires whether a pseudo-inferior product record exists in the position, and the method specifically comprises the following steps: and storing the positions of the various counterfeit inferior products through a server background, and judging that the target detection is correct when the positions of the suspected counterfeit commodities corresponding to the cameras exist at the positions of the various counterfeit inferior products.

The server background stores the positions of the false inferior products in various places, such as the positions of false inferior products which are proved after receiving the reported information or the positions of false inferior products which are proved by other manual examination and judgment.

S5: based on the judging method of S4, obtaining the identification result of the commodity target detection model identification of the preset number of camera shooting pictures, making a commodity target detection verification data set, thereby judging the identification accuracy and the recall rate of the commodity target detection model, when the recall rate is lower than a first threshold value, inputting the description, the efficacy and the appearance of a target commodity, searching commodities with the similarity higher than a second threshold value with the target commodity as first similar commodities on the internet based on a text and image similarity calculation method, obtaining the name and the contact mode of a first similar commodity manufacturer, sending a federal learning request to the first similar commodity manufacturer, and judging whether the first similar commodity manufacturer has a first similar commodity target detection data set or not according to request return information;

the identification accuracy of the commodity target detection model is specifically as follows: the number of times of correct recognition results of the commodity target detection model/the total number of times of recognition and recognition of the commodity target detection model;

the commodity target detection model identification recall rate specifically comprises the following steps: the commodity target detection model identifies the correct times of the identification result/the total number of the false inferior products actually existing in the commodity target detection verification data set;

when the recall rate is lower than a first threshold value, inputting description, efficacy and appearance of a target commodity, searching commodities with similarity higher than a second threshold value with the target commodity on the Internet as first similar commodities based on a text and image similarity algorithm, acquiring names and contact ways of manufacturers of the first similar commodities, sending a federal learning request to the manufacturers of the first similar commodities, and judging whether the manufacturers of the first similar commodities have a first similar commodity target detection data set according to request return information; when the commodity target detection model recall rate is lower than a first threshold value, similar commodities are considered to be obtained for federal learning, certain calculated amount is reduced, and calculation resources are saved.

S6: through a federated learning network platform, on the basis of a horizontal federated learning target detection algorithm, after label name alignment and data access are carried out, a global federated learning model is trained;

the federated learning network platform includes: a client participant, a server and a federal learning framework;

the client participants in the federated learning network platform may be two or more: respectively, a target commodity manufacturer, the first similar commodity manufacturer, and may include other similar commodity manufacturers. The server side in the federal learning network platform can be a cloud server, and the federal learning framework can be a micro-fate.

After a client in the federated learning network platform deploys a federated learning framework, the client mainly executes the following steps:

a. preprocessing a target detection data set stored locally at a client;

preprocessing the target detection data set stored locally at the client, including but not limited to label name alignment preprocessing;

the aligning of the label name comprises: unifying label names of the target commodity manufacturer and a first similar commodity manufacturer in a target detection data set stored in a client into a fused label commodity;

the aligning of the label name further comprises: keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in a target detection data set stored in a client, wherein the label names are a target commodity and a first target commodity respectively;

the preprocessing of the target detection data set stored locally at the client further comprises: data access preprocessing:

the data access preprocessing comprises the following steps:

when the label names aligned with the label names are not fusion label commodities, traversing the background similarity, the scene similarity and the light similarity of the target detection data set of the target commodities and the target detection data set of the first similar commodity manufacturer, and when the similarity is higher than a third threshold value, removing the data corresponding to the target detection data set of the first similar commodity manufacturer and not listing the data in a data set of federal learning training;

b. initiating a federal learning training task to other clients;

c. participating in a federal learning task, and training a federal learning model together with other client sides;

d. deploying a global federated learning model to predict and infer locally;

the server in the federal learning network platform can be provided by a cloud server, and mainly executes the following steps: monitoring the connection condition of a client side participant in the federal learning network platform in real time; aggregating the federated learning model uploaded by the client in the federated learning network platform; selecting a client in the federated learning network platform to participate in federated learning training; and uploading and broadcasting a global federal learning model to a client in the federal learning network platform.

The global federal learning model updated by the federal learning network platform can achieve the following effects:

the method can be distributed to client-side participants in the federated learning network platform, so that the participants benefit, the data sample sources are enlarged, and the generalization capability of a global federated learning model is improved;

s7: inquiring whether a pseudo-inferior product record exists in the position through the server background, calculating the region weight of the pseudo-inferior product, and obtaining a region weight table of the pseudo-inferior product, wherein the method specifically comprises the following steps:

inquiring the positions of the pseudo-inferior products of all regions through the server background, and counting the number of the pseudo-inferior products appearing in each region, wherein the regions can take regions, cities or streets as units, and combining a normalization algorithm according to the number of the pseudo-inferior products to obtain a region weight table where the pseudo-inferior products are located;

the region weight table where the pseudo inferior products are located also comprises the region weight table where the pseudo inferior products of each manufacturer are located;

according to the region weight of the fake inferior products, the existence conditions of the target commodity and the first similar commodity in various regions can be judged and compared, and judgment basis is laid for data fusion selection of subsequent federal learning, so that data is more reasonably utilized, and the federal learning accuracy is improved.

S8: acquiring the whole-network authorized camera, traversing the video stream read by the whole-network authorized camera based on the global federated learning model, intercepting a picture according to a preset frame frequency, carrying out target identification, identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture, and really predicting whether a commodity manufacturer corresponding to a result is up-to-date and identifying authenticity;

the identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture further includes:

a1: identifying label text information in the picture by ocr technology, determining commodity names, judging whether the commodity names corresponding to the global federated learning model prediction result are consistent or not, and if not, judging that the prediction result is not trusted; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

a2: when the global federal learning model prediction result is the target commodity or the first similar commodity, combining the region weight table of the false inferior commodity, obtaining a prediction result value according to the confidence value of the global federal learning model prediction result and the weight value of a commodity manufacturer corresponding to the prediction result in the region weight table of the false inferior commodity, and judging that the prediction result is not reliable when the prediction result value is lower than a preset fourth threshold value; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

a3: when the global federal learning model prediction result is the fusion tag commodity, determining a commodity with higher regional weight as a predicted commodity in the target commodity and the first target commodity according to a regional weight table where the fake inferior goods are located and the location of a camera corresponding to the picture, and distributing a manufacturer corresponding to the predicted commodity to carry out home-going authentication;

the method for identifying the truth of the commodities distributed and predicted by the corresponding manufacturers comprises the following steps:

b1: the manufacturer sends reminding information to a sales merchant according to the location of the camera corresponding to the picture, the sales merchant scans the two-dimensional code of the predicted commodity to determine whether the predicted commodity is a genuine commodity or not and feeds back the information, and if the predicted commodity is not a genuine commodity, the sales merchant checks the goods and obtains evidence at home;

when a sales merchant sells a counterfeit, the sales merchant does not necessarily cooperate with the survey, and therefore needs to check the goods at home.

S9: counting the accuracy of the global federated learning model based on the prediction result of the global federated learning model in the full-network authorized camera in the step S8 and the methods in the steps S4 to S5, and adjusting the method for label name alignment preprocessing and data access preprocessing in the step S6 when the accuracy is lower than a fifth threshold preset by a manufacturer;

c1: the method for adjusting the data access preprocessing comprises the following steps: when the label name of the global federal learning model is not a fusion label commodity, traversing the background similarity, the scene similarity and the light similarity of the target commodity target detection data set and the first similar commodity manufacturer target detection data set, and when the similarity is higher than a third threshold value, removing the data corresponding to the first similar commodity manufacturer target detection data set and not listing the data in a federal learning training data set;

the method can emphasize the background, scene and light which are not possessed by both federal learning and training parties, and improves the accuracy of the model on the premise of improving the generalization ability;

the calculation process of traversing the background similarity, the scene similarity and the light similarity of the target detection data set of the target commodity and the target detection data set of the first similar commodity manufacturer can be carried out in a safe third-party cloud server, so that the data privacy safety is improved;

d1: the preprocessing method for adjusting the alignment of the name of the label comprises the following steps: judging the overlapping rate of the target commodity and the first similar commodity in each region based on the region weight table where the pseudo inferior product is located, and keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in the commodity target detection data set stored in the client when the overlapping rate is higher than a sixth threshold, wherein the label names are the target commodity and the first target commodity respectively;

when the overlapping rate is lower than a sixth threshold value, unifying the label names of the target commodity manufacturer and the first similar commodity manufacturer in the commodity target detection data set stored in the client into a fusion label commodity;

the method can solve the problem of business privacy after home inspection due to confusion of the counterfeit commodities in various regions, and reduces the privacy exposure condition and the error distribution condition of manufacturers.

S10: and (4) retraining the global federated learning model based on the methods of the steps S6-S9, counting the accuracy of the global federated learning model until the accuracy of the global federated learning model is higher than a fifth threshold preset by a manufacturer, and linking the target commodity description to the block chain network for disclosure by a data access preprocessing method and a label name alignment preprocessing method with the accuracy reaching a seventh threshold of a cooperation manufacturer, training parameters and a training network architecture.

In this embodiment, a computer device includes a memory and a processor, where the memory is configured to store at least one program, and the processor is configured to load the at least one program to execute the anti-fake traceability federal learning training method for a block chain in embodiments S1 to S10, so as to achieve the same technical effects as those described in the embodiments.

In this embodiment, a storage medium stores a program executable by a processor, and the program executable by the processor is used to execute the anti-fake traceability federated learning method for the block chain in embodiments S1 to S10 when executed by the processor, so as to achieve the same technical effects as those described in the embodiments.

It should be noted that, unless otherwise specified, when a feature is referred to as being "fixed" or "connected" to another feature, it may be directly fixed or connected to the other feature or indirectly fixed or connected to the other feature. Furthermore, the descriptions of upper, lower, left, right, etc. used in the present disclosure are only relative to the mutual positional relationship of the constituent parts of the present disclosure in the drawings. As used in this disclosure, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. In addition, unless defined otherwise, all technical and scientific terms used in this example have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the description of the embodiments herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in this embodiment, the term "and/or" includes any combination of one or more of the associated listed items.

It will be understood that, although the terms first, second, third, etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one type of element from another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of the present disclosure. The use of any and all examples, or exemplary language ("e.g.," such as "or the like") provided with this embodiment is intended merely to better illuminate embodiments of the invention and does not pose a limitation on the scope of the invention unless otherwise claimed.

It should be recognized that embodiments of the present invention can be realized and implemented in computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with the computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner according to the methods and figures described in the detailed description. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.

Further, operations of processes described in this embodiment can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described in this embodiment (or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) collectively executed on one or more processors, by hardware, or combinations thereof. The computer program includes a plurality of instructions executable by one or more processors.

Further, the method may be implemented in any type of computing platform operatively connected to a suitable interface, including but not limited to a personal computer, mini computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the invention may be embodied in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optically read and/or write storage medium, RAM, ROM, or the like, such that it may be read by a programmable computer, which when read by the storage medium or device, is operative to configure and operate the computer to perform the procedures described herein. Further, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein.

A computer program can be applied to input data to perform the functions described in the present embodiment to convert the input data to generate output data that is stored to a non-volatile memory. The output information may also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including particular visual depictions of physical and tangible objects produced on a display.

The above description is only a preferred embodiment of the present invention, and the present invention is not limited to the above embodiment, and any modifications, equivalent substitutions, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention as long as the technical effects of the present invention are achieved by the same means. The technical solution and/or the embodiments thereof may be variously modified and varied within the scope of the present invention.

Claims (9)

1. An anti-counterfeiting traceability federated learning training method based on a block chain is characterized by comprising the following steps:

building a block chain network, and chaining the commodity certified product traceability data;

making a commodity target detection data set, combining a target detection framework based on the commodity target detection data set, and training a commodity target detection model;

acquiring a full-network authorized camera, traversing the video stream read by the full-network authorized camera through the commodity target detection model, intercepting a picture according to a preset frame frequency, carrying out target identification, and identifying whether a target commodity exists in the picture; if the target commodity exists, judging whether the target commodity circulates to the position where the camera is located in the commodity certified product traceability data which is chained in the block chain network or not, and if not, judging that the target commodity which circulates to the position where the camera is located is suspected to be a fake commodity;

acquiring the position of the suspected counterfeit commodity corresponding to the camera, and judging whether the target detection result of the suspected counterfeit commodity is correct or not through manual inspection or server background inquiry whether a counterfeit record exists in the position or not;

judging the identification accuracy and the recall rate of the commodity target detection model, inputting the description, efficacy and appearance of a target commodity when the recall rate is lower than a first threshold, searching commodities with the similarity higher than a second threshold with the target commodity as first similar commodities on the Internet based on a text and image similarity algorithm, acquiring the name and the contact way of a first similar commodity manufacturer, and sending a federal learning request to the first similar commodity manufacturer;

through a federated learning network platform, on the basis of a horizontal federated learning target detection algorithm, after label name alignment and data access are carried out, a global federated learning model is trained;

inquiring whether a pseudo-inferior product record exists in the position through the server background, and calculating the region weight of the pseudo-inferior product to obtain a region weight table of the pseudo-inferior product;

acquiring the whole-network authorized camera, traversing the video stream read by the whole-network authorized camera based on the global federated learning model, intercepting a picture according to a preset frame frequency, carrying out target identification, identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture, and determining the authenticity of the commodity manufacturer corresponding to the prediction result; the identifying whether the target commodity or the first similar commodity or the fusion label commodity exists in the picture and determining the authenticity identification of the commodity manufacturer corresponding to the prediction result comprises the following steps: when the global federal learning model prediction result is the fusion tag commodity, determining a commodity with higher regional weight as a predicted commodity in the target commodity and the first target commodity according to a regional weight table where the fake inferior goods are located and the location of a camera corresponding to the picture, and distributing a manufacturer corresponding to the predicted commodity to carry out home-going authentication;

counting the accuracy of the global federated learning model based on the prediction result of the global federated learning model in the full-network authorized camera, and adjusting the methods of label name alignment preprocessing and data access preprocessing when the accuracy is lower than a fifth threshold preset by a manufacturer;

retraining the global federal learning model, counting the accuracy of the global federal learning model until the accuracy of the global federal learning model is higher than a fifth threshold preset by a manufacturer, and linking a data access preprocessing method, a label name alignment preprocessing method, training parameters and a training network architecture to the block chain network for disclosure.

2. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the block chain network is built, and the commodity certified product traceability data is linked up, including: building a block chain network based on a Baidu super chain SuperChain, and chaining commodity certified product traceability data, wherein the commodity certified product traceability data comprises: the commodity is distributed and distributed to information from production and processing to warehouse-in and warehouse-out, the commodity is sold in stores and the geographical position at present, and the commodity source tracing data is identified by scanning two-dimensional code labels on the commodity to read the data on the chain.

3. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the whole-network authorized camera is a camera authorized by the whole network to a manufacturer for target detection, and after authorization, the manufacturer legally obtains video stream data of the camera, wherein the video stream data comprise a street-along camera of a sales shop, a camera inside the sales shop, a large-scale business super camera and a shopping square camera.

4. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the obtaining of the position of the suspected counterfeit commodity corresponding to the camera, and judging whether the target detection result of the suspected counterfeit commodity is correct through manual inspection or server background inquiry whether a counterfeit record exists in the position includes:

the manual checking of whether a counterfeit record exists includes: the method comprises the steps of detecting whether a suspected counterfeit commodity is a counterfeit commodity or not, judging whether a counterfeit commodity exists or not, judging whether a target commodity is included in a picture corresponding to the suspected counterfeit commodity or not, judging whether a target detection result of the suspected counterfeit commodity is incorrect or not if the suspected counterfeit commodity does not exist, and judging whether the target detection result of the suspected counterfeit commodity is correct or not if the suspected counterfeit commodity exists;

the server background inquires whether a pseudo-inferior product record exists in the position, and the method specifically comprises the following steps: storing the positions of the various counterfeit inferior products through a server background, and judging that the target detection is correct when the positions of the suspected counterfeit commodities corresponding to the cameras exist at the positions of the various counterfeit inferior products;

the server background stores the positions of the false inferior products in various places, namely the positions of the false inferior products verified after receiving the reported information or the positions of the false inferior products verified and judged by other manual work.

5. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: through the federal learning network platform, after label name alignment and data access are carried out based on a horizontal federal learning target detection algorithm, a global federal learning model is trained, and the method comprises the following steps:

the federated learning network platform includes: a client participant, a server and a federal learning framework;

the client side participants in the federal learning network platform are two or more than two: respectively a target commodity manufacturer, the first similar commodity manufacturer and other similar commodity manufacturers;

a server side in the federal learning network platform is a cloud server, and a federal learning framework is a micro-user's fate;

after a client in the federated learning network platform deploys a federated learning framework, the following steps are executed, including:

a. preprocessing a target detection data set stored locally at a client;

preprocessing a target detection data set stored locally at a client, wherein the preprocessing comprises the preprocessing of label name alignment;

the aligning of the label name comprises: unifying label names of the target commodity manufacturer and a first similar commodity manufacturer in a target detection data set stored in a client into a fused label commodity;

the aligning of the label name further comprises: keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in a target detection data set stored in a client, wherein the label names are a target commodity and a first target commodity respectively;

the preprocessing of the target detection data set stored locally at the client further comprises: data access preprocessing:

the data access preprocessing comprises the following steps:

when the label names aligned with the label names are not fusion label commodities, traversing the background similarity, the scene similarity and the light similarity of the target detection data set of the target commodities and the target detection data set of the first similar commodity manufacturer, and when the similarity is higher than a third threshold value, removing the data corresponding to the target detection data set of the first similar commodity manufacturer and not listing the data in a data set of federal learning training;

b. initiating a federal learning training task to other clients;

c. participating in a federal learning task, and training a federal learning model together with other client sides;

d. deploying a global federated learning model to predict and infer locally;

the server side in the federal learning network platform is provided by a cloud server, and executes the following steps of: monitoring the connection condition of a client side participant in the federal learning network platform in real time; aggregating the federal learning model uploaded by the client in the federal learning network platform; selecting a client in the federated learning network platform to participate in federated learning training; and uploading and broadcasting a global federal learning model to a client in the federal learning network platform.

6. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the server background is used for inquiring whether the position has a false inferior product record or not, and calculating the region weight of the false inferior product to obtain a region weight table of the false inferior product, and the method comprises the following steps:

inquiring the positions of the pseudo-inferior products of all regions through the server background, counting the number of the pseudo-inferior products appearing in each region, wherein the regions take regions, cities or streets as units, and combining a normalization algorithm according to the number of the pseudo-inferior products to obtain a region weight table where the pseudo-inferior products are located;

the region weight table of the false inferior products also comprises the region weight table of the false inferior products of each manufacturer.

7. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the acquiring of the full-network authorized camera, traversing the video stream read by the full-network authorized camera based on the global federal learning model, intercepting a picture according to a preset frame frequency, performing target identification, identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture, and determining whether a commodity manufacturer corresponding to a prediction result identifies authenticity, includes:

the identifying whether the target commodity or the first similar commodity or the fusion tag commodity exists in the picture further includes:

a1: identifying label text information in the picture by ocr technology, determining commodity names, judging whether the commodity names corresponding to the global federated learning model prediction result are consistent or not, and if not, judging that the prediction result is not trusted; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

a2: when the global federal learning model prediction result is the target commodity or the first similar commodity, combining the region weight table of the false inferior commodity, obtaining a prediction result value according to the confidence value of the global federal learning model prediction result and the weight value of a commodity manufacturer corresponding to the prediction result in the region weight table of the false inferior commodity, and judging that the prediction result is not reliable when the prediction result value is lower than a preset fourth threshold value; if the two are consistent, commodity manufacturers corresponding to the distribution prediction results carry out home-going authentication;

the method for identifying the authenticity of the distribution forecast commodity at home by a manufacturer corresponding to the distribution forecast commodity comprises the following steps:

b1: the manufacturer sends reminding information to a sales merchant according to the location of the camera corresponding to the picture, the sales merchant scans the two-dimensional code of the predicted commodity to determine whether the predicted commodity is a genuine commodity or not and feeds back the information, and if the predicted commodity is not a genuine commodity, the sales merchant checks the goods and obtains evidence at home.

8. The anti-counterfeiting traceability federal learning training method based on the blockchain as claimed in claim 1, wherein: the method for carrying out statistics on the accuracy of the global federated learning model based on the prediction result of the global federated learning model in the full-network authorized camera comprises the following steps of:

c1: the method for adjusting the data access preprocessing comprises the following steps: when the label name of the global federal learning model is not a fusion label commodity, traversing the background similarity, the scene similarity and the light similarity of the target commodity target detection data set and the first similar commodity manufacturer target detection data set, and when the similarity is higher than a third threshold value, removing the data corresponding to the first similar commodity manufacturer target detection data set and not listing the data in a federal learning training data set;

d1: the preprocessing method for adjusting the alignment of the name of the label comprises the following steps: judging the overlapping rate of the target commodity and the first similar commodity in each region based on the region weight table where the pseudo inferior product is located, and when the overlapping rate is higher than a sixth threshold value, keeping the label names of the target commodity manufacturer and the first similar commodity manufacturer in a commodity target detection data set stored in a client, wherein the label names are respectively the target commodity and the first target commodity;

and when the overlapping rate is lower than a sixth threshold value, unifying the label names of the target commodity manufacturer and the first similar commodity manufacturer in the commodity target detection data set stored in the client into a fused label commodity.

9. A computer apparatus comprising a memory for storing at least one program and a processor for loading the at least one program to perform the method of any one of claims 1-8.

Images