Medical vending platform mobile management method, device and equipment and storage medium

Abstract

The invention relates to the technical field of medical vending platforms, and discloses a medical vending platform mobile management method, device and equipment and a storage medium, and the method comprises the steps: creating an initial commodity shelving scheme based on the operation data of a vending machine of a medical vending platform, and collecting the image data of medical commodities; carrying out residual learning and two-stage feature extraction to obtain medical commodity feature data; inputting a medical commodity identification model to carry out medical commodity identification to obtain a medical commodity identification result; carrying out inventory detection, generating commodity inventory data, and sending the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; inputting the RFID tag information and the replenishment task instruction into a replenishment execution unit, generating replenishment operation data, and updating real-time inventory information; the initial commodity shelving scheme is adjusted based on the real-time inventory information and the historical sales data, a target commodity shelving scheme is generated, and the method improves the operation efficiency and the service quality of a medical sales platform.

Description

Technical Field

[0001] The present invention relates to the technical field of medical vending platforms, and in particular to a mobile management method, device, equipment and storage medium for a medical vending platform. Background Art

[0002] Self-service vending services for medical supplies have become an important way to meet the public's demand for convenient purchase of medical supplies. Traditional medical vending platforms mainly rely on manual management, and have problems such as low efficiency and poor real-time performance in terms of product replenishment, inventory monitoring and sales analysis, making it difficult to meet users' needs for timely purchase of medical supplies.

[0003] The existing mobile management system of medical vending platforms generally uses a single RFID technology or barcode recognition technology for commodity management. This method is prone to recognition errors when there are many types of commodities and complex placements, and cannot effectively deal with abnormal situations such as misplacement and occlusion of commodities, resulting in inaccurate inventory information and affecting the scientific nature of replenishment decisions. In addition, the current medical vending platform lacks an intelligent commodity shelf optimization mechanism, and cannot dynamically adjust the commodity placement strategy according to the sales characteristics of different time periods and regions, nor can it realize the intelligent allocation and path optimization of replenishment tasks, resulting in a waste of human resources and reducing the platform's operating efficiency. Summary of the invention

[0004] The present invention provides a mobile management method, device, equipment and storage medium for a medical vending platform, and the present invention improves the operation efficiency and service quality of the medical vending platform.

[0005] In a first aspect, the present invention provides a mobile management method for a medical vending platform, the mobile management method for a medical vending platform comprising: Create an initial product listing plan based on the vending machine operation data of the medical vending platform and collect medical product image data; Performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data; Inputting the medical commodity feature data into a medical commodity identification model to perform medical commodity identification to obtain a medical commodity identification result; Performing inventory detection according to the medical commodity identification result to generate commodity inventory data, and sending the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; Inputting the RFID tag information and the replenishment task instruction into the replenishment execution unit, generating replenishment operation data, and updating the real-time inventory information; The initial product listing plan is adjusted based on the real-time inventory information and historical sales data to generate a target product listing plan.

[0006] In a second aspect, the present invention provides a mobile management device for a medical vending platform, the mobile management device for a medical vending platform comprising: The acquisition module is used to create an initial product listing plan based on the vending machine operation data of the medical vending platform and to collect medical product image data; An extraction module, used for performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data; an identification module, used for inputting the medical commodity feature data into a medical commodity identification model to perform medical commodity identification and obtain a medical commodity identification result; A detection module, used to perform inventory detection according to the medical commodity identification result, generate commodity inventory data, and send the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; A generation module, used to input the RFID tag information and the replenishment task instruction into the replenishment execution unit, generate replenishment operation data, and update the real-time inventory information; The adjustment module is used to adjust the initial product listing plan based on the real-time inventory information and historical sales data to generate a target product listing plan.

[0007] The third aspect of the present invention provides a medical vending platform mobile management device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the medical vending platform mobile management device executes the above-mentioned medical vending platform mobile management method.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned medical vending platform mobile management method.

[0009] In the technical solution provided by the present invention, accurate identification of medical products and real-time inventory monitoring are achieved through deep learning and computer vision technology in conjunction with an RFID identification system; a two-stage feature extractor and a residual learning method are used to effectively improve the accuracy of product identification in complex environments; by establishing a dynamic product shelving optimization mechanism, intelligent adjustment of product layout and automatic optimization of replenishment strategies are achieved; a mobile management terminal and a replenishment execution unit are introduced to build a complete replenishment-acceptance-shelf closed-loop management system, which significantly improves the operating efficiency and service quality of the medical vending platform; through intelligent analysis of real-time inventory information and sales data, an adaptive optimization mechanism for sales models is established, which achieves reasonable allocation of product resources and effective control of operating costs.

[0010] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0011] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of an embodiment of a mobile management method for a medical vending platform in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of a mobile management device for a medical vending platform in an embodiment of the present invention; Figure 3 Schematic diagram of an embodiment of a mobile management device for a medical vending platform in an embodiment of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0015] To facilitate understanding of this embodiment, a mobile management method for a medical vending platform disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Create an initial product listing plan based on the vending machine operation data of the medical vending platform and collect medical product image data; It is understandable that the execution subject of the present invention may be a medical vending platform mobile management device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0016] Specifically, data cleaning is performed on the commodity sales records of each vending machine in the medical vending platform to remove invalid data, erroneous data and outliers, ensuring that the obtained sales records can reflect the actual sales situation. After data cleaning, valid sales data is obtained, and the valid sales data is classified according to commodity category and sales time to form classified sales data. The sales situation is concretized to specific commodity categories and sales time periods. Time series analysis is performed on the classified sales data to determine the sales trend and fluctuation of commodities in each time period, and the sales curve of each time period is obtained. By calculating the sales curve, the sales frequency of commodities in each time period is obtained, which reflects the sales activity of commodities in a specific time period, and the sales time period parameters are obtained. The sales time period parameters are input into the commodity turnover rate calculation model to calculate the turnover speed of various commodities. The turnover speed of commodities reflects the sales cycle of commodities, that is, the speed from the listing of commodities to the sale of commodities. Through the commodity turnover speed data, the safety inventory of various commodities is set accordingly to ensure that each commodity has sufficient inventory in the vending machine to cope with demand fluctuations, and the replenishment threshold parameters are obtained. Spatial distribution analysis is performed on the commodity placement position data of each vending machine to obtain the heat data of the commodity location. The heat data of the commodity location mainly reflects the sales frequency of the commodity at a certain location and the interaction between the customer and the location. According to the heat data of the commodity location, the commodity placement is prioritized to obtain the location priority data, and the commodities with high sales potential are placed in a more eye-catching location that is convenient for customers to use. The location priority data is analyzed for correlation with the sales time period parameters to obtain the matching data of the location and time period. The matching data of the location and time period can reveal which commodities can be placed in which location to maximize their sales in a specific time period. Based on the matching data, the commodity location distribution matrix is ​​generated to obtain the commodity layout parameters, and the optimal placement method of each commodity in the vending machine is provided, which helps to improve the exposure rate and sales efficiency of the commodity. The initial commodity shelf plan is generated according to the replenishment threshold parameters, commodity layout parameters and sales time period parameters to maximize the sales potential of the commodities in the vending machine and meet the needs of customers in different time periods. At the same time, based on the initial commodity shelf plan, the remaining quantity of each commodity in the vending machine is monitored in real time to timely grasp the out-of-stock risk of the commodity. By monitoring the remaining quantity of the commodity, the out-of-stock risk value of each commodity is calculated, and then the priority of image acquisition is determined according to the out-of-stock risk value. For products with a higher risk of out-of-stock, image acquisition is prioritized to ensure that changes in inventory and fluctuations in demand can be identified in a timely manner. Based on the product layout parameters, the display of products in the vending machine is spatially mapped to obtain the specific coordinate data of the products. Using these coordinate data, the acquisition angle of the image acquisition device is set to ensure that the full view and details of each product can be captured. The image acquisition device is controlled according to the priority of image acquisition, and acquisition is performed at an adaptive angle to ensure the integrity and accuracy of the image data. Image data of medical products is obtained through adaptive acquisition.

[0017] 102. Perform residual learning and two-level feature extraction on the medical product image data to obtain medical product feature data; Specifically, the brightness of medical product image data is analyzed to obtain the image brightness distribution matrix, which describes the brightness distribution of the image in each area. Based on the image brightness distribution matrix, the image data is divided into multiple blocks to form an image block sequence. In the residual learning process, the features of different parts of the image are better extracted in a targeted manner to improve the adaptability and accuracy of the model to different brightness areas. The image block sequence is input into the backbone feature extraction layer of the residual network, which consists of three residual units. Each residual unit contains two 3×3 convolutional layers and a short-circuit connection. The short-circuit connection design effectively alleviates the gradient vanishing problem in the deep network and enhances the training effect of the network. The ReLU activation function is used in each convolutional layer to increase the nonlinear expression ability of the network, and the batch normalization process is used to improve the stability and convergence speed of the training. After being processed by these three residual units, the backbone feature map is obtained, which contains the basic feature information of medical products. The backbone feature map is calculated by residual mapping, and the features are further adjusted to capture more subtle feature changes in the image to obtain a residual feature map. The residual feature map is element-wise added to the original image block sequence to obtain enhanced image data. The enhanced image data is input to the first-level feature extraction network of the two-level feature extractor. The feature extraction network adopts a pyramid structure consisting of five convolutional layers. The maximum pooling operation is used after each convolution layer. This design can extract the features of the product at different scales and obtain a multi-scale feature map. The multi-scale feature map contains the feature information of the product at different spatial resolutions. Channel attention calculation is performed on the multi-scale feature map to obtain a channel weight matrix. By assigning different weights to different channels, important feature channels are highlighted, and irrelevant or noisy channels are suppressed. According to the channel weight matrix, the multi-scale feature map is weighted to obtain the first-level feature data. The first-level feature data is input to the second-level feature extraction network of the two-level feature extractor. The second-level feature extraction network adopts four parallel dilated convolution branches, and the dilated rates of these convolution branches are set to 1, 2, 4, and 8 respectively. The use of dilated convolution enables the network to increase the receptive field with fewer parameters, capture feature information in a larger range, and obtain multi-receptive field feature maps. In order to improve the effectiveness of the multi-receptive field feature map, a spatial attention calculation is performed on it to obtain a spatial weight matrix. The spatial attention calculation aims to highlight the spatial areas in the image that are most important for product identification and suppress insignificant areas. According to the spatial weight matrix, the multi-receptive field feature map is weighted fused to obtain the second-level feature data. The second-level feature data is processed by the feature integration module. The feature integration module contains two fully connected layers and one Softmax layer. The function of the two fully connected layers is to further integrate and transform the extracted features to make the features more compact and discriminative. The Softmax layer is used to normalize the features and convert them into probability distributions to obtain the final feature data of medical products.

[0018] 103. Inputting the medical commodity feature data into the medical commodity identification model to perform medical commodity identification and obtain a medical commodity identification result; Specifically, the feature data of medical products are input into the product recognition backbone network for feature extraction and recognition. The product recognition backbone network adopts the DenseNet structure. DenseNet (densely connected network) has high feature reusability. By implementing dense jump connections in the network, features are fully shared between layers, improving the efficiency of feature extraction and the training effect of the network. The network contains four densely connected blocks, each of which has six convolutional layers. The connection between the convolutional layers adopts dense jump connections. The output of each convolution layer will be passed to all subsequent layers to form a feature sharing mechanism, and a dense map of product features is obtained. Multi-scale feature fusion operations are performed on the dense map of product features to unify features of different scales into the same dimension for subsequent recognition. The number of channels of feature maps of different scales is unified to 256 through a 1×1 convolution layer, which effectively reduces the computational complexity while maintaining the integrity of the features. The spatial dimensions of each feature map are adjusted to the same size by bilinear interpolation to obtain product feature fusion data. The commodity feature fusion data is input into the type recognition branch network for commodity type recognition. The type recognition branch network contains three parallel attention modules, each of which is composed of a channel attention unit and a spatial attention unit in series. The channel attention unit is used to assign weights to different channels in the feature map, highlight important channels and suppress unimportant feature information; while the spatial attention unit focuses on the features in the spatial position to strengthen the important regional features related to the commodity category. After being processed by these three parallel attention modules, the commodity type feature vector is obtained. The commodity type feature vector is input into the type classifier for classification and recognition. The type classifier contains two fully connected layers and one Softmax layer. The output dimension of the first fully connected layer is set to 512, which is used to reduce and integrate the commodity type features, and the output dimension of the second fully connected layer is the same as the number of categories of medical commodities, which is used to map the features to the specific commodity category space. The Softmax layer converts the output result into the probability distribution of each category, and finally obtains the commodity type information. While recognizing the commodity type, the location of the commodity is detected. The commodity feature fusion data is input into the location detection branch network for location recognition. The location detection branch network adopts the feature pyramid network (FPN) structure. The FPN structure realizes the effective fusion of multi-scale features through top-down feature pyramid and lateral connection to obtain a multi-level location feature map. The region proposal calculation is performed on the multi-level location feature map, and the candidate region features are generated by three anchor box generators of different scales. Each anchor box generator is used to generate candidate regions of different sizes and proportions to ensure that the product can be effectively identified in the image regardless of its size. The non-maximum suppression operation is used to eliminate duplicate candidate regions and retain the most representative candidate region features. The dimension of each candidate region feature is set to 256.The candidate region features are input into the position regressor, which contains four fully connected layers. Each layer uses the ReLU activation function to increase the nonlinear expression ability of the network and better fit the position features of the product. The position regressor outputs four coordinate values, which represent the bounding box position of the product in the image, that is, the specific location information of the product. The type information of the product and the product location information obtained by the position detection are label aligned. Through the label alignment module, the type of the product is associated and matched with its location information to obtain a complete medical product recognition result.

[0019] 104. Perform inventory detection based on the medical product identification result, generate product inventory data, and send the product inventory data to the mobile management terminal to generate a replenishment task instruction; Specifically, the commodity type information and commodity location information in the medical commodity identification result are input into the vending machine inventory detection unit, and the commodity type information is classified and statistically processed to generate a commodity category quantity table to display the specific quantity of each commodity in the vending machine. At the same time, according to the commodity location information, a commodity aisle distribution map is generated to identify the specific location of each commodity in the vending machine. The commodity category quantity table is compared with the commodity layout parameters in the initial commodity shelf plan. By comparing the difference between the current commodity inventory and the original plan, it is identified which commodities are missing in the vending machine, and commodity missing data is generated. Based on the commodity missing data, a replenishment demand matrix is ​​established to describe the replenishment demand for each missing commodity. The commodity aisle distribution map and the commodity layout parameters are spatially mapped and analyzed, and by comparing the actual commodity placement position and the initial planned position, the deviation data of the commodity position is obtained, indicating the deviation between the commodity placement position in the vending machine and the planned plan. Based on these deviation data, aisle adjustment suggestions are generated to optimize the placement of commodities in the vending machine, make the commodities more in line with the original plan, improve the customer's accessibility to the commodities and the overall operating efficiency of the vending machine. Priority sorting is performed on the replenishment demand matrix, and which products need to be replenished first is determined according to the urgency of replenishment. In order to determine the urgency of replenishment, the replenishment level threshold is set based on the replenishment threshold parameter to obtain graded replenishment demand data, and the replenishment demand of different products is graded according to the urgency. The graded replenishment demand data and the channel adjustment suggestions are integrated and processed to obtain a comprehensive commodity inventory data. In order to ensure the security of commodity inventory data during transmission, the commodity inventory data is encrypted through the data encryption module to obtain encrypted inventory data. The encrypted inventory data is sent to the mobile management terminal through the wireless communication module. After receiving the data, the mobile management terminal performs identity authentication to ensure that only authorized personnel can access the inventory data. After identity authentication, the encrypted inventory data is decrypted to obtain the terminal inventory data. According to the terminal inventory data and the sales period parameters, a time-sharing replenishment plan is generated. According to the sales situation in different time periods, a reasonable replenishment time schedule is formulated to ensure that replenishment can be completed in time before the sales peak, so as to maximize customer satisfaction. Based on the time-sharing replenishment plan, the replenishment path is optimized so that the shortest and most effective replenishment route can be selected when executing the replenishment task, thereby improving the replenishment efficiency and reducing the operating cost. The replenishment path data is input into the replenishment scheduling module, which performs task allocation calculations based on the current staffing information. The replenishment tasks are reasonably allocated to the available staff to ensure that the replenishment tasks can be completed efficiently. Through the above steps, the replenishment task instructions are generated, and the replenishment task content, replenishment time and replenishment route of each staff member are clarified, so that the entire replenishment process can be automated and efficient, ensuring that the goods in the vending machine are always in sufficient state to meet medical needs.

[0020] 105. Input the RFID tag information and the replenishment task instruction into the replenishment execution unit, generate replenishment operation data, and update the real-time inventory information; Specifically, the replenishment path data and the graded replenishment demand data in the replenishment task instruction are input into the replenishment execution unit, and the replenishment execution unit performs task planning for each replenishment site according to the replenishment path data to obtain the replenishment execution task sequence. The RFID reader is configured based on the replenishment execution task sequence, and the RFID tag information of the replenishment goods is scanned by the RFID reader to obtain the tag data of the replenishment goods. The tag data of the replenishment goods is matched and verified with the commodity category quantity table to ensure that the actual replenished goods are consistent with the planned goods, and the commodity verification results are obtained. After the commodity verification is completed, the replenishment execution status data is generated according to the verification results. The replenishment execution status data records the execution status of each replenishment task, including whether the replenishment is successfully completed and whether the goods meet the requirements. In order to optimize the replenishment process, the replenishment execution status data and the channel adjustment suggestions are processed in coordination to obtain the channel layout guidance data, and the placement position of each commodity in the vending machine is clarified to ensure that the goods can be displayed according to the optimal layout. In the process of commodity replenishment, the replenishment progress is monitored in real time. By monitoring the product verification results and replenishment execution status data, abnormal situations in the replenishment process are discovered in time, and replenishment abnormal signals are generated. Replenishment abnormal signals include problems such as incorrect product quantity and tag reading failure. In order to process abnormal signals, they are compared with replenishment threshold parameters to generate replenishment warning data. Replenishment warning data effectively reminds managers that there are problems with certain replenishment operations and need to be adjusted in time to avoid affecting inventory status. The generated replenishment warning data is sent to the mobile management terminal through the wireless communication module. After receiving these warning data, the mobile management terminal will classify them and obtain the replenishment adjustment plan. Different response strategies are formulated according to the severity of the warning. For more urgent problems, the system will give priority to solving them, while for minor deviations, they will be corrected at the next replenishment. The replenishment adjustment plan can ensure that the replenishment task can be adjusted in a timely and reasonable manner when problems occur, ensuring the smooth progress of the replenishment process. Replenishment operation data is generated based on the replenishment adjustment plan and the cargo channel layout guidance data, recording each step of replenishment as well as the placement and adjustment plan of the goods. The replenishment operation data is written into the database of the medical vending platform to generate a replenishment history record. Data fusion processing is performed on replenishment history records and product layout parameters to update the real-time inventory information of the medical vending platform. Through the fusion process, the vending platform can adjust its inventory data according to the latest replenishment situation to ensure the accuracy and timeliness of inventory information. The updated real-time inventory information is fed back to the vending machine inventory detection unit, which monitors and manages the status of the goods in the vending machine accordingly.

[0021] 106. Adjust the initial product listing plan based on real-time inventory information and historical sales data to generate a target product listing plan.

[0022] Specifically, the real-time inventory information and replenishment history records are processed according to the commodity category quantity table to obtain commodity circulation data, which reflects the changes in the entry and exit of various commodities in the vending machine. The commodity circulation data is analyzed in the time dimension to obtain the sales of commodities in different periods and generate commodity sales samples. According to the commodity sales samples and the commodity position deviation data, a sales position association matrix is ​​constructed. The sales position association matrix is ​​used to describe the relationship between the sales volume of commodities and their positions in the vending machine, and effectively identify which positions have a greater impact on commodity sales. Then, a cluster analysis is performed on the sales position association matrix to obtain commodity area heat data. The commodity area heat data reveals the sales activity of different areas in the vending machine, that is, which areas have more concentrated commodity sales in certain time periods. The commodity area heat data is cross-validated with the graded replenishment demand data to obtain the commodity dynamic sales rate data, which reflects the sales speed and sales frequency of the commodity, so as to determine whether certain commodities are unsalable or hot-selling in a specific time period. Based on the dynamic sales rate data, a commodity partition configuration plan is generated. Optimize the display position of goods according to the sales situation, for example, place goods with higher sales rate in a position where customers can more easily access them, so as to improve the overall sales efficiency. Perform spatial mapping operation on the goods partition configuration plan and the aisle layout guidance data to obtain the optimized layout data. The optimized layout data includes the optimal placement of goods in the vending machine to maximize the display effect of the goods and increase sales opportunities. Update the layout parameters of the goods according to the optimized layout data. At the same time, dynamically update the replenishment threshold parameters based on the goods sales rate data to generate updated replenishment threshold parameters. The replenishment threshold parameters determine the safety inventory of each product in the vending machine. Dynamically updating the replenishment threshold parameters ensures that the inventory of goods in the vending machine can be replenished in time when sales fluctuate greatly, and there will be no shortage or backlog. The updated replenishment threshold parameters are input into the replenishment warning module, and the replenishment warning module issues an alarm in time based on the new threshold so that the management personnel can adjust the replenishment tasks in time. Perform time period association analysis based on the time-sharing replenishment plan and the replenishment execution status data to obtain time period characteristic data. The time period characteristic data describes the sales characteristics of goods in different time periods, for example, the sales volume of certain goods increases or decreases significantly in certain time periods. Based on the time period characteristic data, the sales period parameters are updated, and the new sales period parameters more accurately reflect the current sales rules and trends. The updated product layout parameters, replenishment threshold parameters, and sales period parameters are subjected to pattern fusion operations to obtain the sales characteristics of the goods, including the layout of the goods in the vending machine, replenishment strategy, and sales rules. Based on the sales characteristics of the goods, the parameters of the initial product listing plan are updated to obtain the final target product listing plan.The target product listing plan is formed after multiple optimizations and adjustments based on the initial plan, combined with real-time inventory information, historical sales data and product sales characteristics, to maximize the sales efficiency of the vending machine and meet the purchasing needs of customers. The target product listing plan is written into the operation database of the medical vending platform to ensure that the vending platform can operate according to the optimal product display and replenishment strategy in the subsequent operation.

[0023] In the embodiment of the present invention, through deep learning and computer vision technology, in conjunction with the RFID identification system, accurate identification and real-time inventory monitoring of medical products are achieved; a two-stage feature extractor and a residual learning method are adopted to effectively improve the accuracy of product identification in complex environments; by establishing a dynamic product shelving optimization mechanism, intelligent adjustment of product layout and automatic optimization of replenishment strategy are achieved; by introducing mobile management terminals and replenishment execution units, a complete replenishment-acceptance-shelf closed-loop management system is constructed, which significantly improves the operating efficiency and service quality of the medical vending platform; through intelligent analysis of real-time inventory information and sales data, an adaptive optimization mechanism for sales models is established, which realizes the rational allocation of product resources and effective control of operating costs.

[0024] In a specific embodiment, the process of executing step 101 may specifically include the following steps: Clean the sales records of the products of each vending machine in the medical vending platform to obtain valid sales data, and classify the valid sales data according to product category and sales time to obtain classified sales data; Perform time series analysis on classified sales data to obtain sales curves for each period, calculate the sales frequency of goods in each period based on the sales curves, and obtain sales period parameters; Input the sales period parameters into the commodity turnover rate calculation model to obtain commodity turnover speed data, set the safety inventory of each commodity according to the commodity turnover speed data, and obtain the replenishment threshold parameters; Perform spatial distribution analysis on the product placement location data of each vending machine to obtain product location heat data, and prioritize the product placement locations according to the product location heat data to obtain location priority data; Perform correlation analysis on the location priority data and the sales period parameters to obtain the matching data of the location and the period, generate the commodity location distribution matrix based on the matching data, and obtain the commodity layout parameters; Generate an initial product placement plan based on replenishment threshold parameters, product layout parameters, and sales period parameters, and monitor the remaining quantity of each product in the vending machine based on the initial product placement plan to obtain a product out-of-stock risk value, and determine the priority of image acquisition based on the product out-of-stock risk value; The merchandise display in the vending machine is spatially mapped according to the merchandise layout parameters to obtain merchandise coordinate data, and the acquisition angle of the image acquisition device is set based on the merchandise coordinate data. According to the image acquisition angle and the priority of image acquisition, the image acquisition device is controlled to perform adaptive angle acquisition to obtain medical merchandise image data.

[0025] Specifically, incomplete, unreasonable or duplicate data is removed from the sales records of the vending machine, and the data obtained can truly reflect the actual sales situation. For example, some repeated transactions recorded due to vending machine failures and erroneous records caused by network delays need to be eliminated. After this process, valid sales data is obtained. The valid sales data is classified according to product category and sales time to obtain classified sales data. Data classification helps to understand the sales patterns of different categories of goods in different time periods. Time series analysis is performed on the classified sales data. By modeling the sales situation in different time periods, the sales curve of each product in each time period is obtained. Assume that the function of the sales volume of the product changing over time is ,in Indicates Category of products in time Analyze the sales curve and calculate the sales frequency of the product in each period, recorded as , the formula is: ; in, Indicates The average sales volume of the same product in the entire time period. is the total length of the time period. By calculating the sales frequency , we can get the sales activity of the goods in each period, and then get the sales period parameters, which reflect the demand fluctuation of the goods in a specific period of time. The sales period parameters are input into the commodity turnover rate calculation model. The commodity turnover rate is used to measure the speed from the listing to the sale of the goods. The calculation formula of commodity turnover rate is: ; in, Indicates The turnover rate of the commodity category, Indicates the total sales volume of the product in a specific time period. Indicates the average inventory of goods. By calculating the turnover rate of goods , and obtain the turnover speed data of the goods. According to the turnover speed, set the safety inventory of each type of goods, recorded as , and its calculation formula is: ; in, Indicates the replenishment cycle. Safety stock Ensure that each product has sufficient inventory during the replenishment cycle to cope with demand fluctuations and obtain the replenishment threshold parameter. In order to optimize the layout of the vending machines, the spatial distribution analysis of the product placement data of each vending machine is performed to obtain the product location heat data. The location heat data reflects the popularity of different locations in the vending machine and is calculated by recording the location frequency of customers purchasing products. Assume that the heat of a certain location is , the formula is: ; in, Indicates at time Time vending machine The number of purchases at a location. , prioritize the positions of the goods in the vending machine and obtain the position priority data. Perform correlation analysis on the position priority data and the sales period parameters to calculate the matching degree of different goods in different positions and time periods, and generate the matching degree data of the position and time period. The matching degree is measured by the correlation coefficient. Assuming that a certain product In Location The matching degree at , and its calculation formula is: ; in, and Respectively indicate products Average sales frequency and location The average popularity of the product. Through these matching data, the product location distribution matrix is ​​generated to obtain the product layout parameters. Based on the replenishment threshold parameters, product layout parameters and sales period parameters, the initial product shelf plan is generated to describe the best placement position and required inventory for each product in the vending machine. Based on the initial product shelf plan, the remaining quantity of each product in the vending machine is monitored to obtain the out-of-stock risk value of the product. Out-of-stock risk value

[0026] ; in, Indicates the current inventory. , determine the products that need to be imaged first so that they can be replenished in time. According to the product layout parameters, the display of the products in the vending machine is spatially mapped to obtain the coordinate data of each product. Assuming that the product The position coordinates in the vending machine are These coordinate data help determine the acquisition angle of the image acquisition device The acquisition angle is calculated based on the location of the product and the installation position of the camera. The formula is: ; in, is the coordinate of the camera. Based on the image acquisition angle and the priority of image acquisition, the image acquisition device is controlled to perform adaptive angle acquisition to obtain image data of the medical product.

[0027] In a specific embodiment, the process of executing step 102 may specifically include the following steps: Performing brightness analysis on the medical product image data to obtain an image brightness distribution matrix, and performing block processing on the medical product image data according to the image brightness distribution matrix to obtain an image block sequence; The image block sequence is input into the backbone feature extraction layer of the residual network, and features are extracted through three residual units. Each residual unit contains two 3×3 convolutional layers and a short-circuit connection. The convolutional layer uses the ReLU activation function and batch normalization to obtain the backbone feature map. Perform residual mapping calculation on the backbone feature map to obtain a residual feature map, and perform element-level addition operation on the residual feature map and the image block sequence to obtain enhanced image data; The enhanced image data is input into the first-level feature extraction network of the two-level feature extractor. The first-level feature extraction network contains a pyramid structure consisting of five convolutional layers. After each convolution layer, the maximum pooling operation is used to obtain a multi-scale feature map. Perform channel attention calculation on the multi-scale feature map to obtain the channel weight matrix, and perform weighted processing on the multi-scale feature map according to the channel weight matrix to obtain the first-level feature data; The first-level feature data is input into the second-level feature extraction network of the two-level feature extractor. The second-level feature extraction network adopts four parallel dilated convolution branches, and the dilation rates are set to 1, 2, 4, and 8 respectively, to obtain a multi-receptive field feature map. Perform spatial attention calculation on the multi-receptive field feature map to obtain a spatial weight matrix, and perform weighted fusion on the multi-receptive field feature map according to the spatial weight matrix to obtain the second-level feature data; The second-level feature data is processed by a feature integration module, which includes two fully connected layers and one Softmax layer, to obtain the feature data of medical products.

[0028] Specifically, the brightness of the medical product image data is analyzed to obtain the image brightness distribution matrix. The image brightness distribution matrix is ​​used to describe the brightness value of each pixel in the image, forming a two-dimensional matrix. Assume that the input image is , whose size is ,in and Represent the height and width of the image respectively, then the brightness distribution matrix Each element of Calculated by the following formula: ; in, , , Respectively represent the image at position The pixel values ​​of the red, green and blue channels at . According to the above formula, the brightness distribution matrix Reflects the brightness information of each position in the image, so that the image can be divided into blocks according to these brightness values. Divide the original image into multiple smaller areas so that more detailed feature extraction can be performed on these areas. Divided into size image blocks, then we get the image block sequence ,in Indicates the number of image blocks. Each image block Include The image blocks are used as independent inputs to facilitate subsequent feature extraction. The image block sequence is input into the backbone feature extraction layer of the residual network. The backbone feature extraction layer of the residual network adopts a structure composed of three residual units, each of which contains two The convolution layer is used to extract features from the input image block. The convolution kernel size of each convolution layer is , its output feature map is expressed as: ; in, is the input feature map of the previous layer, represents the weight of the current convolutional layer, is a bias term, * indicates a convolution operation, BN indicates batch normalization, which is used to improve the training stability and convergence speed of the model, and ReLU indicates an activation function, which is used to increase the nonlinear expression ability of the network. The function of the short-circuit connection is to add the input directly to the output of the convolution layer to form a skip connection, so that the gradient can be better transmitted and avoid the gradient vanishing problem in the deep network. After being processed by three residual units, the backbone feature map is obtained. . For the backbone feature map Perform residual mapping calculation to obtain the residual feature map The purpose of residual mapping is to capture subtle changes in feature maps and enhance the model's ability to learn local details. With the original image block sequence Perform element-level addition operations to obtain enhanced image data , and its calculation formula is: ; The information of the original image is fused with the extracted residual features to enhance the feature representation ability of the image. The enhanced image data is input into the first-level feature extraction network of the two-level feature extractor. The first-level feature extraction network adopts a pyramid structure consisting of five convolutional layers. The maximum pooling operation is used after each convolution layer to reduce the spatial size of the feature map and enhance the network's perception of features of different scales. Assume that the input feature map of the first-level network is , the output after each layer of convolution is: ; in, For the The output feature map of the layer convolution, and Respectively After five layers of convolution and maximum pooling, a multi-scale feature map is obtained. These feature maps contain product information at different scales, which helps capture the global and local features of products. Perform channel attention calculation to highlight the most valuable channels in the feature map and suppress irrelevant features. Channel attention weight matrix The calculation formula is: ; in, and are weights and biases respectively, Represents an activation function (usually Sigmoid), which is used to map the output to the interval (0, 1). The multi-scale feature map is weighted according to the channel weight matrix to obtain the first-level feature data : ; The first-level feature data is input into the second-level feature extraction network of the two-level feature extractor. The second-level feature extraction network uses four parallel dilated convolution branches, and the dilated rates are set to 1, 2, 4, and 8 respectively. The purpose of dilated convolution is to increase the receptive field without increasing the parameters, so as to capture features in a larger range. The output of the dilated convolution is , and its calculation formula is: ; in, The void rate is The convolution operation, and Respectively Through four parallel dilated convolution branches, we can obtain multi-receptive field feature maps. . For multiple receptive field feature maps Perform spatial attention calculations to highlight the most important spatial locations in the feature map. Spatial attention weight matrix The calculation formula is: ; The multi-receptive field feature maps are weighted fused according to the spatial weight matrix to obtain the second-level feature data : ; The second-level feature data is integrated through the feature integration module The feature integration module contains two fully connected layers and one Softmax layer. The function of the first fully connected layer is to reduce the dimension of the feature and obtain the intermediate feature. : ; in, and are the weights and biases of the first fully connected layer. The output dimension of the second fully connected layer is the same as the number of medical product categories and is used for classification prediction. The output is : ; Through the Softmax layer Normalization is performed to obtain the probability of each product category and the characteristic data of medical products. The calculation formula of Softmax is: ; in, Indicates that the product belongs to The probability of the class.

[0029] In a specific embodiment, the process of executing step 103 may specifically include the following steps: The characteristic data of medical products are input into the product recognition backbone network. The product recognition backbone network adopts the DenseNet structure, which includes four densely connected blocks. Six convolutional layers are set in each densely connected block. Dense skip connections are used between adjacent convolutional layers to obtain a dense map of product characteristics. Perform multi-scale feature fusion operations on the dense feature map of the product. Use a 1×1 convolution layer to unify the number of channels of feature maps of different scales to 256, and use bilinear interpolation to adjust the spatial size of the feature map to the same size to obtain the product feature fusion data. The product feature fusion data is input into the type recognition branch network. The type recognition branch network contains three parallel attention modules. Each attention module is composed of a channel attention unit and a spatial attention unit in series to obtain a product type feature vector. The product type feature vector is input into the type classifier. The type classifier includes two fully connected layers and a Softmax layer. The output dimension of the first fully connected layer is 512, and the output dimension of the second fully connected layer is the same as the number of medical product categories, so as to obtain the product type information; The product feature fusion data is input into the location detection branch network. The location detection branch network adopts the FPN structure, including a top-down feature pyramid and lateral connections, to obtain a multi-level location feature map; Perform region proposal calculation on the multi-level position feature map, and obtain candidate region features through three anchor box generators of different scales and non-maximum suppression operations. The dimension of each candidate region feature is 256. The candidate region features are input into the position regressor. The position regressor consists of four fully connected layers. Each layer uses the ReLU activation function. The last layer outputs four coordinate values ​​to represent the bounding box coordinates of the product location to obtain the product location information. The product type information and the product location information are subjected to label alignment processing, and the product type and the location information are associated and matched through the label alignment module to obtain the medical product identification result.

[0030] Specifically, the product recognition backbone network is used to perform deep feature extraction on the input medical product feature data. The product recognition backbone network adopts the DenseNet (densely connected network) structure, which contains multiple densely connected blocks. The network consists of four densely connected blocks, each of which contains six convolutional layers, and dense jump connections are used between adjacent convolutional layers. This dense connection method can effectively improve the reusability of features and enhance the expression ability of the network. For the input medical product feature data , after being processed by the first densely connected block, the output feature representation is: ; in, Indicates Concat is a convolution operation. In this way, the output of each convolution layer is concatenated with the output of all previous layers to form an output with richer features. After four such densely connected blocks, the feature dense map of the product is obtained. This feature-intensive map contains detailed feature descriptions of medical products at different levels and different spatial locations. Perform multi-scale feature fusion operations to integrate feature information from different scales and enhance the feature description capability. The convolution layer unifies the number of channels of feature maps of different scales to 256 to reduce the computational complexity while ensuring full expression of information. The feature map after convolution is , and its calculation formula is: ; in, represents the weight of the convolution kernel, is the bias term, and * is the convolution operation. The bilinear interpolation method is used to adjust the spatial size of the feature map to the same size to obtain the product feature fusion data. The role of bilinear interpolation is to spatially align features of different scales for subsequent feature integration and processing. The obtained product feature fusion data The input is sent to the type recognition branch network to identify the product category. The type recognition branch network contains three parallel attention modules, each of which is composed of a channel attention unit and a spatial attention unit in series. The channel attention unit weights each channel in the feature map to highlight important channel features. Channel weight matrix The calculation formula is: ; in, is the weight of the channel weight matrix, is the bias, Represents the activation function (usually the Sigmoid function), which is used to limit the weight value to the interval (0, 1). The spatial attention unit is used to calculate the weight of each spatial position to highlight the most valuable spatial position in the feature map. Spatial weight matrix The calculation formula is: ; Through the processing of channel attention and spatial attention, the type recognition branch network obtains the product type feature vector , contains the category-related features of medical products. The data is input into the type classifier to classify the product. The type classifier consists of two fully connected layers and a Softmax layer. The output dimension of the first fully connected layer is 512, which is used to reduce and integrate the product type features. Output It is expressed as: ; in, and are the weights and biases of the first fully connected layer. The output dimension of the second fully connected layer is the same as the number of medical product categories, which is used to map the features to the specific product category space. The output is : ; Through the Softmax layer Perform normalization to obtain the probability distribution of each product category and obtain the product type information : ; in, Indicates that the product belongs to At the same time, the product features are integrated into the data Input to the position detection branch network, the position detection branch network adopts the FPN (feature pyramid network) structure, the FPN structure contains a top-down feature pyramid and lateral connections to generate a multi-level position feature map The advantage of FPN is that it can simultaneously utilize the semantic information of high-level features and the spatial details of low-level features to achieve more accurate position detection. Perform region proposal calculations and use three anchor box generators of different scales to generate candidate regions. Each anchor box generator is used to generate candidate regions of different sizes and proportions to ensure that products can be effectively detected regardless of their size in the image. Non-maximum suppression operations are used to remove duplicate and redundant candidate regions and retain the most representative candidate region features. ,in Represents the index of the candidate region. The dimension of each candidate region feature is 256, which is used to describe the characteristic information of the product in the region. The input is sent to the position regressor, which contains four fully connected layers. Each layer uses the ReLU activation function to enhance the nonlinear expression ability of the network. After the four-layer fully connected processing, the position regressor outputs four coordinate values. , represents the bounding box coordinates of the product location, where and Represent the center position of the bounding box, and Represent the width and height of the bounding box respectively. These coordinate values ​​constitute the location information of the product, which is used to accurately locate the position of the product in the image. The product type information and the product location information are label aligned, and the product type and location information are associated and matched through the label alignment module to obtain the final medical product recognition result.

[0031] In a specific embodiment, the process of executing step 104 may specifically include the following steps: The commodity type information and commodity location information in the medical commodity identification result are input into the vending machine inventory detection unit, the commodity type information is classified and statistically processed to obtain a commodity category quantity table, and a commodity aisle distribution map is generated according to the commodity location information; Compare and calculate the commodity layout parameters in the initial commodity listing plan with the commodity category quantity table to obtain the commodity missing data, and establish a replenishment demand matrix based on the commodity missing data; Perform spatial mapping analysis on the product aisle distribution map and product layout parameters to obtain product location deviation data, and generate aisle adjustment suggestions based on the product location deviation data; Prioritize and sort the replenishment demand matrix, set the replenishment level threshold based on the replenishment threshold parameter, and obtain graded replenishment demand data; The hierarchical replenishment demand data and the channel adjustment suggestions are integrated and processed to obtain commodity inventory data, and the commodity inventory data is encrypted through a data encryption module to obtain encrypted inventory data; The encrypted inventory data is sent to the mobile management terminal through the wireless communication module, and identity authentication and data decryption are performed on the mobile management terminal to obtain the terminal inventory data; Generate a time-sharing replenishment plan based on terminal inventory data and sales period parameters, and optimize the replenishment path based on the time-sharing replenishment plan to obtain replenishment path data. The replenishment path data is input into the replenishment scheduling module, and task allocation calculations are performed in combination with personnel configuration information to obtain replenishment task instructions.

[0032] Specifically, the commodity type information is classified and counted to generate a commodity category quantity table. The commodity type information is output by the commodity recognition model and contains the specific category of each commodity in the vending machine, such as masks, disinfectants, thermometers, etc. By classifying and counting the commodity type information, the quantity of each commodity is obtained, which is recorded as ,in Represents the category of the product. Representative Category The number of goods. Table of the number of commodity categories Represented as a vector: ; in, is the total number of different commodity categories in the vending machine. This quantity table intuitively shows the inventory status of various commodities in the vending machine. At the same time, according to the commodity location information in the medical commodity recognition results, a commodity aisle distribution map is generated. Assume that the location of each aisle in the vending machine is ,in The number of the aisle is represented by the center coordinate of each product. By integrating these location information, we can get the product aisle distribution map. , which describes the specific placement of each product in the vending machine. The product layout parameters in the initial product listing plan Compare and contrast operations are performed to detect whether there are any missing items in the vending machine. Product layout parameters Describe the expected quantity of each product when it is initially listed, denoted as The missing data of commodities are calculated by the following formula: ; in, Indicates category The missing quantity for the item. If the product is missing data, it means that the product is out of stock and needs to be replenished. , establish a replenishment demand matrix , which describes the replenishment demand for each product category. and product layout parameters Perform spatial mapping analysis to obtain product location deviation data. Product location deviation data is used to describe the difference between the actual product placement location and the initial planned location. Suppose the expected location of a product is , while its actual position is , then the position deviation It is expressed as: ; By calculating the position deviation of all commodities, we can get the commodity position deviation data matrix Based on the position deviation data, aisle adjustment suggestions are generated to optimize the placement of goods in the vending machine and ensure that the position of goods is more reasonable to improve customer accessibility and purchase convenience. Perform priority sorting calculations to determine which products need to be replenished first. Based on replenishment threshold parameters , set replenishment level threshold Determine the urgency of replenishment. Assume that the replenishment threshold parameter is , then the replenishment level is expressed as: ; in, Indicates category Items that need to be replenished first. Indicates that the product does not need urgent replenishment. According to the replenishment level, the graded replenishment demand data is obtained . The replenishment demand data will be graded Integrate with channel adjustment suggestions to obtain product inventory data The data includes the current inventory quantity, replenishment requirements and placement adjustment suggestions for each product. The product inventory data is encrypted using the data encryption module to obtain encrypted inventory data. Protect the privacy and integrity of data and avoid attacks or tampering during transmission. Encrypted inventory data is transmitted to the wireless communication module. After receiving the data, the mobile management terminal will perform identity authentication to ensure that only authorized personnel can access the sensitive information. Decrypt to get the terminal inventory data , used for subsequent replenishment management and decision-making. Based on terminal inventory data and sales period parameters , generate a time-sharing replenishment plan. The sales period parameter describes the sales of each product in different time periods, such as the peak sales periods in the morning and at noon. The goal of the time-sharing replenishment plan is to reasonably arrange the replenishment time according to the sales peaks and troughs to maximize customer demand and reduce the risk of out-of-stock products. The time-sharing replenishment plan is represented as a matrix , where each element represents the replenishment arrangement of a certain product in a certain period of time. , optimize the replenishment path and obtain the replenishment path data The purpose of replenishment path optimization is to find the optimal replenishment route to reduce replenishment time and operating costs. The optimization of replenishment path is achieved by solving the traveling salesman problem, where the objective function of path length is: ; in, Indicates replenishment site and The distance between is the number of replenishment stations. By minimizing the path length , find the optimal replenishment path. Input into the replenishment scheduling module, combined with current staffing information , perform task allocation calculations and obtain replenishment task instructions . Restocking Task Instructions Contains the specific tasks of each replenishment staff, including detailed information such as the type of goods to be replenished, quantity, replenishment time and replenishment route.

[0033] In a specific embodiment, the process of executing step 105 may specifically include the following steps: Input the replenishment path data and the graded replenishment demand data in the replenishment task instruction into the replenishment execution unit, perform task planning for the replenishment sites according to the replenishment path data, and obtain the replenishment execution task sequence; The RFID reader is configured based on the replenishment execution task sequence, and the RFID tag information of the replenishment goods is scanned by the RFID reader to obtain the replenishment goods tag data; Match and verify the replenishment product label data with the product category quantity table to obtain the product verification result, and generate the replenishment execution status data based on the product verification result; The replenishment execution status data and aisle adjustment suggestions are processed collaboratively to obtain aisle placement guidance data, and the replenishment products are spatially arranged according to the aisle placement guidance data; Monitor the replenishment progress in real time based on the product verification results and replenishment execution status data, obtain replenishment abnormality signals, and compare the replenishment abnormality signals with the replenishment threshold parameters to generate replenishment warning data; Send the replenishment warning data to the mobile management terminal through the wireless communication module, perform graded processing on the replenishment warning data, and obtain a replenishment adjustment plan; Generate replenishment operation data based on the replenishment adjustment plan and the cargo lane layout guidance data, and write the replenishment operation data into the database of the medical vending platform to obtain the replenishment history record; Data fusion processing is performed on replenishment history records and product layout parameters to update the real-time inventory information of the medical vending platform, and the real-time inventory information is fed back to the vending machine inventory detection unit.

[0034] Specifically, the replenishment path data and the hierarchical replenishment demand data in the replenishment task instruction are combined to plan the specific replenishment operation. The replenishment path data describes the visit order of each replenishment station, while the hierarchical replenishment demand data contains the types and quantities of goods that need to be replenished at each station. Based on this information, the replenishment execution unit formulates a replenishment execution task sequence. Represents replenishment path data, where , indicating that the replenishment station. Represents hierarchical replenishment demand data, including the types and quantities of goods required to be replenished at each replenishment station. and Combined, we get the replenishment execution task sequence , which is of the form: ; in, Indicates that it is at the replenishment site The types and quantities of goods that need to be replenished are determined by the RFID reader / writer. The RFID reader / writer is configured based on the replenishment execution task sequence to ensure that the RFID reader / writer can accurately identify the tag information of the replenishment goods. The RFID tag information of each replenishment product is scanned by the RFID reader / writer to obtain the replenishment product tag data. ,in Indicates The label information of replenishment products. These label information contains the unique identification of each product, which is used to match and verify the product information in the system. Verify the matching with the commodity category quantity table to ensure that the types and quantities of the replenished commodities are consistent with the demand. Suppose the commodity category quantity table is , represents the expected inventory quantity of each commodity. and Match and get the product verification results , which is of the form: ; in, Indicates The verification results of the following commodities: Indicates that the verification has passed. Indicates that the verification failed. If some products fail the verification, it means that there is an error in the replenishment process, such as the wrong type of replenishment product or insufficient quantity. Generate replenishment execution status data based on the product verification results , used to record the execution status of replenishment tasks. Coordinate with cargo lane adjustment suggestions to generate cargo lane layout guidance data , which contains the specific placement and aisle number of each product in the vending machine. The replenishment products are spatially arranged according to the aisle placement guidance data, and each product is placed in the vending machine according to the specified position. During the execution of the replenishment task, the replenishment progress is monitored in real time to promptly detect and handle any abnormal situations. According to the product verification results and replenishment execution status data , get the replenishment abnormal signal The replenishment exception signal is used to alert the system to any anomalies that occur during the replenishment process, such as insufficient quantity of goods, wrong product type, or replenishment delay. Compare and generate replenishment warning data , to determine which exceptions need to be handled immediately and which can be resolved later. The formula for replenishment warning is: ; in, Indicates an exception that needs to be handled immediately. Indicates that it can be processed later. The replenishment warning data is sent to the mobile management terminal through the wireless communication module. After receiving the data, the mobile management terminal performs graded processing on the warning data to obtain a replenishment adjustment plan. Different response strategies are formulated according to the severity of the warning. More urgent problems are solved first, while minor deviations are corrected at the next replenishment. Based on the replenishment adjustment plan and the aisle layout guidance data, the final replenishment operation data is generated. The replenishment operation data includes every step of replenishment, the placement of goods, and any necessary adjustments. The replenishment operation data is written into the database of the medical vending platform to generate a replenishment history record. The replenishment history record provides the vending platform with a detailed operation track for subsequent inventory management and analysis, helping to optimize the replenishment strategy. The replenishment history record is fused with the product layout parameters to update the real-time inventory information of the medical vending platform. Assume that the product layout parameters are , the replenishment history is , get updated real-time inventory information through data fusion : ; in, Indicates data fusion operation. Real-time inventory information It contains the current inventory quantity, placement and replenishment history of each product. The updated real-time inventory information is fed back to the vending machine inventory detection unit so that the system can understand the inventory status of the products in the vending machine in real time and make replenishment and adjustments in time.

[0035] In a specific embodiment, the process of executing step 106 may specifically include the following steps: The real-time inventory information and replenishment history records are associated with each other according to the commodity category quantity table to obtain commodity circulation data, and the commodity circulation data is analyzed in the time dimension to obtain commodity sales samples; A sales location association matrix is ​​constructed based on product sales samples and product location deviation data, and cluster analysis is performed on the sales location association matrix to obtain product regional popularity data; Cross-validate the product area popularity data with the graded replenishment demand data to obtain product turnover rate data, and generate a product zoning configuration plan based on the product turnover rate data; Perform spatial mapping operations on the commodity partition configuration plan and cargo lane layout guidance data to obtain optimized layout data, and update commodity layout parameters based on the optimized layout data; Dynamically update and calculate the replenishment threshold parameters based on the commodity sales rate data to obtain updated replenishment threshold parameters, and input the updated replenishment threshold parameters into the replenishment warning module; Perform time period correlation analysis based on the time-sharing replenishment plan and replenishment execution status data to obtain time period characteristic data, and update sales time period parameters based on the time period characteristic data; Performing pattern fusion operation on the updated commodity layout parameters, the updated replenishment threshold parameters, and the updated sales period parameters to obtain commodity sales features; The parameters of the initial product listing plan are updated based on the product sales characteristics to obtain a target product listing plan, and the target product listing plan is written into the operation database of the medical vending platform.

[0036] Specifically, real-time inventory information and replenishment history records are integrated to form a complete commodity circulation record. Real-time inventory information Describes the current quantity of each item in the vending machine, and the replenishment history Contains detailed information for each replenishment operation. Used to define the category of each product, by With replenishment history Perform data association to obtain commodity flow data . Record the changes in each category of goods over a period of time, including increases (restocking) and decreases (sales) in inventory. The flow data is , which is expressed as: ; in, Indicates time Goods of the moment The inventory volume, Indicates time By analyzing the time dimension of commodity circulation data, we can obtain commodity sales samples, including sales changes of various commodities in different time periods. We can build a sales location association matrix based on commodity sales samples and commodity location deviation data. The commodity location deviation data describes the deviation between the location of each commodity in the vending machine and the expected location, which can help analyze whether the location of the commodity affects its sales. The elements of the sales location association matrix are Indicates product category In Location The correlation between sales and location is calculated as: ; in, Display products In time Sales volume, For goods The average sales volume, Indicates location The deviation value of is the average value of the position deviation. By clustering the sales position association matrix, we can obtain the product area heat data, which represents the sales activity of different areas in the vending machine, so as to identify the impact of the product placement on its sales. We can cross-validate the product area heat data with the graded replenishment demand data to obtain the product turnover rate data. The product turnover rate reflects the sales speed and inventory turnover of the product, and is used to determine which products are selling well and which products are unsalable. The calculation formula is: ; in, Display products Sales volume, For goods The average inventory level of a product. By calculating the turnover rate data, a product partition configuration plan is generated. The plan is used to replan the placement area of ​​the product, such as placing products with higher turnover rates in locations that are easier for customers to see and use. A spatial mapping operation is performed on the product partition configuration plan and the aisle layout guidance data to obtain the optimized layout data. The optimized layout data contains the optimal placement position for each product in the vending machine. The spatial mapping operation combines the partition configuration plan with the aisle layout data to ensure that the placement of the products is in line with sales rules and can maximize the use of the space resources of the vending machine. Based on the optimized layout data, the product layout parameters are updated to make the placement of the products in the vending machine more scientific and reasonable. At the same time, the replenishment threshold parameters are dynamically updated and calculated based on the product turnover rate data to obtain the updated replenishment threshold parameters. The replenishment threshold parameter determines when a certain type of goods needs to be replenished. By analyzing the turnover rate, if the turnover rate of a certain type of goods is high, its replenishment threshold is lowered accordingly to reduce the risk of out-of-stock; conversely, if the turnover rate of a certain type of goods is low, its replenishment threshold is appropriately increased to avoid inventory backlogs. The update formula of the replenishment threshold is: ; in, For goods The old replenishment threshold of is the adjustment coefficient, is the average turnover rate of all commodities. Through this formula, the replenishment threshold is adjusted dynamically to make it more in line with the actual sales of the commodities. The updated replenishment threshold parameters are input into the replenishment warning module so that a warning can be issued in time when the inventory is insufficient. The time period association analysis is performed based on the time-sharing replenishment plan and the replenishment execution status data to obtain the time period characteristic data. The time period characteristic data describes the sales characteristics of different commodities in each time period. For example, the sales volume of certain commodities increases or decreases significantly in a specific time period. Based on the time period characteristic data, the sales time period parameters are updated to optimize the replenishment and sales strategies of the vending machine to better meet customer needs. The updated commodity layout parameters, the updated replenishment threshold parameters and the updated sales time period parameters are subjected to pattern fusion operation to obtain the commodity sales characteristics. The pattern fusion operation combines feature information from different sources to generate a global description of product sales features, which is used to guide product listing and replenishment decisions. Product sales features It includes the product's sales situation, location layout, replenishment strategy and time characteristics, which is a comprehensive description of the product's sales performance. According to the product sales characteristics, the parameters of the initial product listing plan are updated to obtain the target product listing plan. The target product listing plan is the result of optimization based on the initial plan, combined with real-time inventory, sales samples, sales rate, location popularity and other factors, which can maximize customer satisfaction and improve the operating efficiency of the vending machine. Update using the following formula: ; in, Indicates the products in the initial listing plan Parameters, is the adjustment coefficient, which is used to control the impact of sales characteristics on the final listing plan. The data is written into the operation database of the medical vending platform to ensure that the vending machine can operate according to the optimal product display and replenishment strategy in subsequent operations.

[0037] The above describes the mobile management method of the medical vending platform in the embodiment of the present invention. The following describes the mobile management device of the medical vending platform in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a mobile management device for a medical vending platform includes: The acquisition module 201 is used to create an initial product shelf plan based on the vending machine operation data of the medical vending platform and to acquire medical product image data; An extraction module 202, for performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data; The identification module 203 is used to input the medical commodity feature data into the medical commodity identification model to perform medical commodity identification and obtain a medical commodity identification result; The detection module 204 is used to perform inventory detection according to the medical commodity identification result, generate commodity inventory data, and send the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; A generation module 205, for inputting the RFID tag information and the replenishment task instruction into the replenishment execution unit, generating replenishment operation data, and updating the real-time inventory information; The adjustment module 206 is used to adjust the initial product listing plan based on the real-time inventory information and historical sales data to generate a target product listing plan.

[0038] Through the coordinated cooperation of the above components, through deep learning and computer vision technology, in conjunction with the RFID identification system, accurate identification and real-time inventory monitoring of medical products are achieved; the use of a two-stage feature extractor and residual learning method effectively improves the accuracy of product identification in complex environments; by establishing a dynamic product shelving optimization mechanism, intelligent adjustment of product layout and automatic optimization of replenishment strategies are achieved; by introducing mobile management terminals and replenishment execution units, a complete replenishment-acceptance-shelf closed-loop management system is constructed, which significantly improves the operational efficiency and service quality of the medical vending platform; through intelligent analysis of real-time inventory information and sales data, an adaptive optimization mechanism for sales models is established, which achieves the rational allocation of product resources and effective control of operating costs.

[0039] above Figure 2 The mobile management device of the traditional Chinese medicine vending platform in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The mobile management device of the medical vending platform in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0040] Figure 33 is a structural diagram of a medical vending platform mobile management device provided by an embodiment of the present invention. The medical vending platform mobile management device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the medical vending platform mobile management device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the medical vending platform mobile management device 300 to implement the steps of the above-mentioned medical vending platform mobile management method.

[0041] The medical vending platform mobile management device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the mobile management device for medical vending platforms shown does not constitute a limitation on the mobile management device for medical vending platforms provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0042] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the medical vending platform mobile management method.

[0043] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0044] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0045] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A mobile management method for a medical vending platform, characterized in that: The method comprises: Create an initial product listing plan based on the vending machine operation data of the medical vending platform and collect medical product image data; Performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data; Inputting the medical commodity feature data into a medical commodity identification model to perform medical commodity identification to obtain a medical commodity identification result; Performing inventory detection according to the medical commodity identification result to generate commodity inventory data, and sending the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; Inputting the RFID tag information and the replenishment task instruction into the replenishment execution unit, generating replenishment operation data, and updating the real-time inventory information; The initial product listing plan is adjusted based on the real-time inventory information and historical sales data to generate a target product listing plan.

2. The medical vending platform mobile management method according to claim 1, characterized in that: The method of creating an initial product shelf plan based on the vending machine operation data of the medical vending platform and collecting medical product image data includes: Performing data cleaning on the commodity sales records of each vending machine in the medical vending platform to obtain valid sales data, and classifying the valid sales data according to commodity category and sales time to obtain classified sales data; Performing time series analysis on the classified sales data to obtain sales curves for each time period, calculating the sales frequency of commodities in each time period based on the sales curves, and obtaining sales time period parameters; Input the sales period parameter into the commodity turnover rate calculation model to obtain commodity turnover speed data, set the safety inventory of each commodity according to the commodity turnover speed data, and obtain the replenishment threshold parameter; Perform spatial distribution analysis on the commodity placement location data of each vending machine to obtain commodity location heat data, and prioritize the commodity placement locations according to the commodity location heat data to obtain location priority data; Performing correlation analysis on the location priority data and the sales period parameters to obtain matching data between the location and the period, generating a commodity location distribution matrix based on the matching data to obtain commodity layout parameters; generating an initial product shelf plan according to the replenishment threshold parameter, the product layout parameter and the sales period parameter, and monitoring the remaining quantity of each product in the vending machine based on the initial product shelf plan to obtain a product out-of-stock risk value, and determining a priority order of image acquisition according to the product out-of-stock risk value; The commodity display in the vending machine is spatially mapped according to the commodity layout parameters to obtain commodity coordinate data, and the acquisition angle of the image acquisition device is set based on the commodity coordinate data. According to the image acquisition angle and the priority of the image acquisition, the image acquisition device is controlled to perform adaptive angle acquisition to obtain medical commodity image data.

3. The medical vending platform mobile management method according to claim 2, characterized in that: The performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data includes: Performing brightness analysis on the medical product image data to obtain an image brightness distribution matrix, and performing block processing on the medical product image data according to the image brightness distribution matrix to obtain an image block sequence; Inputting the image block sequence into the backbone feature extraction layer of the residual network, performing feature extraction through three residual units, each residual unit comprising two 3×3 convolutional layers and a short-circuit connection, the convolutional layer adopts a ReLU activation function and batch normalization processing, and obtaining a backbone feature map; Performing residual mapping calculation on the backbone feature map to obtain a residual feature map, and performing element-level addition operation on the residual feature map and the image block sequence to obtain enhanced image data; Inputting the enhanced image data into a first-level feature extraction network of a two-level feature extractor, wherein the first-level feature extraction network comprises a pyramid structure consisting of five convolutional layers, and a maximum pooling operation is used after each convolution layer to obtain a multi-scale feature map; Performing channel attention calculation on the multi-scale feature map to obtain a channel weight matrix, and performing weighted processing on the multi-scale feature map according to the channel weight matrix to obtain first-level feature data; The first-level feature data is input into the second-level feature extraction network of the two-level feature extractor, wherein the second-level feature extraction network adopts four parallel dilated convolution branches, and the dilated rates are set to 1, 2, 4, and 8 respectively, to obtain a multi-receptive field feature map; Performing spatial attention calculation on the multi-receptive field feature map to obtain a spatial weight matrix, and performing weighted fusion on the multi-receptive field feature map according to the spatial weight matrix to obtain second-level feature data; The second-level feature data is processed by a feature integration module, wherein the feature integration module includes two fully connected layers and one Softmax layer, to obtain medical product feature data.

4. The medical vending platform mobile management method according to claim 3, characterized in that: The step of inputting the medical commodity feature data into a medical commodity identification model to perform medical commodity identification to obtain a medical commodity identification result includes: Inputting the medical product feature data into a product recognition backbone network, wherein the product recognition backbone network adopts a DenseNet structure, includes four densely connected blocks, each of which has six convolutional layers, and dense skip connections are used between adjacent convolutional layers to obtain a dense map of product features; A multi-scale feature fusion operation is performed on the dense feature map of the product, the number of channels of the feature maps of different scales is unified to 256 through a 1×1 convolution layer, and the spatial size of the feature maps is adjusted to the same size by using bilinear interpolation to obtain product feature fusion data; Inputting the commodity feature fusion data into a type recognition branch network, wherein the type recognition branch network comprises three parallel attention modules, each of which is composed of a channel attention unit and a spatial attention unit in series, to obtain a commodity type feature vector; Input the commodity type feature vector into a type classifier, wherein the type classifier comprises two fully connected layers and a Softmax layer, wherein the output dimension of the first fully connected layer is 512, and the output dimension of the second fully connected layer is the same as the number of medical commodity categories, to obtain commodity type information; Inputting the commodity feature fusion data into the position detection branch network, the position detection branch network adopts an FPN structure, including a top-down feature pyramid and lateral connections, to obtain a multi-level position feature map; Performing region proposal calculation on the multi-level position feature map, obtaining candidate region features through three anchor box generators of different scales and non-maximum suppression operations, each of which has a dimension of 256; Input the candidate region features into a position regressor, wherein the position regressor comprises four fully connected layers, each layer uses a ReLU activation function, and the last layer outputs four coordinate values ​​representing the coordinates of the bounding box of the commodity position to obtain the commodity position information; The commodity type information and the commodity location information are subjected to label alignment processing, and the commodity type and the location information are associated and matched by a label alignment module to obtain a medical commodity identification result.

5. The medical vending platform mobile management method according to claim 4, characterized in that: The method of performing inventory detection according to the medical commodity identification result, generating commodity inventory data, and sending the commodity inventory data to the mobile management terminal to generate a replenishment task instruction includes: Inputting the commodity type information and commodity location information in the medical commodity identification result into the vending machine inventory detection unit, classifying and statistically processing the commodity type information to obtain a commodity category quantity table, and generating a commodity aisle distribution map according to the commodity location information; Comparing the commodity category quantity table with the commodity layout parameters in the initial commodity listing plan, obtaining commodity missing data, and establishing a replenishment demand matrix based on the commodity missing data; Performing spatial mapping analysis on the commodity aisle distribution map and the commodity layout parameters to obtain commodity position deviation data, and generating aisle adjustment suggestions based on the commodity position deviation data; Prioritizing and calculating the replenishment demand matrix, setting a replenishment level threshold based on the replenishment threshold parameter, and obtaining graded replenishment demand data; Integrate the graded replenishment demand data and the channel adjustment suggestion to obtain commodity inventory data, and encrypt the commodity inventory data through a data encryption module to obtain encrypted inventory data; The encrypted inventory data is sent to the mobile management terminal through the wireless communication module, and identity authentication and data decryption are performed at the mobile management terminal to obtain the terminal inventory data; A time-sharing replenishment plan is generated according to the terminal inventory data and the sales period parameters, and replenishment path optimization is performed based on the time-sharing replenishment plan to obtain replenishment path data, and the replenishment path data is input into a replenishment scheduling module, and task allocation calculation is performed in combination with personnel configuration information to obtain replenishment task instructions.

6. The medical vending platform mobile management method according to claim 5, characterized in that: The step of inputting the RFID tag information and the replenishment task instruction into the replenishment execution unit, generating replenishment operation data, and updating the real-time inventory information includes: Inputting the replenishment path data and the graded replenishment demand data in the replenishment task instruction into a replenishment execution unit, performing task planning for replenishment sites according to the replenishment path data, and obtaining a replenishment execution task sequence; The RFID reader is configured based on the replenishment execution task sequence, and the RFID tag information of the replenishment commodity is scanned by the RFID reader to obtain the replenishment commodity tag data; Matching and verifying the replenishment commodity label data with the commodity category quantity table to obtain a commodity verification result, and generating replenishment execution status data according to the commodity verification result; The replenishment execution status data and the aisle adjustment suggestion are collaboratively processed to obtain aisle placement guidance data, and the replenishment commodities are spatially arranged according to the aisle placement guidance data; real-time monitoring of the replenishment progress according to the commodity verification result and the replenishment execution status data, obtaining a replenishment abnormality signal, and comparing the replenishment abnormality signal with the replenishment threshold parameter to generate replenishment warning data; The replenishment warning data is sent to the mobile management terminal through a wireless communication module, and the replenishment warning data is processed in a graded manner to obtain a replenishment adjustment plan; Generate replenishment operation data based on the replenishment adjustment plan and the cargo lane placement guidance data, and write the replenishment operation data into a database of the medical vending platform to obtain a replenishment history record; Data fusion processing is performed on the replenishment history records and the commodity layout parameters, the real-time inventory information of the medical vending platform is updated, and the real-time inventory information is fed back to the vending machine inventory detection unit.

7. The medical vending platform mobile management method according to claim 6, characterized in that: The adjusting the initial product listing plan based on the real-time inventory information and historical sales data to generate a target product listing plan includes: Performing data association between the real-time inventory information and the replenishment history records according to a commodity category quantity table to obtain commodity circulation data, and performing time dimension analysis on the commodity circulation data to obtain commodity sales samples; Constructing a sales location association matrix based on the product sales samples and the product location deviation data, and performing cluster analysis on the sales location association matrix to obtain product area popularity data; Cross-validating the commodity area popularity data with the graded replenishment demand data to obtain commodity turnover rate data, and generating a commodity zoning configuration plan based on the commodity turnover rate data; Performing a spatial mapping operation on the commodity partition configuration scheme and the cargo lane placement guidance data to obtain optimized layout data, and updating commodity layout parameters according to the optimized layout data; Dynamically update and calculate the replenishment threshold parameter based on the commodity sales rate data to obtain an updated replenishment threshold parameter, and input the updated replenishment threshold parameter into a replenishment warning module; Performing time period association analysis according to the time-sharing replenishment plan and the replenishment execution status data to obtain time period characteristic data, and updating sales time period parameters based on the time period characteristic data; Performing a pattern fusion operation on the updated commodity layout parameters, the updated replenishment threshold parameters, and the updated sales period parameters to obtain commodity sales features; The initial product listing plan is updated with parameters based on the product sales characteristics to obtain a target product listing plan, and the target product listing plan is written into an operation database of the medical vending platform.

8. A mobile management device for a medical vending platform, characterized in that: The device is used to execute the medical vending platform mobile management method according to any one of claims 1 to 7, the device comprising: The acquisition module is used to create an initial product listing plan based on the vending machine operation data of the medical vending platform and to collect medical product image data; An extraction module, used for performing residual learning and two-level feature extraction on the medical commodity image data to obtain medical commodity feature data; an identification module, used for inputting the medical commodity feature data into a medical commodity identification model to perform medical commodity identification and obtain a medical commodity identification result; A detection module, used to perform inventory detection according to the medical commodity identification result, generate commodity inventory data, and send the commodity inventory data to the mobile management terminal to generate a replenishment task instruction; A generation module, used to input the RFID tag information and the replenishment task instruction into the replenishment execution unit, generate replenishment operation data, and update the real-time inventory information; The adjustment module is used to adjust the initial product listing plan based on the real-time inventory information and historical sales data to generate a target product listing plan.

9. A mobile management device for a medical vending platform, characterized in that: The medical vending platform mobile management device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the medical vending platform mobile management device to execute the medical vending platform mobile management method according to any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the medical vending platform mobile management method according to any one of claims 1 to 7 is implemented.

Images