Distributed cashier method and system
Through federated transfer learning and weighted average aggregation algorithms, the intelligent collaboration problem caused by data differences and model inconsistencies in the distributed cash register system was solved, and efficient and intelligent collaborative optimization of cash register terminals was achieved, thereby improving the overall service quality and operational efficiency.
Patent Information
- Application Number
- CN202510899965.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-01
- Publication Date
- 2025-09-26
AI Technical Summary
The data differences and inconsistent model performance of each cash register terminal in the distributed cash register system lead to uneven levels of intelligent collaboration, affecting the overall service quality and operational efficiency.
The knowledge transfer model in federated transfer learning is used to analyze the common features and transferable knowledge among cash register data, and the model parameters are aggregated through the weighted average aggregation algorithm, and optimized and adjusted in combination with the intelligent function performance indicators.
It has improved the intelligent collaboration level of the distributed cash register system, achieved effective integration and collaborative optimization of data from various terminals, and improved overall service quality and operational efficiency.
Smart Images
Figure CN120706716A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of cashier technology, and in particular to a distributed cashier method and a distributed cashier system. Background Art
[0002] In traditional retail scenarios, checkout work is often concentrated at fixed checkout counters. During peak shopping periods, customers are often forced to wait in long queues to check out, resulting in a poor customer experience and affecting store operational efficiency. With the continuous development of intelligent technology and the diversification of retail formats, the demand for more flexible, efficient, and intelligent checkout methods is becoming increasingly prominent. Some existing decentralized checkout attempts either lack effective intelligent coordination mechanisms, prone to data confusion and settlement errors, or are insufficiently intelligent and cannot fully utilize various intelligent devices and data analysis capabilities to optimize the checkout process.
[0003] The present invention provides a distributed cashier method, which is used to solve the problem that in a distributed cashier system, the level of intelligent collaboration is uneven due to evaluation indicators such as data differences among cashier terminals and inconsistent model performance, which affects the overall service quality and operational efficiency. Summary of the Invention
[0004] The present invention provides a distributed cashier method and system to solve the problem in the prior art that the intelligent collaboration level is uneven due to evaluation indicators such as data differences among cashier terminals and inconsistent model performance, which affects the overall service quality and operational efficiency.
[0005] On the one hand, the present invention provides a distributed cashier method, including: S1: collecting cashier data of each cashier terminal.
[0006] S2: Preprocess the cash register data to obtain preprocessed cash register data.
[0007] S3: Use the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between the pre-processed cash register data to obtain a trained cash register data model.
[0008] S4: Based on the preprocessed cashier data and the weight of the preprocessed cashier data, a weighted average aggregation algorithm is used to aggregate the parameters in the trained cashier data model to obtain the global knowledge transfer model parameters.
[0009] S5: Use the preset intelligent function performance indicators to measure the level of collaboration of distributed cashiers, and optimize and adjust the cashier terminals that are below the preset standards to obtain optimized distributed cashiers.
[0010] According to a distributed checkout method provided by the present invention, in step S1, checkout data includes: product-related data including product image data, product barcode data, and basic product attribute data; transaction-related data including transaction time data, transaction amount data, and payment method data; customer membership data, return and exchange data, and cashier operation data.
[0011] According to a distributed cashier method provided by the present invention, in step S2, the pre-processing step includes: Data cleaning: Clean the cash register data to obtain cleaned cash register data.
[0012] Data labeling: Label the commodity-related data in the cleaned cash register data to obtain labeled cash register data.
[0013] Feature extraction: Convert the labeled cash register data into feature vectors.
[0014] According to the distributed checkout method provided by the present invention, in step S3, common characteristics include daily peak sales hours common to all checkout terminals, similar proportions of commonly used payment methods, the relative stability of popular product categories, and common patterns in customer purchase combinations. Transferable knowledge includes methods for busy terminals to respond to complex customer needs, promotional techniques for specific scenarios, and product identification tips, accumulated from some checkout terminals.
[0015] According to a distributed cashier method provided by the present invention, in step S3, the specific steps of using the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between pre-processed cashier data are as follows: S31: A shared layer is set after the input layer of each preprocessed cashier data model, and the common feature matrix of the shared layer is extracted through multi-data joint training and gradient optimization.
[0016] S32: Calculate the feature difference and correlation between the pre-processed cashier data based on the shared layer common feature matrix to obtain a difference correlation matrix.
[0017] S33: Design a migration mapping function based on the difference association matrix to map the feature space of the preprocessed cashier data to the target data space, compare the model performance under different mapping parameters, and select the optimal parameter combination.
[0018] S34: Integrate the migration mapping function into the federated learning framework to monitor the performance indicators of each terminal model in real time. When performance fluctuations are detected, the mapping function parameters are dynamically adjusted using the shared layer common feature matrix.
[0019] According to a distributed cashier method provided by the present invention, in step S4, the specific steps of aggregating the parameters in the trained cashier data model using a weighted average aggregation algorithm are as follows: S41: Collect cash register data model parameters of all distributed cash register terminals.
[0020] S42: Calculate the weight of the pre-processed cashier data of each terminal.
[0021] S43: For each cashier data model parameter, calculate the weighted average of the parameters according to the weight of the terminal pre-processed cashier data to obtain the aggregation parameter θ global .
[0022] S44: Normalize the aggregated parameters to obtain normalized aggregated parameters θ global .
[0023] S45: normalize the aggregation parameter θ global Update the parameters of the cash register data model.
[0024] S46: Dynamically adjust the weight distribution strategy of terminal data based on the performance feedback of the cash register data model.
[0025] According to a distributed cashier method provided by the present invention, in step S42, the calculation formula for the weight of the pre-processed cashier data is: ; Where, ω i is the final weight of the i-th pre-processed cashier data, n is the number of pre-processed cashier data, F k is the global weight of the kth quality assessment indicator, S i,k is the score of the i-th preprocessed cashier data on the k-th quality evaluation index, It is the sum of the total scores of all pre-processed cash register data.
[0026] According to a distributed checkout method provided by the present invention, in step S5, the preset intelligent function performance indicators include: transaction processing efficiency index, transaction accuracy index, customer experience index, and system collaboration index.
[0027] Transaction processing efficiency indicators include average transaction time, transaction throughput, and peak-hour processing capacity.
[0028] Transaction accuracy indicators include transaction error rate, product identification accuracy, and amount matching rate.
[0029] Customer experience indicators include customer waiting time and customer satisfaction.
[0030] System collaboration indicators include model parameter consistency, knowledge transfer effectiveness, and load balance.
[0031] According to a distributed cashier method provided by the present invention, the calculation formula for the transaction error rate is: ; Where, E i is the number of occurrences of type i error, m is the total number of error types, and n is the total number of transactions.
[0032] On the other hand, the present invention further provides a distributed cash register system, comprising: a data collection module for collecting cash register data of each cash register terminal.
[0033] The data preprocessing module preprocesses the cash register data to obtain preprocessed cash register data.
[0034] The data analysis module is used to use the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between pre-processed cash register data to obtain a trained cash register data model.
[0035] The data aggregation module is used to aggregate the parameters in the trained cashier data model using a weighted average aggregation algorithm based on the preprocessed cashier data and the weight of the preprocessed cashier data to obtain the global knowledge transfer model parameters.
[0036] The data evaluation module is used to measure the level of collaboration of distributed cash registers using preset intelligent function performance indicators, and to optimize and adjust cash register terminals that are below the preset standards to obtain optimized distributed cash registers.
[0037] The distributed cashier method and system provided by the present invention uses a federated transfer learning knowledge transfer model to mine common and transferable knowledge, and a weighted average aggregation algorithm to obtain global model parameters, and then measures and optimizes the parameters based on intelligent function performance indicators. This solves the problem of uneven intelligent collaboration levels in distributed cashier systems due to evaluation indicators such as data differences among cashier terminals and inconsistent model performance, which affects overall service quality and operational efficiency.
[0038] The weighted average aggregation algorithm combines pre-processed cash register data from different terminals according to their respective weights, ensuring that each terminal's data plays an appropriate role in the overall system. The knowledge transfer model, on the other hand, focuses on discovering the common characteristics and transferable knowledge among these data. The combination of the two can comprehensively and deeply integrate the value inherent in multi-source data, avoiding the limitations of a single data source or single analytical method. This allows dispersed data resources across various entities to collaborate and play a greater role, collectively contributing to the improvement of the intelligence level of the entire distributed cash register system.
[0039] The transferable knowledge identified through the knowledge transfer model allows different cash register terminals, inventory management systems, and marketing management systems to better understand each other's data characteristics and business connections, enabling the sharing of experience and knowledge. Processing the integrated data using a weighted average aggregation algorithm further strengthens collaboration among various entities at the data level, making collaboration between different parts of the system closer and smoother, and overall moving towards a more optimized operational state, such as achieving seamless coordination between cash register, inventory management, and marketing activities.
[0040] The common features analyzed by the knowledge transfer model help extract universal feature representations. These features reflect common patterns and regularities across data from different entities. When the data is aggregated using a weighted average aggregation algorithm and used for subsequent model training, the model can learn from this data incorporating common features, reducing over-reliance on specific terminal data. This enhances the model's generalization across diverse business scenarios and data distributions, enabling it to make accurate and effective predictions and decisions even with new data inputs.
[0041] The weighted average aggregation algorithm assigns weights based on evaluation metrics such as the quality and quantity of each subject's data. This allows high-quality, large-scale data to have a greater impact during the aggregation process, guiding the model to better learn key information. The transferable knowledge mined by the knowledge transfer model provides additional learning guidance for the model, helping it converge more quickly to optimal parameter states and avoid falling into local optimal solutions. This improves model training efficiency and ultimate performance, for example, by increasing the accuracy of product recognition at checkout terminals and the precision of predicting customer preferences.
[0042] The weighted average aggregation algorithm considers the weight differences between entities when fusing data, ensuring that common features are reflected in the overall data while also not neglecting the individual characteristics of different entity data. The knowledge transfer model, while analyzing common features and transferable knowledge, also retains the data characteristics and knowledge unique to each entity based on its own business scenario, while exploring commonalities. The combination of these two approaches enables a distributed checkout system to leverage commonalities for collaborative optimization while respecting the individual differences between terminals and business systems, meeting diverse business needs. Different checkout terminals may operate in different environments and target different customer groups. This combined approach allows the system to flexibly adapt to specific circumstances. For example, checkout terminals in large shopping malls and small street shops can achieve unified basic functional optimization through common features and aggregated data. Furthermore, knowledge transfer and appropriate weighting can be used to adapt to their different sales characteristics and customer preferences, providing personalized intelligent services and improving the overall system's adaptability to complex business environments.
[0043] As business evolves, data from various entities will constantly change, and new features and knowledge will gradually emerge. This combination of the two allows for dynamic adjustment of weighted average aggregation and knowledge transfer content based on new data, continuously uncovering new common features and transferable knowledge, and continuously optimizing data fusion and model training. This ensures that the distributed cash register system can maintain efficient and intelligent operation over the long term and adapt to ever-changing business operational needs. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0045] Figure 1 This is a flow chart of a distributed cashier method provided by an embodiment of the present invention; Figure 2 This is a module diagram of a distributed cash register system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0046] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.
[0047] The following combination Figure 1-Figure 2 A distributed cashier method and system of the present invention is described.
[0048] Figure 1 It is a flow chart of a distributed cashier method provided by an embodiment of the present invention.
[0049] like Figure 1 As shown, an embodiment of the present invention provides a distributed cashier method and system, the method comprising: S1: Collect data from various cash register terminals and related business systems, specifically targeting the business data they generate. For example, cash register terminals collect data such as product images and transaction records. Marketing management systems collect feedback on promotional campaign effectiveness and customer engagement data.
[0050] Product data includes product image data, product barcode data, and product basic attribute data.
[0051] Product image data: In a checkout system with product image recognition capabilities, the cash register terminal uses an installed camera or other equipment to capture images of products as customers place them at the checkout counter for payment. These images can be used for subsequent product identification and category verification, such as distinguishing between products that appear similar but differ in brand or specification, to ensure accurate checkout. Image data includes information about the product's appearance from different angles and positions. For example, for bottled beverages, images from multiple angles, including upright and sideways positions, can be provided, helping to accurately identify specific product details such as the model and packaging version.
[0052] Product barcode data: Cash register terminals are equipped with scanners and other devices for scanning barcodes or QR codes on products. Barcode data uniquely identifies the product. Scanning it quickly retrieves detailed information such as the product name, price, and inventory from the system database, enabling rapid entry and pricing. For gift box combinations or shampoo in different sizes, barcodes can accurately distinguish the specific product attributes, ensuring settlement based on the correct price for each item or combination.
[0053] Basic product attributes: Cash registers record basic product attributes, including category, brand, specifications, and color. This information facilitates categorized statistics and analysis, such as sales percentages of different product categories and the popularity of various brands. This provides data support for merchants' product management and marketing strategy development.
[0054] Transaction-related data includes transaction time data, transaction amount data, and payment method data.
[0055] Transaction time data: This records the specific time of each transaction, including year, month, day, hour, minute, and second. This helps analyze sales patterns across different time periods, such as which time of day has the highest customer traffic and sales, morning, noon, or evening? It also helps identify the best sales days of the week, helping businesses plan staffing schedules, restocking, and other operational activities.
[0056] Transaction amount data: records the total amount of each transaction, and subdivides the total original price of the goods, discount amount, actual payment amount and other specific amount information to facilitate subsequent financial accounting, sales performance statistics, and analysis of the impact of promotional activities on sales.
[0057] Payment method data: Clearly record the payment methods used by customers, such as cash, bank card, QR code scanning, membership card, and facial recognition. Analyzing the proportion of payment methods used can help understand customer payment preferences and help merchants evaluate different payment channels based on factors such as transaction fees and security, further optimizing payment services and partner payment institutions.
[0058] Transaction serial number data: Each transaction generates a unique serial number, which serves as an identifier within the system, similar to a transaction ID number. This number can be used to quickly and accurately locate specific transaction records during subsequent inquiries, reconciliations, returns, and exchanges, facilitating various operations and data association.
[0059] Customer data: If a customer is a member, the cash register will capture their membership level, including regular, silver, and gold. Different levels correspond to different discounts, points, and other benefits. The customer's basic personal information related to consumption, such as name and contact information, is recorded to facilitate subsequent member service notifications and marketing campaign push. The customer's past purchase categories, purchase frequency, and average purchase amount are also recorded to analyze member consumption preferences and loyalty.
[0060] Return and Exchange Data: When a product is returned or exchanged, the cash register records the product information (similar to the original product barcode, name, and specifications), the reason for the return or exchange (quality issue, customer dissatisfaction, unsuitable size, etc.), the time of return or exchange, and the transaction serial number involved. This data is crucial for analyzing product quality and customer satisfaction, as well as optimizing product management and after-sales service strategies. For example, if a certain product category is frequently returned or exchanged due to quality issues, merchants need to promptly communicate with the supplier to resolve the issue.
[0061] Cashier operation data: This records cashier operations at the checkout terminal, including the order in which items are scanned, input commands, and operation timestamps. This data can be used to monitor cashier compliance and prevent operational errors or violations. It also helps analyze efficiency bottlenecks in the checkout process, enabling optimization and overall efficiency improvement.
[0062] S2: Detailed labeling of collected data. Product images are clearly labeled with key attributes such as product category, brand, and specifications. Transaction records are labeled with customer purchase combinations, spending amount, purchase time, and membership status.
[0063] Product image annotation: For product images captured at checkout terminals, a structured labeling system must be constructed through multi-dimensional annotation. Pre-process the checkout data by selecting the product using a rectangular or polygonal box, and labeling the product category, brand name, and specifications. For combined products, the attributes of each sub-product must be separately annotated.
[0064] Add color descriptions, texture features, and key identifiers. These annotations can be achieved through pixel-level labeling using image segmentation technology.
[0065] Record the shooting environment, placement, and image quality for subsequent data enhancement. The annotation tool can use the open-source VGG Image Annotator pre-trained model to automatically generate candidate boxes for manual correction.
[0066] Using feature extraction methods, the labeled data is converted into feature vectors that can be used for model training. Product image data is processed through the convolutional and pooling layers of a convolutional neural network to extract visual features.
[0067] Structured data such as transaction records are used to construct feature vectors through engineering methods such as data encoding, normalization, and feature combination to ensure that the data can be effectively input into the model for training.
[0068] S3: Use the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between the pre-processed cash register data to obtain a trained cash register data model.
[0069] S31: Use the knowledge transfer method in federated transfer learning to analyze the common features and transferable knowledge between different pre-processed cash register data. By setting up a shared layer in the model and building a specific migration mapping mechanism, the transfer and sharing of knowledge between different pre-processed cash register data can be achieved, reducing the differences in intelligent functions caused by the distribution differences of each pre-processed cash register data. The effective knowledge of extracting popular product features trained by a busy cash register terminal can be transferred to the models of other terminals or related business systems, allowing the various pre-processed cash register data models to learn from each other and promote the coordinated improvement of overall intelligent functions.
[0070] S32: Based on the common characteristics of each pre-processed cash register data reflected in the output of the shared layer, deeply analyze the differences and correlations between each pre-processed cash register data, and accordingly design a mapping function or module that can realize the migration of knowledge from one pre-processed cash register data to the corresponding model part of another pre-processed cash register data.
[0071] When constructing a shared layer, it is trained with multiple preprocessed cash register data sets, allowing it to learn relatively abstract feature patterns shared by different preprocessed cash register data sets. For example, for product image data, regardless of which preprocessed cash register data set it comes from, the shared layer may learn some common basic visual features such as product appearance outline and texture. For transaction record-related data, common statistical features such as purchase frequency and consumption amount range may be learned. The process of training the shared layer typically uses a gradient-based optimization algorithm. In each iteration, a small batch of samples from each preprocessed cash register data set is combined to calculate the gradient and update the parameters of the shared layer, so that the output of the shared layer can best reflect the common parts of each preprocessed cash register data set.
[0072] S33: Apply the constructed migration mapping mechanism to model training, compare the performance of each pre-processed cashier data model under different parameter settings, fine-tune the parameters and structure of the migration mapping mechanism based on the comparison results, and output the optimized migration mapping mechanism.
[0073] Construct a specific migration mapping mechanism. This requires an in-depth analysis of the differences and connections between each pre-processed cash register data, and the design of mapping functions or mapping modules based on the business logic and the potential connections between data features. For example, if it is found that the customer's purchase combination in the cash register terminal data is related to the product combination recommended by the promotion in the marketing management system, a mapping mechanism can be constructed to convert the customer's purchase combination characteristics extracted from the cash register terminal into a representation that matches the product combination recommendation characteristics in the marketing management system through a specific mathematical transformation (which can be linear transformation, nonlinear neural network mapping, etc.), thereby realizing the migration of knowledge from one pre-processed cash register data to the model part corresponding to another pre-processed cash register data.
[0074] S34: Use the optimized migration mapping mechanism for federated transfer learning, continuously monitor the performance of each pre-processed cash register data model after integrating the common features of the shared layer with the migrated knowledge, retrospectively analyze and adjust the model architecture and parameters for any problems that arise, and continuously optimize the knowledge migration effect to enhance the overall intelligent collaboration function.
[0075] After determining the transfer mapping mechanism, its parameters and structure are continuously adjusted and optimized through experimentation and verification. For example, during training, the performance of the model on various preprocessed cashier data is compared with and without the transfer mapping mechanism and with different parameter settings (for example, the changes in evaluation metrics such as accuracy and recall for the prediction tasks corresponding to each preprocessed cashier data). Based on performance feedback, the weights in the mapping function and the method of nonlinear transformation are fine-tuned to ensure that the transfer mapping mechanism can effectively and accurately transfer transferable knowledge to the model of the target preprocessed cashier data, while avoiding overfitting or the transfer of erroneous knowledge that may lead to degraded model performance.
[0076] Throughout the federated transfer learning process, the performance of each pre-processed cash register data model is continuously monitored after incorporating the common features learned in the shared layer and the knowledge transferred by the transfer mapping mechanism. Observe whether each pre-processed cash register data has improved in its own business-related intelligent tasks, such as the accuracy of product recognition at the cash register terminal and the effectiveness of promotional activity recommendations in the marketing management system. If problems such as poor performance or overfitting occur, timely backtracking and analysis are conducted to determine which link in the shared layer or transfer mapping mechanism has deviated. Targeted adjustments are then made to the model architecture and parameters to continuously optimize the effectiveness of knowledge transfer. This allows the pre-processed cash register data models of each terminal to fully utilize the common features and transferable knowledge between different pre-processed cash register data, achieving a coordinated improvement in overall intelligent functions.
[0077] S4: Based on the preprocessed cashier data and the weight of the preprocessed cashier data, a weighted average aggregation algorithm is used to aggregate the parameters in the trained cashier data model to obtain the global knowledge transfer model parameters.
[0078] The specific steps for using the weighted average aggregation algorithm to aggregate the parameters in the trained cash register data model are as follows: S41: Collect the cashier data model parameters of all distributed cashier terminals, denoted as {θ1, θ2, ..., θ k}, θ k is the model parameter set of the kth terminal.
[0079] S42: Calculate the weight of the pre-processed cashier data of each terminal, and assign a weight {ω1, ω2, ..., ω n}.
[0080] The calculation formula for the weight of preprocessed cashier data is: ; Where, ω i is the final weight of the i-th pre-processed cashier data, n is the number of pre-processed cashier data, F k is the global weight of the kth quality assessment indicator, S i,k is the score of the i-th preprocessed cashier data on the k-th quality evaluation index, is the sum of the total scores of all pre-processed cashier data. The quality assessment indicators k include completeness, accuracy, timeliness, relevance, and consistency.
[0081] S43: Calculate the weighted average of parameters: For each cashier data model parameter, perform weighted average according to the corresponding terminal weight. k , its global aggregate value θ global The calculation formula is: ; Where θ i，k is the kth model parameter of the i-th terminal.
[0082] For category-type parameters, a weighted voting mechanism is used to count the frequency of each category in each terminal, multiply it by the corresponding terminal weight, and sum it up. The category with the highest weighted frequency is selected as the global parameter.
[0083] S44: Normalize the aggregated parameters to ensure that the parameter range is consistent with the original model. If the parameters deviate too much, apply L1 / L2 regularization to constrain the parameter range to prevent overfitting.
[0084] S45: The aggregated parameters θglobal are updated as parameters of the global knowledge transfer model, and each terminal downloads the updated parameters for further training.
[0085] S46: Dynamically adjust the weight distribution strategy of each terminal data based on model performance feedback.
[0086] S5: Use the preset intelligent function performance indicators to measure the level of collaboration of distributed cashiers, and optimize and adjust the cashier terminals that are below the preset standards to obtain optimized distributed cashiers.
[0087] Performance indicators include: transaction processing efficiency indicators, transaction accuracy indicators, customer experience indicators, and system collaboration indicators.
[0088] Transaction processing efficiency indicators include average transaction time, transaction throughput, and peak-hour processing capacity.
[0089] Transaction time measures the average time from the start to the completion of a single transaction, reflecting the checkout speed. The formula for calculating average transaction time is: ; Where Ti is the time taken for the i-th transaction, and n is the total number of transactions.
[0090] Transaction throughput is the number of transactions processed per unit time, reflecting the system's concurrency capability. The calculation formula for transaction throughput is: ; Where n is the total number of transactions and t is the length of the time period.
[0091] Peak processing capacity refers to the transaction processing efficiency during business peak periods.
[0092] Transaction accuracy indicators include transaction error rate, product identification accuracy, and amount matching rate.
[0093] The transaction error rate is the proportion of errors that occur during the transaction process, including product identification errors, amount calculation errors, etc. The calculation formula for the transaction error rate is: ; Where, E i is the number of occurrences of type i error, m is the total number of error types, and n is the total number of transactions.
[0094] Product recognition accuracy is the accuracy of identifying products through images or barcodes. The formula for calculating product recognition accuracy is: ; Where a is the number of correctly identified products, and d is the total number of identified products.
[0095] Amount matching rate: The degree of matching between the transaction amount and the order amount.
[0096] ; Where q i is the actual amount, f i is the order amount, and n is the total number of transactions.
[0097] Customer experience indicators include customer waiting time and customer satisfaction.
[0098] Customer waiting time is the average waiting time from when customers queue up to when they start their transactions.
[0099] Customer satisfaction is a customer satisfaction score collected through questionnaires or rating systems, usually an average of 1-5 points.
[0100] System collaboration indicators include model parameter consistency, knowledge transfer effectiveness, and load balance.
[0101] Model parameter consistency is the similarity between each terminal model parameter and the global model, which measures the effectiveness of collaborative training. The formula for calculating model parameter consistency is: ; Where θ i is the model parameter vector of the i-th terminal, θ global is the global model parameter vector, and cos(·) is the cosine similarity function.
[0102] The effectiveness of knowledge transfer is to evaluate the performance improvement after knowledge is transferred from the source terminal to the target terminal. The calculation formula for the effectiveness of knowledge transfer is: ; Where σ1 is the model accuracy after migration, and σ2 is the model accuracy before migration.
[0103] Load balancing measures the degree of balance in the transaction loads processed by each cash register terminal. The calculation formula for load balancing is: ; Where σ(Ti) is the standard deviation of the transaction processing time of each terminal, It is the average transaction processing time of each terminal.
[0104] Based on the results of the difference analysis, targeted optimization and adjustment strategies are formulated. For issues with intelligent functions, the corresponding terminals can be instructed to increase data collection, optimize data annotation quality, and adjust model structures or parameters. For deficiencies in collaborative decision-making, collaborative decision-making rules can be further improved, parameter settings in intelligent algorithms can be optimized, or information communication mechanisms between pre-processed cash register data can be strengthened. Through continuous optimization and improvement, the overall performance and effectiveness of the distributed cash register system's intelligent collaboration can be continuously improved, ensuring the long-term stable and efficient operation of the system.
[0105] Through the above systematic and comprehensive steps, we can effectively solve the problems of large differences in intelligent functions and insufficient collaborative intelligence in the distributed cash register system in terms of intelligent collaboration, and create a highly collaborative, intelligent and efficient distributed cash register system to better meet the diverse needs of business operations and improve the overall cash register efficiency and service quality.
[0106] In summary, this embodiment provides a distributed checkout method that uses a federated transfer learning knowledge transfer model to mine common and transferable knowledge, and a weighted average aggregation algorithm to obtain global model parameters. The method then measures and optimizes the parameters based on intelligent function performance indicators. This solves the problem of uneven intelligent collaboration levels in distributed checkout systems due to evaluation indicators such as data differences among checkout terminals and inconsistent model performance, which affects overall service quality and operational efficiency.
[0107] The weighted average aggregation algorithm combines pre-processed cash register data from different terminals according to their respective weights, ensuring that each terminal's data plays an appropriate role in the overall system. The knowledge transfer model, on the other hand, focuses on discovering the common characteristics and transferable knowledge among these data. The combination of the two can comprehensively and deeply integrate the value inherent in multi-source data, avoiding the limitations of a single data source or single analytical method. This allows dispersed data resources across various entities to collaborate and play a greater role, collectively contributing to the improvement of the intelligence level of the entire distributed cash register system.
[0108] The transferable knowledge identified through the knowledge transfer model allows different cash register terminals, inventory management systems, and marketing management systems to better understand each other's data characteristics and business connections, enabling the sharing of experience and knowledge. Processing the integrated data using a weighted average aggregation algorithm further strengthens collaboration among various entities at the data level, making collaboration between different parts of the system closer and smoother, and overall moving towards a more optimized operational state, such as achieving seamless coordination between cash register, inventory management, and marketing activities.
[0109] Based on the same general inventive concept, the present invention also protects a distributed cashier system. The distributed cashier system provided by the present invention is described below. The reverse express status identification and tracking management system based on big data described below can be referenced to each other with the distributed cashier method described above.
[0110] Figure 2 This is a module diagram of a distributed cash register system provided by an embodiment of the present invention.
[0111] like Figure 2 As shown, a distributed cash register system includes: Data collection module, used to collect cash register data from each cash register terminal; The data preprocessing module preprocesses the cash register data to obtain preprocessed cash register data; The data analysis module uses the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between pre-processed cash register data to obtain a trained cash register data model. The data aggregation module is used to aggregate the parameters in the trained cashier data model using a weighted average aggregation algorithm based on the pre-processed cashier data and the weight of the pre-processed cashier data to obtain the global knowledge transfer model parameters; The data evaluation module is used to measure the level of collaboration of distributed cash registers using preset intelligent function performance indicators, and to optimize and adjust cash register terminals that are below the preset standards to obtain optimized distributed cash registers.
[0112] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods of each embodiment or certain portions of the embodiments.
[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A distributed cashier method, characterized in that: include: S1: Collect cash register data from each cash register terminal; S2: Preprocessing the cashier data to obtain preprocessed cashier data; S3: Using the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between the pre-processed cashier data, and obtain a trained cashier data model; S4: Aggregating the parameters in the trained cashier data model using a weighted average aggregation algorithm based on the pre-processed cashier data and the weight of the pre-processed cashier data to obtain global knowledge transfer model parameters; S5: Use the preset intelligent function performance indicators to measure the level of collaboration of distributed cashiers, and optimize and adjust the cashier terminals that are below the preset standards to obtain optimized distributed cashiers.
2. A distributed cashier method according to claim 1, characterized in that: In step S1, the cashier data includes: commodity-related data, including commodity image data, commodity barcode data, and commodity basic attribute data; transaction-related data, including transaction time data, transaction amount data, and payment method data; and also includes customer membership data, return and exchange data, and cashier operation data.
3. A distributed cashier method according to claim 1, characterized in that: In step S2, the preprocessing includes: Data cleaning: cleaning the cash register data to obtain cleaned cash register data; Data labeling: labeling the commodity-related data in the cleaning cash register data to obtain labeled cash register data; Feature extraction: converting the labeled cashier data into feature vectors.
4. A distributed cashier method according to claim 1, characterized in that: In step S3, the common features are the daily peak sales hours existing in all cash register terminals, the similar proportions of customers' commonly used payment methods, the relative stability of popular product categories, and the common matching patterns of customers' purchase combinations; the transferable knowledge is the response methods of busy terminals to complex customer needs, promotional activities techniques in specific scenarios, and product identification tips accumulated from some cash register terminals.
5. A distributed cashier method according to claim 1, characterized in that: In step S3, the specific steps of using the knowledge transfer model in federated transfer learning to analyze the common features and transferable knowledge between the pre-processed cash register data are as follows: S31: Setting a shared layer after the input layer of each pre-processed cashier data model, and extracting the common feature matrix of the shared layer through multi-data joint training and gradient optimization; S32: Calculating the feature difference and correlation between the pre-processed cashier data according to the shared layer common feature matrix to obtain a difference correlation matrix; S33: Designing a migration mapping function based on the difference association matrix, mapping the feature space of the pre-processed cashier data to the target data space, comparing the model performance under different mapping parameters, and selecting the optimal parameter combination; S34: Integrate the migration mapping function into the federated learning framework, monitor the performance indicators of each terminal model in real time, and when performance fluctuations are detected, dynamically adjust the mapping function parameters using the shared layer common feature matrix.
6. A distributed cashier method according to claim 1, characterized in that: In step S4, the specific steps of using the weighted average aggregation algorithm to aggregate the parameters in the trained cashier data model are as follows: S41: Collect cashier data model parameters of all distributed cashier terminals; S42: Calculate the weight of pre-processed cashier data of each terminal; S43: For each cashier data model parameter, calculate the weighted average of the parameters according to the weight of the cashier data pre-processed by the terminal to obtain the aggregation parameter θ global ; S44: Normalize the aggregated parameters to obtain normalized aggregated parameters θ global ; S45: The normalized aggregation parameter θ global Update the parameters of the cashier data model; S46: Dynamically adjust the weight distribution strategy of the terminal data according to the performance feedback of the cashier data model.
7. A distributed cashier method according to claim 6, characterized in that: In step S42, the calculation formula for the weight of the pre-processed cashier data is: ； Where, ω i is the final weight of the i-th pre-processed cashier data, n is the number of pre-processed cashier data, F k is the global weight of the kth quality assessment indicator, S i,k is the score of the i-th preprocessed cashier data on the k-th quality evaluation index, It is the sum of the total scores of all pre-processed cash register data.
8. A distributed cashier method according to claim 1, characterized in that: In step S5, the preset intelligent function performance indicators include: transaction processing efficiency indicator, transaction accuracy indicator, customer experience indicator, and system collaboration indicator; Transaction processing efficiency indicators include average transaction time, transaction throughput, and peak-hour processing capacity; Transaction accuracy indicators include transaction error rate, product identification accuracy, and amount matching rate; Customer experience indicators include customer waiting time and customer satisfaction; System collaboration indicators include model parameter consistency, knowledge transfer effectiveness, and load balance.
9. A distributed cashier method according to claim 8, characterized in that: The calculation formula for transaction error rate is: ； Where, E i is the number of occurrences of type i error, m is the total number of error types, and n is the total number of transactions.
10. A distributed cashier system, applied to a distributed cashier method according to any one of claims 1 to 9, characterized in that: The distributed cash register system includes: Data collection module, used to collect cash register data from each cash register terminal; A data preprocessing module preprocesses the cashier data to obtain preprocessed cashier data; A data analysis module is used to analyze the common features and transferable knowledge between the pre-processed cashier data using the knowledge transfer model in federated transfer learning to obtain a trained cashier data model; A data aggregation module is used to aggregate the parameters in the trained cashier data model using a weighted average aggregation algorithm based on the pre-processed cashier data and the weight of the pre-processed cashier data to obtain global knowledge transfer model parameters; The data evaluation module is used to measure the level of collaboration of distributed cash registers using preset intelligent function performance indicators, and to optimize and adjust cash register terminals that are below the preset standards to obtain optimized distributed cash registers.