Goods shelf management method and device and storage medium

Abstract

The invention discloses a goods shelf management method and device and a storage medium, and relates to the technical field of intelligent retail and warehouse management, and the goods shelf management method comprises the steps: collecting the state change data of a target commodity in a goods shelf through a goods shelf sensor, and obtaining the image data associated with the goods shelf; based on the state change data and the image data, determining behavior characteristics and attention popularity of the customer to the target commodity; according to the historical sales data and the current sales strategy of the target commodity, predicting the sales trend of the target commodity in a future preset time period, and obtaining a sales prediction result; and generating a corresponding optimized sales strategy based on the sales prediction result, the current sales strategy, the behavior characteristics and the attention popularity. According to the method, the commodity state change data and the shelf image data are integrated, customer behaviors and commodity attention popularity are accurately analyzed, the future sales trend is predicted in combination with the historical sales data, and the optimized sales strategy is generated, so that the intelligent level and sales efficiency of shelf management are improved.

Description

Technical Field

[0001] This application relates to the field of smart retail and warehouse management technology, and in particular to a shelf management method, equipment and storage medium. Background Technology

[0002] In smart retail systems, shelf replenishment decisions are primarily based on historical sales data and inventory monitoring results. However, this data only reflects the net change in inventory and cannot reconstruct the complete interaction process between the product and the customer before it is sold, including behaviors such as the product being picked up, put back, touched multiple times, and the customer's observation of the shelf. Therefore, when making replenishment judgments, the system lacks awareness of customer behavior, making it difficult to accurately assess the actual attention and potential demand for the product, thus affecting the accuracy of replenishment decisions.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a shelf management method, equipment and storage medium, which aims to solve the technical problem that the shelf management system does not accurately assess the actual demand for goods.

[0005] To achieve the above objectives, embodiments of this application provide a shelf management method, the shelf management method comprising: The system collects status change data of target products on the shelf using shelf sensors, and also acquires image data associated with the shelf; wherein, the status change data includes weight changes and displacement changes caused by the target products leaving or returning to the shelf; Based on the state change data and the image data, determine the customer's behavioral characteristics and level of interest in the target product; Based on the historical sales data and current sales strategy of the target product, predict the sales trend of the target product in a future preset time period to obtain the sales forecast result; Based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of attention, an optimized sales strategy is generated for the target product.

[0006] In one embodiment, before the step of predicting the sales trend of the target product within a preset future time period based on historical sales data and current sales strategies, and obtaining a sales forecast result, the shelf management method further includes: Based on the status change data collected by the shelf sensors, the sales quantity, sales time and corresponding sales price of the target product in multiple historical periods are determined, and the historical sales data of the target product is obtained.

[0007] In one embodiment, the step of determining the customer's behavioral characteristics and level of interest in the target product based on the state change data and the image data includes: The image data is subjected to target detection to identify customers lingering in front of the shelf and their interactive actions with the target products. The interaction action and the state change data are time-aligned and correlated to determine whether the interaction action results in the target product being taken out or put back. The behavioral characteristics are generated based on at least one of the following: the customer's dwell time in the area where the target product is located, the duration of visual attention, the number of times the product is picked up, the number of times the product is put back, and the tactile behavior that does not cause a change in state. Based on the statistical distribution and changing trends of the behavioral characteristics over multiple time periods, the customer's level of attention to the target product is determined.

[0008] In one embodiment, the step of predicting the sales trend of the target product within a preset future time period based on the historical sales data and current sales strategy of the target product, and obtaining the sales forecast result, includes: The historical sales data and the current sales strategy are used as input features and fed into a pre-trained sales prediction model. Based on the sales forecasting model, the expected sales volume of the target product within a preset future time period is output as the sales forecast result.

[0009] In one embodiment, the step of generating an optimized sales strategy for the target product based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of public interest includes: Based on the behavioral characteristics, the number of times the target product was taken and the number of times it was put back are obtained, as well as the actual number of times the target product was purchased. The actual number of purchases is the number of events in which the product was taken without being put back. Calculate the customer conversion rate of the target product based on the number of times it was taken and the number of times it was actually purchased; Based on the level of public interest, the target sales volume for the target product is determined. When the sales forecast result is lower than the target sales volume and the customer conversion rate is lower than the preset conversion threshold, the optimized sales strategy is generated based on the current sales strategy. The optimized sales strategy includes adjusting the price, promotion plan and / or display position of the target product.

[0010] In one embodiment, the step of determining the target sales volume corresponding to the target product based on the level of public interest includes: Obtain the historical popularity and corresponding actual sales volume of multiple reference products belonging to the same category as the target product; Based on the historical attention and the corresponding actual sales volume, establish a mapping relationship between the attention and the target sales volume; Based on the mapping relationship and the current popularity of the target product, the target sales volume is obtained.

[0011] In one embodiment, after the step of generating an optimized sales strategy based on the current sales strategy when the sales forecast result is lower than the target sales volume and the customer conversion rate is lower than a preset conversion threshold, the shelf management method further includes: Monitor the actual customer behavior data and sales performance of the target product after the implementation of the optimized sales strategy; The customer conversion rate is updated based on the actual customer behavior data, and the sales forecast result is updated based on the sales performance. When the updated customer conversion rate fails to reach the preset improvement threshold, or when the gap between the updated sales forecast and the target sales volume fails to narrow to the preset range, the parameters of the optimized sales strategy are iteratively adjusted.

[0012] In one embodiment, after the step of generating an optimized sales strategy for the target product based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of attention, the shelf management method further includes: Based on the sales forecast results, determine the expected sales volume of the target product within a preset future time period; In response to the promotional plan or price adjustment instruction included in the optimized sales strategy, the expected sales volume is adjusted by a coefficient to generate the final demand forecast result for the target product; Based on the demand forecast results and the current inventory status of the target product, a replenishment plan for the target product is generated.

[0013] This application also provides a shelf management device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the shelf management method described above.

[0014] This application embodiment also provides a storage medium, which is a computer-readable storage medium, and stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the shelf management method described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: This embodiment achieves real-time perception and quantitative analysis of customer-product interaction by collecting status change data of target products and acquiring image data associated with the shelf. This overcomes the limitations of traditional methods, which lack identification of pre-purchase behavior and rely solely on final transaction results, leading to biased and delayed sales analysis. Based on the collected status change and image data, the system can extract customer behavior characteristics in front of the shelf and calculate the target product's popularity, providing real-time and objective quantitative evidence for subsequent sales strategy optimization. Furthermore, by combining historical sales data with current sales strategies, sales trend predictions are made, ensuring the predictions reflect current market dynamics and improving the accuracy and timeliness of sales trend judgments. Finally, by integrating sales prediction results, current sales strategies, behavioral characteristics, and popularity, optimized sales strategies that match specific scenarios and are executable are generated. This allows shelf operations to dynamically adjust sales strategies and product displays based on real-time customer behavior feedback, thereby improving product conversion efficiency and inventory turnover, and effectively reducing lost sales opportunities and inventory backlog risks caused by lagging sales analysis or static decision-making. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the shelf management method of this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the shelf management method of this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the shelf management method of this application; Figure 4 This is a flowchart illustrating Embodiment 4 of the shelf management method of this application; Figure 5 This is a schematic diagram of the structure of the shelf management equipment in the hardware operating environment of the shelf management method in this application embodiment.

[0017] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0019] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0020] In smart retail systems, shelf replenishment decisions are primarily based on historical sales data and inventory monitoring results. However, this data only reflects the net change in inventory and cannot reconstruct the complete interaction process between the product and the customer before it is sold, including behaviors such as the product being picked up, put back, touched multiple times, and the customer's observation of the shelf. Therefore, when making replenishment judgments, the system lacks awareness of customer behavior, making it difficult to accurately assess the actual attention and potential demand for the product, thus affecting the accuracy of replenishment decisions.

[0021] In view of the above problems, this application proposes a shelf management method, which collects state change data of target products on the shelf through shelf sensors and obtains image data associated with the shelf; wherein, the state change data includes weight changes and displacement changes caused by the target product leaving or returning to the shelf; based on the state change data and the image data, the behavioral characteristics and attention intensity of customers towards the target product are determined; based on the historical sales data and current sales strategy of the target product, the sales trend of the target product in the future within a preset time period is predicted to obtain a sales forecast result; based on the sales forecast result, the current sales strategy, the behavioral characteristics, and the attention intensity, an optimized sales strategy for the target product is generated.

[0022] This application provides a solution that, by collecting data on the status changes of target products and acquiring image data associated with the product shelf, achieves real-time perception and quantitative analysis of the customer-product interaction process. This overcomes the limitations of traditional methods, which lack identification of pre-purchase behavior and rely solely on the final transaction result, leading to one-sided and lagging sales analysis. Based on this, the system can extract customer behavior characteristics in front of the shelf using the collected status change and image data, and calculate the attention level of target products accordingly, providing real-time and objective quantitative basis for subsequent sales strategy optimization. Furthermore, by combining historical sales data with current sales strategies, sales trend predictions are made, ensuring that the prediction results reflect current market dynamics and improving the accuracy and timeliness of sales trend judgment. Finally, by integrating sales prediction results, current sales strategies, behavioral characteristics, and attention levels, optimized sales strategies that match specific scenarios and are executable are generated. This allows shelf operations to dynamically adjust sales strategies and product displays based on real-time feedback from customer behavior, thereby improving product conversion efficiency and inventory turnover rate, and effectively reducing the risk of lost sales opportunities and inventory backlog caused by lagging sales analysis or static decision-making.

[0023] It should be noted that the executing entity in this embodiment can be an intelligent computing device or system deployed in a retail environment. This entity typically includes a data acquisition module, an edge computing unit, and a central processing server. Specifically, the data acquisition module consists of various sensors (such as weight and displacement sensors) and image acquisition devices (such as cameras) deployed on the shelf, responsible for collecting raw state change data and image data. The edge computing unit can be integrated locally on the shelf or within the store's local area network for real-time preprocessing, feature extraction, and preliminary analysis of the collected data. The central processing server can be a server cluster located in a local data center or in the cloud, responsible for running complex predictive models, multi-objective optimization algorithms, and strategy generation engines, and interfacing with the store management system (such as an inventory system and a price management system). The following description uses an intelligent shelf management system integrating the above modules as an example to illustrate this embodiment and the subsequent embodiments.

[0024] Please refer to the shelf management method of the first embodiment proposed in this application. Figure 1 The method includes steps S10 to S40: Step S10: Collect status change data of the target product in the shelf through the shelf sensor, and obtain image data associated with the shelf; wherein, the status change data includes weight change and displacement change caused by the target product leaving or returning to the shelf.

[0025] It should be noted that shelf sensors refer to devices integrated or installed on shelves to sense changes in physical quantities. These typically include weight sensors (such as strain gauges and piezoelectric sensors) and displacement sensors (such as infrared photoelectric sensors, laser rangefinders, or gravity-sensing gyroscopes). Their physical support forms include rollers, hooks, and trays. Status change data refers to the set of quantified physical parameter changes captured in real time by sensors when goods are picked up, placed back, moved, or replaced. Image data refers to static images or video streams containing target goods and their surrounding environment captured by cameras, vision sensors, or inspection robots deployed in the shelf area, used to assist in identifying the status of goods and user behavior. Weight change refers to the increase or decrease in the shelf's load-bearing capacity caused by the handling of goods. Displacement change refers to the movement or tilting of goods within the three-dimensional space of the shelf.

[0026] As a feasible implementation method, an array of pressure sensors is deployed under each pallet on the shelf to detect changes in weight distribution. At the same time, millimeter-wave radar or ToF (Time-of-Flight) sensors are installed on the sides or top of the shelf to detect product displacement and hand approach trajectories in a non-contact manner.

[0027] As an example, when a customer picks up a beverage, the pressure sensor in the corresponding shelf detects a decrease in weight. Simultaneously, a millimeter-wave radar installed on top of the shelf scans its coverage area. The point cloud data, processed by a target detection algorithm, identifies the trajectory of the hand approaching the shelf and tracks the movement of the product as it is removed. The pressure sensor records the moment of weight change, and the radar detects the beginning of product displacement. These are synchronized with a control unit inside the shelf, generating a composite state change signal containing time, weight change, and displacement information. This composite state change signal is then sent to an edge computing device deployed locally on the shelf for processing. The edge computing device uses a filtering algorithm to smooth noise in the composite state change signal, extracting the valid signal caused by customer behavior. Furthermore, a camera fixed above the shelf captures a video stream, with each frame timestamped in sync with the sensor data stream. When processing the composite state change signal of weight and displacement, the system automatically extracts the image frame corresponding to the timestamp for subsequent visual verification of whether the product has been taken and for identifying details of customer behavior.

[0028] Step S20: Based on the state change data and the image data, determine the customer's behavioral characteristics and level of attention to the target product.

[0029] It should be noted that behavioral characteristics refer to quantifiable and recordable behavioral patterns of customers during their interaction with target products. These include the duration of customer dwell time in the shelf area where the target product is located, visual attention duration, number of times the target product is picked up, duration of each pick-up, number of times the product is put back, and non-removal interaction behaviors that do not cause changes in state (weight or displacement), such as shaking the product, flipping to examine the packaging, and comparing the product horizontally with other products. Among these, dwell time can be calculated by counting the number of consecutive frames in the image data showing the customer in that shelf area, and visual attention duration can be estimated by analyzing the spatial relationship between the customer's facial orientation and the location of the target product in the image.

[0030] Attention level refers to a comprehensive indicator used to measure the degree to which a target product is noticed or actively interacted with by customers within a preset time period. Its calculation can integrate data from multiple dimensions, including the number of times the product is picked up, the number of times it is put back, the duration of each pick-up, the duration of dwell time, the duration of visual attention, and the number of customers who interact within a unit of time.

[0031] In this embodiment, pattern analysis is first performed on the state change data to detect product interaction events related to the target product, including events such as the target product being picked up, put back, or moved. The occurrence time, duration, and corresponding state change amount of each product interaction event are recorded, and the number of occurrences for each type of product interaction event is accumulated. Specifically, the two state changes of the target product leaving the shelf and the product being put back into the shelf are detected separately. When a change in weight or displacement is detected, it is recorded as a product picking event; when a change in weight or displacement is detected, it is recorded as a product putting event. By continuously tracking the correspondence between these two types of events on the shelf corresponding to the target product, the difference between the total number of product picking events and the total number of product putting events can be calculated, thereby determining the actual number of times the target product was purchased. This method can comprehensively reflect various shopping scenarios, including situations where a customer takes the target product and then returns and puts it back after a considerable period of time.

[0032] Optionally, a time-window-based auxiliary determination mechanism is introduced to determine the event type by detecting changes in weight or displacement within a preset time period. Specifically, when a decrease in weight or displacement is detected followed by a return to its original value within the preset time period, it is recorded as a product return event; when a decrease in weight or displacement is detected but does not return to its original value within the preset time period, it is recorded as a product retrieval event. In this embodiment, the number of retrieval events can be directly counted as actual purchase behavior.

[0033] While detecting product interaction events, image analysis is performed on the collected image data or video streams. For example, a pre-trained hand keypoint detection model and head pose estimation model can be used to identify the relative position of the customer's hand and the shelf where the target product is located in the image frame, as well as the angle of the customer's face relative to the shelf, thereby determining whether the customer's gaze is on the shelf area where the target product is located, and calculating the length of time the customer stays in that shelf area.

[0034] Next, the time-aligned product interaction events are correlated with the image analysis results. For example, the start time of a product retrieval event is matched with an event detected in the image analysis within the same time period where a customer's hand approaches the shelf, thereby obtaining multi-dimensional behavioral feature data including interaction frequency, interaction duration, visual attention duration, and dwell time. Through this correlation, it is possible to effectively distinguish genuine customer interactions from other non-customer activities. Specifically, when the system detects a product interaction event, by examining the image analysis results within the same time period, the identity of the person performing the product interaction action can be determined. If the image analysis identifies the person performing the action as wearing a store clerk uniform, the product interaction event can be marked as a store clerk restocking; if it is identified as a regular customer, it is marked as a customer interaction. In addition, by analyzing the spatiotemporal relationship between the customer's facial orientation and hand movements in the image data, the system can confirm whether the customer's visual focus matches the shelf location corresponding to the product interaction event, thereby determining whether the customer's attention is truly focused on the target product being interacted with, ensuring the accuracy of the behavioral feature analysis.

[0035] Subsequently, the system generates behavioral characteristic records corresponding to the target product. These behavioral characteristic records include: statistical time window, number of times the product was picked up, total picking time, number of times the product was put back, actual purchase number, average picking time per time, total number of people standing in the shelf area where the target product is located, total standing time and average standing time, total visual attention time, and other indicators.

[0036] Finally, based on a pre-defined weighting model or through machine learning algorithms, the aforementioned behavioral characteristic data is comprehensively calculated to obtain a quantified popularity score. Specifically, each behavioral characteristic indicator is assigned a corresponding weight, and a comprehensive value is obtained through weighted calculation. This value is the popularity score of the target product within the statistical time window. This popularity score can be used for horizontal comparison between different products or trend analysis of the same product over different time periods.

[0037] As an example, the system identified two valid product return events and five valid product retrieval events for coffee located in shelf A03 within a ten-minute period from 15:00 to 15:10. The corresponding single retrieval durations were 5 seconds, 7 seconds, 12 seconds, 8 seconds, and 2 seconds, respectively, with a total retrieval time of 34 seconds. Therefore, the actual number of purchases of this coffee product is the number of retrieval events minus the number of product return events, i.e., 5 - 2 = 3 times. Simultaneously, image analysis confirmed that three customers lingered in front of shelf A, with lingering times of 15 seconds, 30 seconds, and 10 seconds, respectively, for a total lingering time of 55 seconds. The system quantifies and scores each behavioral characteristic indicator according to preset weighting rules, where: the weight for product return events is 0.2, the weight for product retrieval events is 0.3, the weight for total retrieval time is 0.1, the weight for actual purchase events is 0.2, the weight for total number of lingering customers is 0.1, and the weight for total lingering time is 0.1. Based on the aforementioned behavioral characteristic data and corresponding weights, the attention level of this coffee product during this period can be calculated as (2×0.2)+(5×0.3)+(34×0.1)+(3×0.2)+(3×0.1)+(55×0.1)=0.4+1.5+3.4+0.6+0.3+5.5=11.7. This calculation result can be normalized or the original value can be retained as needed. This process clearly transforms multi-source behavioral characteristic data into specific behavioral characteristic statistics and quantified attention levels.

[0038] Step S30: Based on the historical sales data and current sales strategy of the target product, predict the sales trend of the target product in a future preset time period to obtain the sales forecast result.

[0039] It should be noted that historical sales data includes past sales volume, sales time (such as seasons and holidays), and time-series data on sales changes during various promotional activities. Current sales strategies refer to marketing parameters such as pricing, discounts, buy-one-get-one-free or bundled sales, and display placement that are currently being implemented. Sales forecasts refer to the expected sales volume or daily sales curve of the target product within a preset future time period (such as the next 24 hours or 7 days), and may further include warnings about the risks of stockouts and slow-moving inventory.

[0040] In this embodiment, before predicting the sales trend of the target product in a future preset time period based on the historical sales data and current sales strategy of the target product and obtaining the sales forecast result, the sales quantity, sales time and corresponding sales price of the target product in multiple historical time periods are first determined based on the status change data collected by the shelf sensor, so as to obtain the historical sales data of the target product.

[0041] As one possible implementation, step S30 includes steps S310 to S320: Step S310: Input the historical sales data and the current sales strategy as input features into the pre-trained sales prediction model.

[0042] Step S320: Based on the sales forecasting model, output the expected sales volume of the target product within a preset future time period as the sales forecast result.

[0043] In this embodiment, a sales forecasting model can be constructed to predict the future sales trend of the target product. Specifically, firstly, historical sales data of the target product is collected and organized, and daily sales volume is arranged in chronological order to form a historical sales time series. At the same time, various parameters in the current sales strategy are converted into structured features. For example, the discount level is represented as a value between 0 and 1, with no discount being 1 and 50% off being 0.5; buy-one-get-one-free activities are converted into equivalent discount rates; and the display location is divided into different levels according to its effectiveness in reaching customers and mapped to discrete integers, for example, the top shelf is set to 3, the second shelf (usually the prime display shelf) is set to 4, the middle shelf is set to 2, and the bottom shelf is set to 1.

[0044] Subsequently, the aforementioned historical sales time series and current sales strategy features are input into the sales forecasting model. This model uses the historical sales time series as the primary input and the corresponding date's promotional status, price changes, seasonal characteristics (e.g., weekday, month), holiday indicators (e.g., whether it is a statutory holiday), and current sales strategy features as auxiliary input features. During the training phase, parameters are optimized by minimizing prediction errors (e.g., MAE or RMSE). During the inference phase, the current sales strategy features are fixed, and a sliding prediction window generates the expected sales volume for a future preset time period. This sales forecasting model can employ time series models (e.g., ARIMA, Prophet) or regression models based on ensemble learning (e.g., Random Forest, XGBoost).

[0045] As an example, to predict the daily sales volume of a beverage for the next three days, the system first generates recent sales time-series features (e.g., the average daily sales volume over the past 7 days) based on the beverage's historical sales data. Simultaneously, it obtains current sales strategy parameters, including quantifying the "buy one get one half price" promotion as an equivalent discount rate of 0.75, quantifying the current display location (second shelf) as a value of 4, and determining the weekday feature and holiday identifier corresponding to the prediction date. Based on these features, the sales forecast model outputs the expected daily sales volume for the next three days, for example, a result of [120, 135, 118] (unit: units). This result is the sales forecast result.

[0046] Step S40: Based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of attention, generate an optimized sales strategy for the target product.

[0047] It should be noted that optimizing sales strategies refers to adjustment suggestions or automatic execution instructions generated to improve the sales performance of target products (such as increasing sales volume or clearing inventory).

[0048] As a feasible implementation method, a multi-objective optimization model can be constructed to generate optimized sales strategies that balance sales efficiency and customer experience. The multi-objective optimization model takes the expected sales revenue corresponding to the sales forecast results (such as the expected sales volume within a preset future time period) as the primary optimization objective; at the same time, it takes user satisfaction indicators reflected by behavioral characteristics and attention intensity (such as actual purchase conversion rate, average holding time, visual attention time, etc.) as constraints or secondary optimization objectives to avoid simply pursuing sales volume at the expense of customer experience (such as excessive price reductions leading to brand devaluation, or frequent promotions causing customers to wait).

[0049] Specifically, the decision variables are first defined, including adjustable strategy parameters such as discount rates. Whether to activate the buy-one-get-one-free promotion Display hierarchy (1 is the bottom layer, 2 is the middle layer, 3 is the top layer, 4 is the second layer), etc. Then, based on the sales forecasting model trained in step S30, a revenue objective function is constructed:

[0050] The expected sales volume is output by the sales forecasting model based on the parameters of the current sales strategy.

[0051] Simultaneously, user satisfaction constraints or objectives are introduced. For example, a satisfaction function is constructed using the mapping relationship between behavioral characteristics and attention intensity established in step S20. Its input is strategy parameters, and its output is the expected level of attention or conversion rate. Constraints can be set:

[0052] That The required attention threshold for maintaining basic customer interest; or, if this is taken as a secondary objective, a bi-objective optimization problem is formed:

[0053] The initial solution uses the current sales strategy parameters. This ensures that sales strategy optimization is based on actual operational conditions. Within the feasible solution space defined by sales strategy parameters (such as discounts not falling below cost price, and physical limitations on the display level of target products), a multi-objective optimization algorithm (such as NSGA-II, MOEA / D, or weighted summation) is used to search within the feasible region of the strategy to obtain a Pareto optimal solution set. This Pareto optimal solution set possesses Pareto optimality, meaning that for any solution, there are no other feasible strategies that can increase expected sales revenue without decreasing customer satisfaction, nor are there any strategies that can increase customer satisfaction without decreasing expected sales revenue. This implies that each solution in this set represents an optimal balance between sales objectives and customer experience.

[0054] Finally, from the Pareto optimal solution set mentioned above, a recommended strategy can be selected according to preset business rules. For example, the solution with expected sales revenue increasing by more than 10% compared to the current strategy and user satisfaction index not lower than the historical average can be selected as the final recommendation. The specific sales strategy parameters corresponding to this solution (such as suggested discount rate, promotional activity format, display location adjustment plan, etc.) constitute the generated optimized sales strategy. The system can output the optimized sales strategy to operations personnel for final review and manual decision-making, or in highly automated scenarios, it can directly drive subsequent automated execution systems based on the optimized sales strategy, such as adjusting electronic price tags or directing robots or robotic arms to adjust product displays. Through this method, the system can pursue sales goals while also considering customer experience, avoiding strategy deviations caused by optimizing a single objective.

[0055] This embodiment achieves real-time perception and quantitative analysis of customer-product interaction by collecting status change data of target products and acquiring image data associated with the shelf. This overcomes the limitations of traditional methods, which lack identification of pre-purchase behavior and rely solely on final transaction results, leading to biased and delayed sales analysis. Based on the collected status change and image data, the system can extract customer behavior characteristics in front of the shelf and calculate the target product's popularity, providing real-time and objective quantitative evidence for subsequent sales strategy optimization. Furthermore, by combining historical sales data with current sales strategies, sales trend predictions are made, ensuring the predictions reflect current market dynamics and improving the accuracy and timeliness of sales trend judgments. Finally, by integrating sales prediction results, current sales strategies, behavioral characteristics, and popularity, optimized sales strategies that match specific scenarios and are executable are generated. This allows shelf operations to dynamically adjust sales strategies and product displays based on real-time customer behavior feedback, thereby improving product conversion efficiency and inventory turnover, and effectively reducing lost sales opportunities and inventory backlog risks caused by lagging sales analysis or static decision-making.

[0056] Based on the above embodiments of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 In the shelf management method, step S20 includes steps S210 to S240: Step S210: Perform target detection on the image data to identify customers lingering in front of the shelf and their interactive actions with the target product.

[0057] It should be noted that interactive actions refer to behaviors in which a customer comes into contact with or intends to approach a target product, such as reaching out, picking up, putting down, touching, shaking, flipping to view, or moving the product while holding it.

[0058] In this embodiment, a deep learning-based object detection model, such as the YOLO series or Faster R-CNN, is used to process the image data. This object detection model can be pre-trained on a dataset containing various customer poses, hand movements, and product scenes, enabling real-time detection of the customer's bounding box, hand position, and the spatial relationship between the hand and the target product from image data or video streams. Specifically, the object detection model processes image frames to identify the customer as a whole, the hand region, and the target product in the image data, and outputs their respective bounding boxes. Based on this, a pre-trained pose estimation model (such as OpenPose) or a hand keypoint detection model (such as MediaPipeHands) is used to locate and track keypoints on the customer's torso, arms, hands, and even fingers. By analyzing the spatial position, motion trajectory, and spatial relationship between these keypoints and the target product's bounding box, such as overlap and relative distance, the system can more accurately determine the type and intent of the interaction. For example, when a hand keypoint is detected entering the product boundary frame and accompanied by a grasping gesture, while the arm keypoint shows an outward movement trajectory, it can be identified as a "picking up" action; if the hand keypoint is within the product boundary frame but has no obvious displacement, and the fingers are in an extended or slightly bent state to explore the packaging surface, it can be identified as a browsing action such as "touching" or "viewing".

[0059] Step S220: Perform time alignment and correlation analysis on the interaction action and the state change data to determine whether the interaction action causes the target product to be taken out or put back.

[0060] In this embodiment, the timestamp of each interaction is recorded and compared with the state change data recorded by the shelf sensors to find the correspondence between the two. For example, if a customer picks up an item at a certain moment, and the quantity, weight, or displacement of that item on the shelf decreases at the same moment, then the interaction can be considered as causing the item to be taken out.

[0061] Step S230: Generate the behavioral features based on at least one of the following: the customer's dwell time in the area where the target product is located, the duration of visual attention, the number of times the product is picked up, the number of times the product is put back, and the touch behavior that does not cause a change in state.

[0062] It should be noted that the dwell time refers to the length of time a customer stays in the shelf area where the target product is located; the visual attention time is the cumulative time that the customer's gaze is focused on the shelf area where the target product is located, determined by analyzing the direction of the customer's face or gaze; the number of times the product is picked up and the number of times it is put back represent the number of times the product is picked up or put back that are determined to be valid product interaction events and cause a significant change in the shelf status; the touch behavior that does not cause a change in status refers to a touch action that does not change the quantity, weight or displacement of the product, or it can also be understood as the customer performing a slight operation on the target product (such as flipping the packaging, shaking to check, lifting it a short distance and putting it back immediately, etc.), but the weight change or displacement caused by the slight operation does not exceed the preset threshold, or although there is an instantaneous change, it recovers to the original state within a very short time window (this time window is different from the preset period of the time window-based auxiliary judgment mechanism in the aforementioned step S20, and is shorter than the preset period, which can be represented as the second preset period), and therefore does not constitute a valid product picking event or product putting event.

[0063] In this embodiment, the determination of touch behavior that does not cause a change in state differs from the time window pattern matching method that simply relies on "detecting a decrease in weight or displacement followed by recovery within a preset time period." The touch behavior mentioned in this embodiment focuses more on the specific physical parameters of the interaction action, such as whether physical characteristics like the amount of weight change, displacement amplitude, and duration of change exceed a preset effective interaction threshold, rather than simply performing "change-recovery" pattern matching by setting a fixed-length time window. This allows for a more accurate distinction between browsing touches and picking behaviors indicating a purchase intention.

[0064] In this embodiment, to determine the dwell time, the spatial range of the shelf area where the target product is located is first obtained, and the total dwell time of the customer in the shelf area is calculated by continuously tracking the customer's movement trajectory in the shelf area.

[0065] Regarding visual attention duration, a multi-strategy fusion approach to gaze analysis can be employed due to potential angle, occlusion, or resolution limitations of customer faces in actual monitoring scenarios. In scenarios where the customer's face is clearly visible in the image data, gaze estimation techniques can be used. By analyzing the position and orientation of key facial points (such as eyes and nose tip), combined with the three-dimensional spatial coordinates of the target product and its shelf, the gaze point can be calculated to determine whether the gaze falls on the shelf area, and the cumulative duration of gaze on the shelf area can be recorded. When the customer's face is not visible or is occluded in the image data, an approximate judgment can be made based on head or body orientation. For example, the facing direction can be inferred from shoulder or torso posture. If the angle between the customer's head or body orientation and the shelf area is less than a preset angle threshold and persists for a preset duration, it is considered valid attention, and the corresponding visual attention duration is recorded. In extreme cases (such as when only the customer's back is visible), indirect inference can be made by combining customer hand movements and pausing behavior. For example, if a hand reaches towards a shelf during a prolonged pausing period, this can also be included as a proxy indicator of visual attention duration. The above methods can be used individually or in combination to adapt to actual deployment scenarios under different viewing angles, lighting conditions, and occlusion conditions.

[0066] The system can directly count the number of times items are taken and returned based on the list of valid product interaction events confirmed in step S20.

[0067] For touch behaviors that do not cause a change in state, collaborative analysis can be performed by combining tracking data of key hand points with displacement sensor data (such as point cloud changes from millimeter-wave radar): by tracking the movement trajectory of key hand points, it can be determined whether the product has been shaken, flipped, or lifted in place, while simultaneously monitoring in real time whether the weight change or displacement amplitude is always below the effective interaction threshold, or whether the weight change or displacement amplitude recovers within a second preset time period; if the above conditions are met, it is determined to be a browsing touch behavior.

[0068] The real-time calculation results of the above indicators will be recorded in a structured manner, collectively forming behavioral characteristics that describe the interaction process between customers and target products.

[0069] Step S240: Based on the statistical distribution and changing trends of the behavioral characteristics over multiple time periods, determine the customer's level of interest in the target product.

[0070] It should be noted that attention intensity is a comprehensive indicator used to measure the degree to which a target product is noticed or actively interacted with by customers within a preset time period. It is a quantified value obtained by statistically analyzing and weighting the values ​​of multiple behavioral characteristic indicators (such as dwell time, number of times the product is picked up, number of times it is put back, and visual attention duration) over different time periods. The analysis of statistical distribution and trends may include calculating the mean, variance, and slope of change over time for the behavioral characteristic indicators, as well as comparing different time periods (such as peak and off-peak hours).

[0071] As a feasible implementation method, the system can summarize and statistically analyze all behavioral characteristics collected within a fixed time window (e.g., every 10 minutes, every hour). For each time window, statistical values ​​of various behavioral characteristic indicators can be calculated first, such as the total number of times items are taken, the total duration of taking items, the average duration of a single take, the total visual attention duration, the average visual attention duration, the total dwell time, and the average dwell time. Subsequently, a multi-indicator comprehensive evaluation method (e.g., weighted summation) is used to calculate the attention score of the target product. In the process of calculating the attention score, preset weights reflecting their importance can be assigned to different behavioral characteristic indicators, such as assigning a higher weight to the number of times items are taken, a medium weight to the visual attention duration, and a basic weight to the dwell time.

[0072] As another feasible implementation method, after calculating the attention score of the target product using a multi-index comprehensive evaluation method, further trend analysis can be performed on the attention scores over multiple consecutive time windows. This includes calculating the moving average, the rate of change, or identifying whether the attention score shows an upward or downward trend. Based on the trend analysis results, the system dynamically adjusts the original attention score for the current time window. Specifically, adjustment coefficients corresponding to different trend states are pre-configured. If the attention score continues to rise over multiple consecutive time windows, the original attention score is moderately increased by the corresponding adjustment coefficient, i.e., original attention score × (1 + adjustment coefficient), to enhance the system's responsiveness to increases in attention. Conversely, if a downward trend is observed, the score can be appropriately decreased, i.e., original attention score × (1 - adjustment coefficient), to avoid misjudgments due to short-term fluctuations. If the trend is stable, the original attention score remains unchanged. Finally, based on the original attention score for the current time window and the adjustment coefficient determined by the trend analysis results, the final attention value of the target product is obtained.

[0073] This implementation method introduces a trend perception mechanism, which enables the attention intensity value to not only reflect the intensity of customer behavior in the current period, but also effectively capture the evolution direction of customer interests, thereby providing a more forward-looking basis for subsequent sales strategy optimization.

[0074] As an example, if a product's popularity score is detected to be 85, 92, and 100 in the past three time windows, indicating a significant upward trend, an upward adjustment factor of 0.1 (i.e., multiplying the original popularity score corresponding to the current time window by 1.1) can be applied to adjust the final popularity value from 100 to 110, thereby triggering promotional or restocking suggestions more sensitively. Conversely, if the popularity score declines continuously, the final popularity value can be appropriately lowered to prevent misjudging demand due to occasional peaks.

[0075] It's important to note that the moving average is a method of smoothing out short-term fluctuations and revealing long-term trends by calculating the average of multiple consecutive windows. Its core principle is to define a fixed-length window that moves sequentially along the timeline, calculating the average of the data within that window after each move.

[0076] As an example, if the popularity scores of a consecutive window sorted by time are S1, S2, S3, S4, S5, etc., and the window size is 3, then the first moving average (corresponding to time point 3) = (S1+S2+S3) / 3; the second moving average (corresponding to time point 4) = (S2+S3+S4) / 3; the third moving average (corresponding to time point 5) = (S3+S4+S5) / 3, and so on, moving forward in sequence.

[0077] Based on the above embodiments of this application, in the third embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 In the shelf management method, step S40 includes steps S410 to S440: Step S410: Based on the behavioral characteristics, obtain the number of times the target product was taken and the number of times it was put back, as well as the actual number of times the target product was purchased. The actual number of purchases is the number of events in which the product was taken without being put back.

[0078] It should be noted that the actual number of purchases is the difference between the number of times items were taken and the number of times they were put back; that is, the number of times items were taken minus the number of times they were put back. This actual number of purchases directly reflects the number of events in which customers took the target item from the shelf and ultimately completed the purchase, and is an important indicator for measuring the actual purchase behavior of the product.

[0079] Step S420: Calculate the customer conversion rate of the target product based on the number of times the product was taken and the number of times it was actually purchased.

[0080] It's important to note that customer conversion rate refers to the efficiency of converting interest in a target product (manifested as picking up the product) into a final purchase. The formula is: Customer Conversion Rate = Actual Purchases / Number of Pickups. This ratio reflects the product's ability to convert customer interest into actual purchases.

[0081] Step S430: Based on the level of attention, determine the target sales volume corresponding to the target product.

[0082] It should be noted that the target sales volume refers to an expected and achievable sales volume reference value set based on the current level of attention the target product receives. This target sales volume is not a fixed historical average, but a sales target dynamically linked to real-time popularity.

[0083] As one possible implementation, step S430 includes steps S4310 to S4330: Step S4310: Obtain the historical popularity and corresponding actual sales volume of multiple reference products belonging to the same category as the target product.

[0084] Step S4320: Based on the historical attention and the corresponding actual sales volume, establish a mapping relationship between the attention and the target sales volume.

[0085] Step S4330: Based on the mapping relationship and the current popularity of the target product, obtain the target sales volume.

[0086] In this embodiment, the system can establish a mapping relationship between popularity and actual sales volume based on the historical sales data of the target product, or by combining the historical popularity and corresponding actual sales volume of reference products in the same category, and convert the current popularity of the target product into the target sales volume accordingly. For example, the system can statistically analyze the average daily sales volume of the same category of products in different popularity intervals to construct a popularity-sales volume comparison table; after obtaining the current popularity of the target product, the corresponding target sales volume can be directly obtained or interpolated according to this comparison table. This mapping relationship can be maintained separately according to product category or store type to improve the accuracy and applicability of the target sales volume.

[0087] As another feasible implementation, a regression prediction model can be trained to determine the target sales volume. This model uses popularity as its core feature and can incorporate auxiliary inputs such as product category, current price, shelf location, seasonality, historical customer conversion rate of the target product, and competitor popularity to output a target sales volume that matches the current market environment. The model can be trained offline using paired samples of historical popularity and corresponding actual sales volume for different time periods. Algorithms such as linear regression, gradient boosting trees (e.g., XGBoost), or neural networks are employed to learn the non-linear impact of popularity on sales volume. After training, the current popularity of the target product and other contextual features are input into the regression prediction model to generate the corresponding target sales volume in real time. Compared to a static lookup table, this implementation method can more flexibly capture the complex non-linear relationship between popularity and sales volume and can adapt to the sales characteristics of different products.

[0088] Step S440: When the sales forecast result is lower than the target sales volume and the customer conversion rate is lower than the preset conversion threshold, the optimized sales strategy is generated based on the current sales strategy. The optimized sales strategy includes adjusting the price, promotion plan and / or display position of the target product.

[0089] It should be noted that the preset conversion threshold is the minimum acceptable conversion rate set based on the historical performance of the target product or the average level of similar products, and is used to identify the problems of "high popularity but low conversion" or "inefficient exposure".

[0090] The core of this embodiment is that when the target product has a high level of attention (specifically reflected in a corresponding target sales volume that is not low), but the sales forecast results indicate that the future sales volume will not reach the expected target sales volume, and the customer conversion efficiency is also low, it indicates that the problem mainly lies in the process of converting customer interest into purchase. Therefore, it is necessary to generate targeted and optimized sales strategies to improve conversion, such as adjusting the price of the target product, promotional plans and / or display location.

[0091] In this embodiment, a rule engine can be used to automatically generate optimized sales strategies. Each rule consists of a condition judgment part and a strategy execution part. The condition judgment part sets specific thresholds or pattern recognition conditions based on multi-dimensional indicators such as sales forecast results, current sales strategies, target sales volume, behavioral characteristics, and popularity. The execution part corresponds to specific strategy suggestions or operation instructions.

[0092] For example, a rule can be set: when the sales forecast results for a target product show that the average daily expected sales volume in the future preset time period is lower than the target sales volume, and the customer conversion rate is lower than the preset conversion threshold, the rule engine determines that the target product has a high interest but low conversion problem, and triggers the generation of optimization strategy suggestions such as "price reduction, limited-time discount, buy-one-get-one-free activity or limited-time promotion".

[0093] As a feasible implementation method, optimized sales strategies can also be generated directly based on a combination of factors, including attention levels and behavioral characteristics such as the number of times a product is picked up, average dwell time, and total visual attention time. For example, when the attention level of a target product is higher than the average level of similar products, but the ratio of the number of times a product is picked up to the visual attention time is lower than a preset threshold (indicating that some people are only looking but not picking up products), the rule engine may determine that the product information is insufficient and trigger the generation of a strategy suggestion to "add product descriptions or trial packs next to the shelf." As another example, when sales forecasts show that the average daily sales volume within a preset time period will be significantly higher than the current available inventory level, the rule engine will trigger the generation of an optimized strategy suggestion to "increase replenishment priority and generate a stocking warning" to prevent stockouts. Furthermore, when the attention level of a target product remains consistently low, but historical sales are stable, and behavioral characteristics show that the average dwell time and visual attention time of customers are significantly lower than the average level of the shelf area, the rule engine determines that the target product's display location is poor and generates an optimized suggestion to "adjust the display location or add directional signs to adjacent high-traffic shelves."

[0094] As another feasible implementation, a sales strategy generation method based on multi-objective optimization can be adopted. This involves constructing the generation of an optimized sales strategy as a mathematical optimization problem. In this problem, the decision variables are adjustable sales strategy parameters, such as price adjustment range, promotional discount rate, buy-one-get-one-free promotions, adjusting display locations, adding directional signage, adding product descriptions, and providing trial packs. The optimization objectives can be set as maximizing expected sales revenue, minimizing inventory backlog costs, and maximizing customer satisfaction (reflected by positive indicators in attention and behavioral characteristics). Constraints include, but are not limited to: the adjusted optimized sales strategy meeting the requirement that sales forecast results are not lower than the target sales volume, customer conversion rate increasing to not lower than a preset conversion threshold, and business constraints such as cost budget for optimizing the sales strategy, discounts not falling below cost price, physical limitations on the display level of target products, and store display rules. Starting with the current sales strategy, the system searches within the feasible solution space of the sales strategy parameters using optimization algorithms (such as linear programming, genetic algorithms, or sequential minimum optimization algorithms) to find the optimal strategy solution or Pareto optimal solution set that simultaneously satisfies multiple business objectives and constraints. Operators can select the optimization strategy to be implemented, or the system can automatically select it based on preset priorities.

[0095] As another feasible implementation, step S440 includes steps S4410 to S4430: Step S4410: Monitor the actual customer behavior data and sales performance of the target product after the implementation of the optimized sales strategy.

[0096] It should be noted that actual customer behavior data refers to data newly collected by the system after the implementation of optimized sales strategies, using shelf sensors and image acquisition devices (such as cameras, vision sensors, or inspection robots deployed in the shelf area), reflecting the interaction status between customers and target products. Examples include new pick-up frequency, return frequency, dwell time, and visual attention duration. Sales performance, on the other hand, refers to the actual sales results generated after the implementation of optimized sales strategies, such as the actual sales volume during the period following the implementation of the optimized sales strategies.

[0097] In this embodiment, after the optimized sales strategy (such as price reduction or adjustment of product display location) takes effect, the system automatically starts the corresponding monitoring cycle. During this monitoring cycle, the system continues to run the process from steps S10 to S30, that is, it continues to collect status change data and image data, and calculates and generates new behavioral characteristics, attention intensity, and actual sales data within the current monitoring cycle in real time. These newly calculated sales results are clearly distinguished and marked on the timeline from the historical sales results before the implementation of the optimized sales strategy, for subsequent effect evaluation.

[0098] Step S4420: Update the customer conversion rate based on the actual customer behavior data, and update the sales forecast result based on the sales performance.

[0099] It should be noted that updated customer conversion rate refers to the recalculated customer conversion rate based on the number of pickups and actual purchases generated within the monitoring period after the implementation of the optimized sales strategy. Updated sales forecast results refer to using the latest sales performance data (such as actual sales volume within the monitoring period after the implementation of the optimized sales strategy or within a pre-set time period) as new input and feeding it into the sales forecast model, thereby continuously updating the forecast value of future sales volume and incorporating the latest market feedback after the optimized sales strategy takes effect.

[0100] As a feasible implementation method, after the new monitoring period ends, the system automatically extracts the total number of pickups and the total number of actual purchases within that monitoring period from the newly calculated sales results data, and recalculates the customer conversion rate. Simultaneously, the system concatenates the latest actual sales volume's corresponding sales time-series sequence with the original historical sales data, inputting this updated sales time-series sequence into the sales forecasting model. The model is then refitted and re-predicted to obtain updated sales forecast results based on the latest market conditions, ensuring the timeliness of the evaluation criteria.

[0101] Step S4430: When the updated customer conversion rate does not reach the preset improvement threshold, or the gap between the updated sales forecast result and the target sales volume does not narrow to the preset range, the parameter iterative adjustment of the optimized sales strategy is triggered.

[0102] It's important to note that the preset improvement threshold is set to measure the effectiveness of the optimized sales strategy; it represents the minimum increase in customer conversion rate required relative to the initial state before the strategy was implemented. The preset range refers to the allowable interval between the sales forecast and the target sales volume; once the gap narrows to within this range, the sales target is considered achievable. Parameter iterative adjustment refers to automatically modifying sales strategy parameters (such as promotional discount rates, price adjustment ranges, buy-one-get-one-free promotions, and adjustments to display locations) based on the feedback from the current optimized sales strategy implementation. This generates a new, optimized sales strategy, which is then executed again, forming a closed-loop optimization cycle of "execution-monitoring-evaluation-adjustment."

[0103] In this embodiment, the system compares the updated customer conversion rate with the customer conversion rate before the implementation of the optimized sales strategy and calculates the percentage increase. If the percentage increase is lower than a preset threshold (e.g., a required increase of over 10% but only a 5% increase), the conversion improvement is deemed to have failed to meet expectations. Simultaneously, the system calculates the absolute or relative gap between the updated sales forecast and the target sales volume. If this gap does not fall within a preset target range (e.g., a required gap to be reduced to within 10% but still a 20% gap), the sales improvement is deemed to have failed to meet expectations. When any one or two conditions are met, the system triggers a sales strategy adjustment process. This adjustment process can be based on a reinforcement learning framework, using the current sales strategy parameters and performance evaluation (conversion rate, sales gap) as state inputs, and the direction of adjustment of the sales strategy parameters (e.g., increasing discounts, changing display locations) as actions. Narrowing the sales gap and increasing customer conversion rate serve as reward signals, training the model to output new sales strategy parameters. Alternatively, a more direct rule could be adopted. For example, if the customer conversion rate is not improved enough, the discount rate could be automatically increased by a certain percentage on the current basis to generate a new round of optimized sales strategy, and then await final review and manual decision-making by operations personnel or automatic execution.

[0104] Based on the above embodiments of this application, in the fourth embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 After step S40, the shelf management method further includes steps S50 to S70: Step S50: Based on the sales forecast results, determine the expected sales volume of the target product within a future preset time period.

[0105] It should be noted that the expected sales volume refers to a quantitative estimate of the number of target products that may be sold within a fixed period of time in the future (such as the next 24 hours, 3 days or a week) based on the current market conditions, customer behavior characteristics and historical sales trends. Its value comes from the sales forecast results generated in step S30 and is usually output in the form of average daily expected sales volume or total expected sales volume.

[0106] As a feasible implementation method, the system can accumulate the probability distribution or point estimates output by the sales forecasting model at a time granularity to obtain the total expected sales volume within a preset future time period. For example, if the sales forecasting model predicts an average daily expected sales volume of 40 units for the next three days, then the total expected sales volume is 120 units.

[0107] Step S60: In response to the promotional plan or price adjustment instruction included in the optimized sales strategy, the expected sales volume is adjusted by a coefficient to generate the final demand forecast result for the target product.

[0108] It should be noted that the demand forecast results are revised forecasts based on the original expected sales volume, taking into account the potential incremental effects of the upcoming optimized sales strategy. These revised forecasts are intended to more accurately reflect the actual demand level after the optimized sales strategy is implemented.

[0109] As a feasible implementation method, the system can pre-establish a mapping relationship library between promotion types and sales growth coefficients. For example, "limited-time 20% off" corresponds to a coefficient of 1.3, "buy one get one free" corresponds to a coefficient of 1.8, and "adjust display position to prime shelf" corresponds to a coefficient of 1.15. When the optimized sales strategy includes multiple instructions, a comprehensive correction coefficient can be calculated using a product or weighted summation method. In addition, correction coefficients can be set separately according to product categories, or the coefficients can be personalized based on historical promotional results, avoiding the use of a uniform amplification ratio for all products.

[0110] As an example, if the original expected sales volume of a product is 100 units as output by the sales forecasting model, and if the optimized sales strategy includes "price reduction of 10%" and "increase in trial products", which correspond to coefficients of 1.25 and 1.1 respectively, then the final demand forecast result is 100 × 1.25 × 1.1 = 137.5 units, which is rounded up to 138 units.

[0111] Step S70: Based on the demand forecast results and the current inventory status of the target product, generate a replenishment plan for the target product.

[0112] It should be noted that the current inventory status refers to the real-time available inventory of the target product in the current store or warehouse that can be immediately used to replenish the shelves; the replenishment plan refers to the quantity of products that are recommended to be replenished from the real-time available inventory to the shelves in order to meet demand forecasts and maintain normal shelf display.

[0113] In this embodiment, the system first calculates the total demand forecast for a preset future time period (if there are forecasts for multiple days, the sum of the average daily expected sales for each day in the forecast results), and simultaneously obtains the current remaining quantity of the target product on the shelves. Then, it calculates the theoretical replenishment demand based on the current inventory status. The calculation formula is: Theoretical replenishment demand = Total demand forecast + Safety stock - Current shelf quantity. Subsequently, the system compares the theoretical replenishment demand with the real-time available inventory, and takes the smaller of the two as the final actual replenishment quantity: Actual replenishment quantity = Min(Theoretical replenishment demand, Real-time available inventory). If the actual replenishment quantity is greater than zero, a corresponding replenishment instruction is generated. Simultaneously, if the theoretical replenishment demand is greater than the current real-time available inventory, the system generates an inventory warning, indicating insufficient available inventory.

[0114] In addition, when multiple products need to be replenished at the same time, the system can prioritize the replenishment plan based on the importance of the products, the level of stockout risk, or the expected turnover rate to optimize the replenishment order.

[0115] As an example, a certain product shelf currently has 20 units, the total demand forecast for the next three days is 300 units, the safety stock is 10 units, and the current real-time available inventory is 130 units. First, the theoretical replenishment demand is calculated as 300 + 10 - 20 = 290 units. Then, the actual replenishment quantity is determined to be Min(290, 130) = 130 units. Since the actual replenishment quantity is greater than zero, the system generates a replenishment plan to "replenish the shelf with 130 units". Simultaneously, because the theoretical replenishment demand (290 units) is greater than the current real-time available inventory (130 units), the system generates an inventory warning, indicating a shortage of 160 units, to remind operations personnel to replenish stock promptly.

[0116] Optionally, based on the generated replenishment plan, if an inventory alert is generated, the system can further coordinate with the backend supply chain system to convert the replenishment plan into a specific ordering plan. In addition to the order quantity, the ordering plan will also consider the supplier's preparation cycle, logistics and transportation time, and the store's sales rhythm to calculate the suggested order placement time, product category, order quantity, and estimated arrival time. This ensures that the replenished inventory is available in time before new replenishment needs arise, thereby maintaining a reasonable inventory level and avoiding stockouts or inventory backlogs.

[0117] This application provides a shelf management device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the shelf management method in the above embodiment 1.

[0118] The following is for reference. Figure 5 The diagram illustrates a structural schematic of a shelf management device suitable for implementing embodiments of this application. The shelf management device in the embodiments of this application may include various hardware and software components for implementing shelf management methods. Figure 5 The shelf management equipment shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of this application.

[0119] like Figure 5 As shown, the shelving management device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the shelving management device. The processing unit 1001, the ROM 1002, and the RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the shelving management equipment to communicate wirelessly or wiredly with other devices to exchange data. While the figures show shelving management equipment with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0120] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0121] The shelving management equipment provided in this application, employing the shelving management method described in the above embodiments, can solve the technical problem of inaccurate assessment of actual demand for goods by the shelving management system. Compared with the prior art, the beneficial effects of the shelving management equipment provided in this application are the same as those of the shelving management method provided in the above embodiments, and other technical features of the shelving management equipment are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0122] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0124] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the shelf management method in the above embodiments.

[0125] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0126] The aforementioned computer-readable storage medium may be included in the shelving management equipment; or it may exist independently and not be assembled into the shelving management equipment.

[0127] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the shelf management device, cause the shelf management device to: collect state change data of target goods on the shelf through shelf sensors, and acquire image data associated with the shelf; wherein the state change data includes weight changes and displacement changes caused by the target goods leaving or returning to the shelf; determine customer behavior characteristics and attention levels for the target goods based on the state change data and the image data; predict the sales trend of the target goods within a preset time period in the future based on the historical sales data and current sales strategy of the target goods, and obtain a sales forecast result; and generate an optimized sales strategy for the target goods based on the sales forecast result, the current sales strategy, the behavior characteristics, and the attention levels.

[0128] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0129] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0130] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0131] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described shelf management method, thereby solving the technical problem of inaccurate assessment of actual demand for goods in shelf management systems. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the shelf management method provided in the above embodiments, and will not be repeated here.

[0132] This application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the shelf management method described above.

[0133] The computer program product provided in this application can solve the technical problem of inaccurate assessment of actual demand for goods in shelf management systems. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the shelf management method provided in the above embodiments, and will not be repeated here.

[0134] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

[0135] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system 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 system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0136] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0137] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A shelf management method characterized by, The shelf management method includes: The system collects status change data of target products on the shelf using shelf sensors, and also acquires image data associated with the shelf; wherein, the status change data includes weight changes and displacement changes caused by the target products leaving or returning to the shelf; Based on the state change data and the image data, determine the customer's behavioral characteristics and level of interest in the target product; Based on the historical sales data and current sales strategy of the target product, predict the sales trend of the target product in a future preset time period to obtain the sales forecast result; Based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of attention, an optimized sales strategy is generated for the target product.

2. The shelf management method according to claim 1, wherein Before the step of predicting the sales trend of the target product within a preset future time period based on the historical sales data and current sales strategy of the target product, and obtaining the sales forecast result, the shelf management method further includes: Based on the status change data collected by the shelf sensors, the sales quantity, sales time and corresponding sales price of the target product in multiple historical periods are determined, and the historical sales data of the target product is obtained.

3. The shelf management method according to claim 1, wherein The step of determining the customer's behavioral characteristics and level of interest in the target product based on the state change data and the image data includes: The image data is subjected to target detection to identify customers lingering in front of the shelf and their interactive actions with the target products. The interaction action and the state change data are time-aligned and correlated to determine whether the interaction action results in the target product being taken out or put back. The behavioral characteristics are generated based on at least one of the following: the customer's dwell time in the area where the target product is located, the duration of visual attention, the number of times the product is picked up, the number of times the product is put back, and the tactile behavior that does not cause a change in state. Based on the statistical distribution and changing trends of the behavioral characteristics over multiple time periods, the customer's level of attention to the target product is determined.

4. The shelf management method according to claim 1, wherein The step of predicting the sales trend of the target product within a preset future time period based on the historical sales data and current sales strategy of the target product, and obtaining the sales forecast result, includes: The historical sales data and the current sales strategy are used as input features and fed into a pre-trained sales prediction model. Based on the sales forecasting model, the expected sales volume of the target product within a preset future time period is output as the sales forecast result.

5. The shelf management method according to claim 1, wherein The step of generating an optimized sales strategy for the target product based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of public interest includes: Based on the behavioral characteristics, the number of times the target product was taken and the number of times it was put back are obtained, as well as the actual number of times the target product was purchased. The actual number of purchases is the number of events in which the product was taken without being put back. Calculate the customer conversion rate of the target product based on the number of times it was taken and the number of times it was actually purchased; Based on the level of public interest, the target sales volume for the target product is determined. When the sales forecast result is lower than the target sales volume and the customer conversion rate is lower than the preset conversion threshold, the optimized sales strategy is generated based on the current sales strategy. The optimized sales strategy includes adjusting the price, promotion plan and / or display position of the target product.

6. The shelf management method according to claim 5, wherein The step of determining the target sales volume corresponding to the target product based on the level of public interest includes: Obtain the historical popularity and corresponding actual sales volume of multiple reference products belonging to the same category as the target product; Based on the historical attention and the corresponding actual sales volume, establish a mapping relationship between the attention and the target sales volume; Based on the mapping relationship and the current popularity of the target product, the target sales volume is obtained.

7. The shelf management method as described in claim 5, characterized in that, When the sales forecast result is lower than the target sales volume and the customer conversion rate is lower than a preset conversion threshold, the optimized sales strategy is generated based on the current sales strategy. The optimized sales strategy includes adjusting the price, promotional plan, and / or display location of the target product. Following this step, the shelf management method further includes: Monitor the actual customer behavior data and sales performance of the target product after the implementation of the optimized sales strategy; The customer conversion rate is updated based on the actual customer behavior data, and the sales forecast result is updated based on the sales performance. When the updated customer conversion rate fails to reach the preset improvement threshold, or when the gap between the updated sales forecast and the target sales volume fails to narrow to the preset range, the parameters of the optimized sales strategy are iteratively adjusted.

8. The shelf management method as described in claim 1, characterized in that, After the step of generating an optimized sales strategy for the target product based on the sales forecast results, the current sales strategy, the behavioral characteristics, and the level of attention, the shelf management method further includes: Based on the sales forecast results, determine the expected sales volume of the target product within a preset future time period; In response to the promotional plan or price adjustment instruction included in the optimized sales strategy, the expected sales volume is adjusted by a coefficient to generate the final demand forecast result for the target product; Based on the demand forecast results and the current inventory status of the target product, a replenishment plan for the target product is generated.

9. A shelf management device, characterized in that, The shelf management device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the shelf management method as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the shelf management method as described in any one of claims 1 to 8.

Images