A store cash register management method and device, computer equipment and storage medium

Abstract

The application discloses a store cash register management method and device, computer equipment and a storage medium, and the method comprises the following steps: acquiring private cash register characteristics of each store, and constructing a potential private cash register risk identification model according to the private cash register characteristics; using the potential private cash register risk identification model to predict the potential private cash register probability of each store, so as to select n stores with the highest potential private cash register probability as first target stores; acquiring a private cash register tool image, and constructing a private cash register tool identification model according to the private cash register tool image; using the private cash register tool identification model to identify the private cash register tool of the first target stores; extracting a cash register area, identifying the staff and customer information in the cash register area, and combining the private cash register tool information to determine whether the corresponding store has a private cash register condition. The application can improve the store cash register management efficiency, and avoid or reduce the private cash register condition of the store.

Description

Technical Field

[0001] This invention relates to the field of computer software technology, and in particular to a store cashier management method, apparatus, computer equipment, and storage medium. Background Technology

[0002] Private cashiering refers to store clerks showing customers their personal QR codes or POS machines, or accepting payments without using the company's designated cash registers, thus pocketing the money that should belong to the store. Since these transactions don't actually go through the store's POS system, the system won't directly reflect the activity, and the amounts involved are usually small, making it difficult to identify stores engaging in private cashiering.

[0003] For businesses with a large number of stores, each store typically has multiple cameras installed, and each camera can provide over ten hours of video footage. Therefore, if we simply rely on image algorithms to detect every frame of every second, it would require enormous computing power, resulting in low detection efficiency. Furthermore, the detection process requires massive GPU computing resources. Considering network factors, it might be necessary to deploy GPU servers for each store, leading to excessively high overall detection costs and making it difficult to implement.

[0004] Therefore, how to improve the efficiency of store checkout management and effectively detect unauthorized checkout transactions in order to reduce or even avoid such transactions is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a store cashier management method, device, computer equipment, and storage medium, aiming to improve the efficiency of store cashier management and avoid or reduce the occurrence of unauthorized cashier transactions in stores.

[0006] In a first aspect, embodiments of the present invention provide a store cashier management method, including:

[0007] Obtain the characteristics of private cash collection in each store, and construct a risk identification model for potential private cash collection stores based on the characteristics of private cash collection; wherein, the characteristics of private cash collection include store operation characteristics, product category characteristics, time period sales characteristics, and private cash collection tool characteristics;

[0008] The potential illicit collection store risk identification model is used to predict the probability of potential illicit collection for each store, and the n stores with the highest probability of potential illicit collection are selected as the first target stores.

[0009] Acquire images of tools used for illicit bribery, and construct a tool recognition model based on the images of the tools used for illicit bribery;

[0010] The private collection tool identification model is used to identify the private collection tool in the first target store, and the first target store containing the private collection tool is output as the second target store, and the private collection tool information corresponding to the private collection tool is output.

[0011] Extract the cashier area from the second target store, identify the employee and customer information within the cashier area, and determine whether there is any private cash collection in the corresponding store by combining the information of the private collection tool.

[0012] Secondly, embodiments of the present invention provide a store POS management device, comprising:

[0013] The first model building unit is used to acquire the private cashier characteristics of each store and construct a risk identification model for potential private cashier stores based on the private cashier characteristics; wherein, the private cashier characteristics include store operation characteristics, product category characteristics, time period sales characteristics, and private cashier tool characteristics;

[0014] The first store selection unit is used to predict the potential private collection probability of each store using the potential private collection store risk identification model, so as to select the n stores with the highest potential private collection probability as the first target stores.

[0015] The second model building unit is used to acquire images of illicit collection tools and build an illicit collection tool recognition model based on the images of illicit collection tools.

[0016] The second store selection unit is used to identify the first target store using the private collection tool identification model, output the first target store containing the private collection tool as the second target store, and output the private collection tool information corresponding to the private collection tool.

[0017] The cashier management unit is used to extract the cashier area of ​​the second target store, identify the employee and customer information in the cashier area, and determine whether there is any private cashiering in the corresponding store by combining the private collection tool information.

[0018] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the store cashier management method as described in the first aspect.

[0019] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the store POS management method as described in the first aspect.

[0020] This invention provides a store checkout management method, apparatus, computer equipment, and storage medium. The method includes: acquiring the characteristics of unauthorized checkout at each store, and constructing a potential unauthorized checkout store risk identification model based on the characteristics; wherein the unauthorized checkout characteristics include store operation characteristics, product category characteristics, time period sales characteristics, and unauthorized checkout tool characteristics; using the potential unauthorized checkout store risk identification model to predict the potential unauthorized checkout probability for each store, and selecting the n stores with the highest potential unauthorized checkout probability as first target stores; acquiring images of unauthorized checkout tools, and constructing an unauthorized checkout tool identification model based on the images; using the unauthorized checkout tool identification model to identify unauthorized checkout tools in the first target stores, and outputting the first target stores containing unauthorized checkout tools as second target stores, and outputting the unauthorized checkout tool information corresponding to the unauthorized checkout tools; extracting the checkout area from the second target stores, identifying employee and customer information within the checkout area, and determining whether unauthorized checkout exists in the corresponding store based on the unauthorized checkout tool information. This invention proposes a risk identification model for stores with potential for unauthorized collection to predict the probability of such collection. Then, it uses unauthorized collection tools and checks the checkout area to determine if there are conditions for unauthorized collection. This improves the efficiency of store checkout management and avoids or reduces unauthorized collection. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a store POS management method provided in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of a sub-process of a store cashier management method provided in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of another sub-process of a store cashier management method provided in an embodiment of the present invention;

[0025] Figure 4 A schematic block diagram of a store POS management device provided in an embodiment of the present invention;

[0026] Figure 5 This is a schematic block diagram of a store POS management device provided in an embodiment of the present invention;

[0027] Figure 6This is another schematic block diagram of a store cashier management device provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0030] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0031] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0032] Please see below. Figure 1 This invention provides a store cashier management method, specifically including steps S101 to S105.

[0033] S101. Obtain the characteristics of private cash collection in each store, and construct a risk identification model for potential private cash collection stores based on the characteristics of private cash collection; wherein, the characteristics of private cash collection include store operation characteristics, product classification characteristics, time period sales characteristics, and private cash collection tool characteristics;

[0034] S102. Using the potential illicit collection store risk identification model, predict the potential illicit collection probability of each store, and select the n stores with the highest potential illicit collection probability as the first target stores.

[0035] S103. Obtain an image of the illicit collection tool and construct an illicit collection tool recognition model based on the image of the illicit collection tool;

[0036] S104. Use the private collection tool identification model to identify the private collection tool in the first target store, and output the first target store containing the private collection tool as the second target store, and output the private collection tool information corresponding to the private collection tool.

[0037] S105. Extract the cashier area of ​​the second target store, identify the employee and customer information in the cashier area, and determine whether there is any private cashiering in the corresponding store by combining the private cashiering tool information.

[0038] In this embodiment, a risk identification model for potential private cash collection stores is first constructed using the characteristics of private cash collection to predict stores with potential private cash collection, namely the first target store. Then, for the first target store, the constructed private cash collection tool identification model is used to identify and determine whether there are images of private cash collection tools. A second target store with a higher probability of private cash collection is further selected. Then, the second target store is identified and determined whether there are corresponding private cash collection factors, such as whether there are employees, customers and corresponding transaction factors in the cashier area at the same time, so as to determine whether the store has private cash collection.

[0039] This embodiment predicts stores with a potential for unauthorized cash collection by constructing a risk identification model. Then, it uses unauthorized cash collection tools and checks the checkout area for conditions that meet the criteria for such collection to determine if it is indeed happening. This improves the efficiency of store checkout management and avoids or reduces unauthorized cash collection. This embodiment can detect unauthorized cash collection in stores, thereby preventing the loss of corporate funds, and achieves the same effect while saving server network resources. Therefore, this embodiment is particularly suitable for enterprises with a large number of chain stores.

[0040] In one embodiment, step S101 includes:

[0041] The contribution value of each of the aforementioned private cashier features was calculated using a random forest model;

[0042] Select the m features with the highest contribution values ​​as input features;

[0043] The input features are fed into the XGBoost algorithm for classification and prediction, thereby constructing the potential private collection store risk identification model.

[0044] In this embodiment, the contribution value of each unauthorized cash collection feature is first determined using a random forest model. The feature with the highest contribution value is then selected as the input feature and fed into the XGBoost algorithm. The XGBoost algorithm classifies and predicts these features to identify stores with a high probability of unauthorized cash collection. In the specific implementation, positive and negative sample labels can be used for predictive learning of the XGBoost algorithm. Here, stores historically confirmed to have engaged in unauthorized cash collection can be used as positive sample label values, and stores without unauthorized cash collection can be used as negative sample label values. In addition, the specific features of the private payment system include store operation features, such as loss rate, time-based discounts, number of pending orders, number of returns, sales revenue, gross profit, customer traffic, and payment type; product category features (product category features), such as pork loss rate, pork discounts, non-pork discounts, and pork sales revenue; time-based sales features, such as the sales percentage of each product category in different time periods, time-based sales revenue, and month-on-month and year-on-year sales growth; and private payment tool features, such as the number of private payment tools, the type of private payment tools (Alipay, WeChat QR code payment, POS machine, etc.), the number of private payments, the weekly frequency of private payment tools, and the daily frequency of private payment tools.

[0045] In one specific embodiment, calculating the contribution value of each of the private cashier features using a random forest model includes:

[0046] The Gini value is used as the contribution value, and the Gini value for each of the private cashier characteristics is calculated according to the following formula:

[0047]

[0048] In the formula, D represents the private cashier feature, k represents the private cashier feature number, and P k Let k represent the feature weights, where k = 1, 2, 3, ...

[0049] In this embodiment, a random forest model is used to determine the contribution of each private collection feature to each decision tree in the random forest. The contribution value is calculated using the Gini value. Then, the average contribution value of each private collection feature to all decision trees in the random forest is taken to obtain the contribution value corresponding to each private collection feature. Finally, the contribution values ​​corresponding to each private collection feature are compared.

[0050] Furthermore, the change in the Gini index is calculated using the following formula:

[0051] N = Gini(D) + Gini(D) 1 )+…+Gini(D i )

[0052] Where N represents the change in the Gini index, and i represents the tree node number; the importance of the nth feature at node m in the decision tree is the change in the Gini index of the i new nodes after node m branches.

[0053] In one embodiment, step S103 includes:

[0054] The image of the smuggling tool is input into the YOLOv5 L6 detection model for training and learning, thereby constructing the smuggling tool recognition model.

[0055] The smuggling tool identification model is optimized and updated using the loss function according to the following formula:

[0056] Loss=γ1L cls +γ2L obj +γ3L loc

[0057] In the formula, Loss represents the loss function, γ1, γ2, and γ3 all represent balance coefficients, and L cls L represents the classification loss. obj L represents the confidence loss. loc This indicates positioning loss.

[0058] In this embodiment, the YOLOv5 L6 detection model is trained using images of payment tools to construct a payment tool recognition model capable of identifying the presence of such tools. It should be noted that, considering the relatively high resolution of the original image (1280*720) and the small size and lack of sufficient detail of the main target (payment tools), traditional detectors struggle to detect them; therefore, YOLOv5 L6 is used to build the model. It is understood that the images of payment tools refer to images containing payment tools, such as WeChat payment codes, Alipay payment codes, POS machines, and other payment collection tools. Furthermore, by annotating the payment tools in the images, the training effect can be improved.

[0059] Additionally, during model training, the optimal anchor values ​​can be calculated for the images of the smuggling tools. This involves regressing predicted boxes based on the initial anchor values, comparing them with the ground truth boxes, calculating the differences, and then iteratively updating the network parameters. The loss function for network training mainly consists of three parts: classification loss, confidence loss, and localization loss. The classification loss L... cls The main calculation focuses on the classification loss of positive samples, i.e., samples containing illicit tools, which determines whether the anchor box matches the corresponding labeled classification; the confidence loss L... objThe calculation involves the loss between the network-predicted target bounding box and the ground truth box. Different weights are assigned to the three target prediction layers (large, medium, and small). Taking a 1280x1280 pixel map as an example, the final output feature map sizes are 40x40, 80x80, and 160x160, respectively. The largest, 160x160, is responsible for detecting small targets, corresponding to the 1280x1280 pixel map, meaning each feature map has an 8x8 receptive field. Since this patent primarily focuses on small target detection, a larger weight is given to the small target prediction layer during training. The localization loss L... loc The calculation is for the localization loss of positive samples, which is the error between the predicted bounding box and the calibration box.

[0060] In one specific embodiment, such as Figure 2 As shown, the step of inputting the image of the smuggling tool into the YOLOv5 L6 detection model for training and learning, thereby constructing the smuggling tool recognition model, includes steps S201 to S204.

[0061] S201. Random data augmentation processing is performed on the image of the smuggling tool through the input of the YOLOv5 L6 detection model;

[0062] S202. Input the image of the smuggling tool after random data augmentation into the backbone layer to extract the image features of the smuggling tool image;

[0063] S203. The image features are enhanced using the neck layer of the YOLOv5 L6 detection model.

[0064] S204. The head layer of the YOLOv5 L6 detection model is used to predict the output of the image features after feature enhancement.

[0065] The YOLOv5 L6 detection model mainly consists of four parts: input, backbone, neck, and head. The input part primarily augments the input image data randomly, using methods such as random horizontal flipping, random scaling, random translation, or Mosaic data augmentation to enrich the sample data and enhance the model's robustness. Random scaling, in particular, adds many small targets, improving the network's robustness. Mosaic data augmentation involves stitching together four images using random scaling, cropping, and arrangement. This not only enriches the dataset and improves the network's robustness but also allows for direct computation of the data from the four images during training, reducing the mini-batch size required for training and enabling good results with a single GPU. The backbone is mainly used to extract features of the detection tools; the neck enhances the features extracted by the backbone; and the head is used for the output to predict the results.

[0066] In one embodiment, such as Figure 3 As shown, step S105 includes steps S301 to S305.

[0067] S301. Use the YOLOv5 L6 detection model to perform image recognition on the second target store to extract the cashier area.

[0068] S302. Obtain employee features and input the employee features into the YOLOv5 L6 detection model to construct an employee identification model;

[0069] S303. Use the YOLOv5 L6 detection model to perform person recognition in the cashier area, and use the employee recognition model to perform employee recognition in the cashier area.

[0070] S304. Based on the results of person recognition and employee recognition, determine whether there are employees and / or customers in the cashier area;

[0071] S305. When both employees and customers are present in the checkout area, the customer information corresponding to the customer is identified using the YOLOv5 L6 detection model; wherein, the customer information includes the customer's surrounding merchandise and the customer's payment method.

[0072] In this embodiment, for the second target store where a private payment tool is found, the judgment is further combined with the private payment tool information. The private payment tool information can be the time point before and after the appearance of the private payment tool. That is, the time point of the appearance of the private payment tool is further refined. An image is captured at the second level to identify whether there is a private payment tool, customers and corresponding customer information (such as whether the customer is holding a mobile phone or other payment terminal) in the transaction area, so as to identify whether there is a private payment situation.

[0073] Specifically, since customer transactions typically occur at the cash register, it's necessary to first identify the cashier area within the store. Further refinement of the cash register area is needed to identify instances of customers and employees making unauthorized transactions. To improve the efficiency of cashier area identification and extraction, a cashier area detection model can be built to specifically identify the cashier area. When building this model, images containing manually labeled cashier areas can be used to train the YOLOv5L6 detection model. Similarly, a loss function is used to update the parameters of the constructed cashier area detection model.

[0074] Furthermore, based on the extracted checkout area, a circle is formed with the cash register on the checkout counter as the center point and the relative size of the cash register as a threshold, enlarging a certain area proportionally to define the transaction area. This entire transaction area, centered on the cash register, can be divided into customer and employee areas to identify customers, cash registers, and employees. To further distinguish between employees and customers, the presence or absence of uniforms can be used for differentiation.

[0075] Typically, store employees wear uniforms and hats, so we can use these uniforms and hats to detect whether someone is an employee, thus building an employee identification model. Similarly, we can use images labeled with uniforms and hats to train the YOLOv5L6 detection model, and we can also use a loss function to update the parameters of the constructed employee identification model.

[0076] Furthermore, in one embodiment, step S105 further includes:

[0077] Obtain transaction information from the second target store; this transaction information may include transaction records, transaction time, etc.

[0078] The system combines the transaction information, the information on the payment tools used for unauthorized payments, the customer information, and the results of the personal identification in the checkout area to determine if unauthorized payments are made.

[0079] Stores typically have designated areas for product selection, transactions, and weighing. Customers usually complete their purchases in the transaction area. Therefore, by simulating the corresponding transaction scenarios and required elements, it's possible to determine if unauthorized cash registers are present. For example, a transaction might involve employees, customers, a POS system, and customers paying via mobile phone (i.e., customer information). Transaction scenarios can be either normal or unauthorized cash register transactions. The difference between unauthorized and normal transactions is that unauthorized cash registers bypass the POS system and instead use a personal QR code for payment. Therefore, the presence of unauthorized cash register tools, employees, customers, and customer mobile payments simultaneously in the transaction area, especially if the payment occurs outside of normal transaction hours, can be used as a criterion for identifying unauthorized cash registers.

[0080] For example, if a private cash register, employee, customer, and customer mobile payment scenario exist in the transaction area, but the POS system stores the transaction information for that moment, then it can be considered that no private cash register has occurred. Furthermore, customer purchase information can be obtained. For instance, if a customer purchased multiple types of goods, but the POS system only stores one type, and the corresponding transaction area at that moment contains a private cash register, employee, customer, and customer mobile payment, then it can be determined that a private cash register has occurred. Alternatively, if a private cash register, employee, and customer are present in the transaction area, but no customer mobile payment is detected, and the customer is identified as paying in cash, then it's possible to directly identify where the employee stored the received cash.

[0081] Figure 4 This is a schematic block diagram of a store checkout management device 400 provided in an embodiment of the present invention. The device 400 includes:

[0082] The first model construction unit 401 is used to obtain the private cashier characteristics of each store and construct a risk identification model for potential private cashier stores based on the private cashier characteristics; wherein, the private cashier characteristics include store operation characteristics, product classification characteristics, time period sales characteristics and private cashier tool characteristics;

[0083] The first store selection unit 402 is used to predict the potential private collection probability of each store using the potential private collection store risk identification model, so as to select the n stores with the highest potential private collection probability as the first target stores.

[0084] The second model building unit 403 is used to acquire images of illicit collection tools and build an illicit collection tool recognition model based on the images of illicit collection tools.

[0085] The second store selection unit 404 is used to identify the first target store using the private collection tool identification model, output the first target store containing the private collection tool as the second target store, and output the private collection tool information corresponding to the private collection tool.

[0086] The cashier management unit 405 is used to extract the cashier area of ​​the second target store, identify the employee and customer information in the cashier area, and determine whether there is any private cashiering in the corresponding store by combining the private collection tool information.

[0087] In one embodiment, the first model building unit 401 includes:

[0088] The contribution value calculation unit is used to calculate the contribution value of each of the aforementioned private cashier features using a random forest model;

[0089] The feature selection unit is used to select the m features with the highest contribution values ​​as input features;

[0090] The classification and prediction unit is used to input the input features into the xgboost algorithm for classification and prediction, thereby constructing the potential private collection store risk identification model.

[0091] In one embodiment, the contribution value calculation unit includes:

[0092] A Gini value calculation unit is used to use the Gini value as the contribution value and calculate the Gini value for each of the private cashier characteristics according to the following formula:

[0093]

[0094] In the formula, D represents the private cashier feature, k represents the private cashier feature number, and P k Let k represent the feature weights, where k = 1, 2, 3, ...

[0095] In one embodiment, the second model building unit 403 includes:

[0096] The training and learning unit is used to input the image of the smuggling tool into the YOLOv5 L6 detection model for training and learning, thereby constructing the smuggling tool recognition model.

[0097] The optimization and update unit is used to optimize and update the smuggling tool identification model according to the following formula using a loss function:

[0098] Loss=γ1L cls +γ2L obj +γ3L loc

[0099] In the formula, Loss represents the loss function, γ1, γ2, and γ3 all represent balance coefficients, and L cls L represents the classification loss. obj L represents the confidence loss. loc This indicates positioning loss.

[0100] In one embodiment, such as Figure 5 As shown, the training and learning unit includes:

[0101] The data augmentation unit 501 is used to perform random data augmentation processing on the image of the smuggling tool through the input of the YOLOv5 L6 detection model.

[0102] The feature extraction unit 502 is used to input the image of the smuggling tool after random data augmentation into the backbone layer to extract the image features of the smuggling tool image;

[0103] Feature enhancement unit 503 is used to perform feature enhancement processing on the image features using the neck layer of the YOLOv5 L6 detection model;

[0104] The output prediction unit 504 is used to predict the output of the image features after feature enhancement processing through the head layer of the YOLOv5 L6 detection model.

[0105] In one embodiment, such as Figure 6 As shown, the cashier management unit 405 includes:

[0106] Image recognition unit 601 is used to perform image recognition on the second target store using the YOLOv5 L6 detection model in order to extract the cashier area therein.

[0107] The employee model building unit 602 is used to acquire employee features and input the employee features into the YOLOv5 L6 detection model to build an employee identification model.

[0108] The person recognition unit 603 is used to perform person recognition in the cashier area using the YOLOv5 L6 detection model, and to perform employee recognition in the cashier area using the employee recognition model.

[0109] The person identification unit 604 is used to determine whether there are employees and / or customers in the cashier area based on the results of person identification and employee identification.

[0110] The information recognition unit 605 is used to identify customer information corresponding to the customer when both employees and customers are present in the checkout area; wherein, the customer information includes the customer's surrounding goods and the customer's payment method.

[0111] In one embodiment, the cashier management unit 405 further includes:

[0112] A transaction information acquisition unit is used to acquire transaction information of the second target store;

[0113] The private cashier detection unit is used to make a private cashier detection by combining the transaction information, private cashier tool information, customer information, and the person recognition results of the cashier area.

[0114] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0115] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0116] This invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the computer device may also include various network interfaces, power supplies, and other components.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0118] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A store checkout management method, characterized in that, include: Obtain the characteristics of private cash collection in each store, and construct a risk identification model for potential private cash collection stores based on the characteristics of private cash collection; wherein, the characteristics of private cash collection include store operation characteristics, product category characteristics, time period sales characteristics, and private cash collection tool characteristics; The potential illicit collection store risk identification model is used to predict the probability of potential illicit collection for each store, and the n stores with the highest probability of potential illicit collection are selected as the first target stores. Acquire images of tools used for illicit bribery, and construct a tool recognition model based on the images of the tools used for illicit bribery; The private collection tool identification model is used to identify private collection tools in the first target store, and the first target store containing private collection tools is output as the second target store, and the private collection tool information corresponding to the private collection tool is also output; the private collection tool information refers to the time points before and after the appearance of the private collection tool. Extract the cashier area of ​​the second target store, identify the employee and customer information within the cashier area, and determine whether there is any private cash collection in the corresponding store by combining the information of the private collection tool. The step of extracting the checkout area of ​​the second target store, identifying the employee and customer information within the checkout area, and determining whether there is any unauthorized checkout activity in the corresponding store by combining the information on the unauthorized checkout tool includes: The YOLOv5 L6 detection model was used to perform image recognition on the second target store in order to extract the cashier area. Employee features are obtained and input into the YOLOv5 L6 detection model to construct an employee identification model. The YOLOv5 L6 detection model is used to perform person recognition in the checkout area, and the employee recognition model is used to perform employee recognition in the checkout area. Based on the results of person recognition and employee recognition, determine whether there are employees and / or customers in the checkout area; When both employees and customers are present in the checkout area, the YOLOv5 L6 detection model is used to identify the customer information corresponding to the customer; wherein, the customer information includes the customer's surrounding merchandise and the customer's payment method; Obtain transaction information from the second target store; The system combines the transaction information, the information on the payment tools used for unauthorized payments, the customer information, and the results of the personal identification in the checkout area to determine if unauthorized payments are made.

2. The store checkout management method according to claim 1, characterized in that, The process of acquiring the characteristics of unauthorized cash collection at each store and constructing a risk identification model for potential unauthorized cash collection stores based on these characteristics includes: The contribution value of each of the aforementioned private cashier features was calculated using a random forest model; Select the m features with the highest contribution values ​​as input features; The input features are fed into the XGBoost algorithm for classification and prediction, thereby constructing the potential private collection store risk identification model.

3. The store checkout management method according to claim 2, characterized in that, The calculation of the contribution value of each of the private cashier features using a random forest model includes: The Gini value is used as the contribution value, and the Gini value for each of the private cashier characteristics is calculated according to the following formula: In the formula, D represents the private cashier feature, k represents the private cashier feature number, and P k Let k represent the feature weights, where k = 1, 2, 3, ...

4. The store checkout management method according to claim 1, characterized in that, The step of acquiring images of illicit bribery tools and constructing a illicit bribery tool recognition model based on the images includes: The image of the smuggling tool is input into the YOLOv5 L6 detection model for training and learning, thereby constructing the smuggling tool recognition model. The smuggling tool identification model is optimized and updated using the loss function according to the following formula: Loss=γ1L cls +γ2L obj +γ3L loc In the formula, Loss represents the loss function, γ1, γ2, and γ3 all represent balance coefficients, and L cls L represents the classification loss. obj L represents the confidence loss. loc This indicates positioning loss.

5. A store checkout management method according to claim 4, characterized in that, The step of inputting the image of the smuggling tool into a YOLOv5 L6 detection model for training and learning, thereby constructing the smuggling tool recognition model, includes: The image of the smuggling tool is subjected to random data augmentation processing through the input of the YOLOv5 L6 detection model. The image of the smuggling tool, after random data augmentation, is input into the backbone layer to extract the image features of the smuggling tool image; The image features are enhanced using the neck layer of the YOLOv5 L6 detection model. The head layer of the YOLOv5 L6 detection model outputs predictions based on the image features after feature enhancement.

6. A store POS management device, characterized in that, include: The first model building unit is used to acquire the private cashier characteristics of each store and construct a risk identification model for potential private cashier stores based on the private cashier characteristics; wherein, the private cashier characteristics include store operation characteristics, product category characteristics, time period sales characteristics, and private cashier tool characteristics; The first store selection unit is used to predict the potential private collection probability of each store using the potential private collection store risk identification model, so as to select the n stores with the highest potential private collection probability as the first target stores. The second model building unit is used to acquire images of illicit collection tools and build an illicit collection tool recognition model based on the images of illicit collection tools. The second store selection unit is used to identify the private collection tool in the first target store using the private collection tool identification model, and output the first target store containing the private collection tool as the second target store, and output the private collection tool information corresponding to the private collection tool; the private collection tool information refers to the time points before and after the appearance of the private collection tool. The cashier management unit is used to extract the cashier area of ​​the second target store, identify the employee and customer information in the cashier area, and determine whether there is any private cashiering in the corresponding store by combining the private collection tool information. The cashier management unit includes: The image recognition unit is used to perform image recognition on the second target store using the YOLOv5 L6 detection model in order to extract the cashier area therein. An employee model building unit is used to acquire employee features and input the employee features into a YOLOv5 L6 detection model to build an employee identification model. The person recognition unit is used to perform person recognition in the cashier area using the YOLOv5 L6 detection model, and to perform employee recognition in the cashier area using the employee recognition model. The person identification unit is used to determine whether there are employees and / or customers in the checkout area based on the results of person identification and employee identification. An information recognition unit is used to identify customer information corresponding to a customer when both employees and customers are present in the checkout area; wherein, the customer information includes the customer's surrounding merchandise and the customer's payment method; The cashier management unit also includes: A transaction information acquisition unit is used to acquire transaction information of the second target store; The private cashier detection unit is used to make a private cashier detection by combining the transaction information, private cashier tool information, customer information, and the person recognition results of the cashier area.

7. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the store checkout management method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the store checkout management method as described in any one of claims 1 to 5.

Images