A fresh food area electronic label management method to maximize sales

Abstract

The present invention discloses a method for managing and controlling electronic tags in a fresh food area for maximizing sales, comprising: evaluating the overall freshness of fresh food; obtaining a historical overall freshness change curve and a sales volume curve within a preset price adjustment period T; obtaining a relationship function between the overall freshness and the sales volume under a price J condition; importing the relationship function between the overall freshness and the sales volume under the price J condition into a trend analysis module, searching for a price range corresponding to the next price adjustment period T, estimating the change curves of the overall freshness and the sales volume under different price conditions within the price range, and calculating the ideal price J for the next price adjustment period according to the following formula * The present invention can estimate the freshness level and sales volume of fresh produce products, and automatically adjust product price tags according to a promotional plan preset by the user, with the goal of maximizing sales, thus greatly reducing the time for manual adjustment and confirmation of product information.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic tag management in fresh food areas, and in particular to a method for managing and controlling electronic tags in fresh food areas for maximizing sales. Background Art

[0002] Electronic paper electronic tags have become popular in supermarkets due to their excellent display effects and convenient price updates, and are gradually replacing the original paper tags. When the price of a product needs to be adjusted, you only need to send the adjusted product price image to the corresponding electronic tag unit to achieve price updates throughout the supermarket. Electronic tags come in a variety of sizes. Taking supermarkets as an example, you can replace the original paper tags with electronic tags of the same size. For example, in the fresh food area, you can use large-screen electronic tags of more than 10 inches and hang them above the fresh food to provide customers with price reminder services.

[0003] The shelf life of fresh food products is short, so merchants usually replenish them in batches. In addition, there are many types of products in the fresh food area, so new products are frequently added. At the same time, many fresh products will lose moisture and reduce their sales probability within the same day. To solve this problem, merchants will use price reduction and other methods to increase the sales probability. After replacing the electronic tag, the staff can check the current status of the fresh food products and manually change them in the background, and then display the change results on the electronic tag.

[0004] However, the above-mentioned electronic tag update method still has the following two technical problems: First, most electronic tags are currently updated through manual background processing, and a small number of research schemes propose to automatically update electronic tags after identifying the goods through some reading devices. The invention with application number 201810299344.5 published on October 18, 2019 discloses a method for automatic identification of commodity information, which can automatically adjust the commodity information on the electronic shelf label in combination with the technology of automatic identification of commodities, greatly improving work efficiency and avoiding information inconsistency caused by mis-sending of commodities. However, the application adopts the RFID reader to realize the identification of commodity information, and can only update the electronic tag corresponding to a single commodity. However, this does not reduce the workload of the staff in the fresh food area. Once the electronic tag corresponds to multiple commodities and the location area occupied by the commodity changes, the price tag cannot be automatically updated. At present, there is no automatic update method for the price tag of large-screen electronic tags that display multiple commodities at the same time. Second, the promotional price adjustment of commodities in the fresh food area needs to be carried out by professionals, and the relevant professionals need to have both relevant knowledge and marketing knowledge of fresh commodities.

[0005] The above two technical problems mean that even if the fresh food area is changed to electronic tags to indicate prices, it cannot significantly reduce the workload of service staff in the fresh food area. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides a method for managing and controlling electronic tags in fresh food areas for maximizing sales. The method can estimate the freshness level and sales volume of goods in the fresh food area, automatically adjust the price tags of goods according to the promotional plans pre-set by the user with the goal of maximizing sales, and greatly reduce the time for manual adjustment and confirmation of product information.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for controlling electronic tags in a fresh food area to maximize sales is provided, which is used to automatically control the display content of electronic tags installed in the fresh food area. The method comprises the following steps:

[0009] S1, constructing a freshness evaluation model, wherein the freshness evaluation model is trained by using a number of stacked sample images of similar fresh products;

[0010] S2, taking images of fresh products in the designated product placement area according to a preset collection cycle, importing the fresh product images into a freshness evaluation model, and evaluating the overall freshness of the fresh products; the fresh products in the designated product placement area are set with flexible prices, and different freshness levels correspond to different price ranges;

[0011] S3, with time as the horizontal axis, fit the historical overall freshness data and corresponding sales data of fresh products evaluated in the n collection cycles from the current moment forward, and obtain the historical overall freshness change curve X(t) and sales volume curve S(t) within the preset price adjustment cycle T;

[0012] S4, according to the preset price adjustment cycle T, the price J of fresh products in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed to obtain the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition:

[0013]

[0014] Where: X T is the overall freshness at the beginning of the historical period T, F1(t) is the freshness loss value caused by water loss, and F2[Y(t)] is the freshness loss value caused by sales volume Y(t);

[0015] S5, import the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition into the trend analysis module, find the price range corresponding to the next price adjustment cycle T, estimate the change curve of the overall freshness and sales volume under different price conditions within the price range, and calculate the ideal price J for the next price adjustment cycle according to the following formula * :

[0016]

[0017] Y * (t) = S[J * , X * (t)]

[0018] X * (t) = X0-F1(t)-F2[Y * (t)]

[0019] Where: X0 is the overall freshness at the current moment, X * (t) is the estimated freshness change curve, Y * (t) is the estimated sales volume change curve.

[0020] To optimize the above technical solutions, the specific measures taken also include:

[0021] Further, in step S1, the freshness evaluation model includes a feature extraction unit, a first feature fusion unit, a second feature fusion unit, an adaptive maximum pooling layer, a global average pooling layer and a fully connected layer connected in sequence;

[0022] The feature extraction unit includes three convolution modules, which perform layer-by-layer feature extraction on the imported fresh product images of the designated product placement area; the first feature fusion unit is used to extract the outputs of the second convolution module and the third convolution module, and perform feature fusion of different sizes; the second feature fusion unit is used to perform feature fusion again on the output of the first feature fusion unit and the output of the first convolution module; the features fused by the second feature fusion unit are subjected to adaptive maximum pooling and global average pooling processing by the adaptive maximum pooling layer and the global average pooling layer respectively, and the global features of the fresh product images are extracted; the fully connected layer is used to classify the extracted global features of the fresh product images, calculate the probability that the global features of the fresh product images belong to each freshness level, and output the freshness level of the current fresh product image.

[0023] Further, in step S4, the price J of fresh products in the current price adjustment period, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed according to the preset price adjustment period T, and the process of obtaining the relationship function S[J, X(t)] between the overall freshness and the sales volume under the price J condition includes the following steps:

[0024] According to different external factors, the overall freshness change curve, sales volume curve, and overall freshness-sales volume relationship curve of fresh products within the price range [J-ΔJ, J+ΔJ] are collected to generate several groups of sample data corresponding to price J to construct a training set; ΔJ is the preset allowable price error of the fresh product;

[0025] A relationship evaluation model is constructed based on a neural network, and the relationship evaluation model is trained using a training set until convergence;

[0026] The price J of fresh products in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are introduced into the relationship evaluation model, and the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition is classified.

[0027] Further, in step S5, the relationship function S[J, X(t)] between the overall freshness and the sales volume under the price J condition is imported into the trend analysis module, and the price range corresponding to the overall freshness X0 at the current moment is found. The process of estimating the change trend of the overall freshness and the sales volume under different price conditions includes the following steps:

[0028] The GRU network model based on the attention mechanism is used to construct a trend analysis module. The output of this trend analysis module is the historical freshness level data of fresh products corresponding to the m sampling periods counted forward from the current moment, and the output is the estimated freshness level data of the m sampling periods counted backward from the current moment;

[0029] Find the corresponding price range based on the estimated freshness level data;

[0030] The call relationship evaluation model is used to estimate the change curves of overall freshness and sales volume under different price conditions within the price range.

[0031] Furthermore, the control method further comprises the following steps:

[0032] A1, generating the correspondence between the electronic tag and the unit product area;

[0033] A2, regularly photographing the shelf image of the unit product area corresponding to each electronic tag within a preset field of view centered on the unit product area of ​​the electronic tag;

[0034] A3, preliminarily divide the shelf image into unit product areas, perform similarity judgment on the images of adjacent unit product areas, regard adjacent unit product areas with similarity higher than the similarity threshold as the same product placement area, preliminarily integrate the unit product areas on the shelf image, generate several product placement areas, and mark each product placement area; then call the product recognition model to identify the product type of each product placement area;

[0035] A4, according to the frequency of occurrence of the product placement area, the product placement area corresponding to only one electronic tag is defined as the first placement area, and the remaining product placement areas are defined as the second placement areas, and the main label of each second placement area is determined according to the image area ratio;

[0036] A5, retrieve the product label image corresponding to each product placement area, splice the retrieved product label images according to the positional relationship of each product placement area on the shelf image, and generate an electronic label price tag; wherein the label image of the second placement area other than the main label is scaled according to the area occupied by the second placement area on the shelf image;

[0037] A6, sends the generated electronic label price tag to the corresponding electronic label for display.

[0038] Furthermore, in step A4, when one of the product placement areas contains more than two products, the product types contained therein are matched with the product types of the adjacent product placement areas. If all product types are matched successfully, the product placement area is cleared; otherwise, a corresponding product placement area is generated according to the product types that failed to match; adjacent product placement areas refer to product placement areas that are within the coverage of the same electronic tag display.

[0039] Furthermore, in step A5, the process of generating the electronic price tag includes the following steps:

[0040] A51, for each electronic tag, read the number of the product placement area and the product type on each product placement area in the current verification cycle, and retrieve the product label image data corresponding to each product placement area according to the product type;

[0041] A52, generating a label template according to the position information of the unit product area occupied by each product placement area, wherein the label template includes a plurality of label filling boxes, and each label filling box corresponds to a product placement area;

[0042] A53, determining whether the electronic label corresponds to a second placement area other than the main label, and if so, calling a scaling component to scale the corresponding label filling frame within a preset image size range according to the area occupied by the second placement area on the shelf image;

[0043] A54, fill the retrieved product label image data into the corresponding label filling box to generate the electronic label price tag within the current collection cycle.

[0044] Furthermore, the control method further comprises the following steps:

[0045] After the electronic tag is updated, the display image data of several electronic tags corresponding to any second placement area are retrieved, and the update result of this electronic tag is estimated by checking the similarity and scaling size of the product label image data corresponding to the second placement area in all retrieved display image data.

[0046] The beneficial effects of the present invention are:

[0047] First, the electronic label management method for the fresh food area of ​​the present invention that maximizes sales can estimate the freshness level and sales volume of the goods in the fresh food area, and automatically adjust the price tags of the goods according to the promotion plan pre-set by the user with the goal of maximizing sales, thereby greatly reducing the time for manual adjustment and confirmation of product information.

[0048] Second, the electronic label management method for the fresh food area of ​​the present invention that maximizes sales can automatically adjust the display image of the large-screen electronic label according to the real-time status of the goods on the shelf; based on the user's visual range, the user can obtain the price information of the product type (including misplaced goods) within the close visual range at any angle, thereby improving the user experience. In addition, since the price label display content can include the price information of all misplaced goods, the probability of manual rearrangement can be reduced by attracting users to buy.

[0049] Third, the electronic tag management method for the fresh food area of ​​the present invention that maximizes sales can realize automatic verification of the update results in combination with the repeated price tags of some commodities, and the verification results are highly reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the electronic tag management method for the fresh food area for maximizing sales of the present invention.

[0051] Figure 2 It is a structural diagram of the freshness evaluation model of the present invention.

[0052] Figure 3 It is a flow chart of the price updating method of the present invention.

[0053] Figure 4 It is a flow chart of the verification method of the present invention. DETAILED DESCRIPTION

[0054] The present invention will now be described in further detail with reference to the accompanying drawings.

[0055] It should be noted that the terms such as "upper", "lower", "left", "right", "front", "back", etc. cited in the invention are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments in their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0056] Figure 1 This is a flow chart of the electronic tag management method for the fresh food area to maximize sales of the present invention. Figure 1 , the control method includes the following steps:

[0057] S1, constructing a freshness evaluation model, wherein the freshness evaluation model is trained by using a number of stacked sample images of similar fresh products.

[0058] S2, taking images of fresh goods in the designated goods display area according to a preset collection cycle, importing the fresh goods images into a freshness evaluation model, and evaluating the overall freshness of the fresh goods; setting flexible prices for the fresh goods in the designated goods display area, and different freshness levels correspond to different price ranges.

[0059] S3, with time as the horizontal axis, fit the historical overall freshness data and corresponding sales data of fresh products evaluated in the n collection cycles from the current moment forward, and obtain the historical overall freshness change curve X(t) and sales volume curve S(t) within the preset price adjustment period T.

[0060] S4, according to the preset price adjustment cycle T, the price J of fresh products in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed to obtain the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition:

[0061]

[0062] Where: X T is the overall freshness at the beginning of historical period T, F1(t) is the freshness loss value caused by moisture loss, and F2[Y(t)] is the freshness loss value caused by sales volume Y(t).

[0063] S5, import the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition into the trend analysis module, find the price range corresponding to the next price adjustment cycle T, estimate the change curve of the overall freshness and sales volume under different price conditions within the price range, and calculate the ideal price J for the next price adjustment cycle according to the following formula * :

[0064]

[0065] Y * (t) = S[J * , X * (t)]

[0066] X * (t) = X0-F1(t)-F2[Y * (t)]

[0067] Where: X0 is the overall freshness at the current moment, X * (t) is the estimated freshness change curve, Y * (t) is the estimated sales volume change curve.

[0068] The preferred technical solution of this embodiment is further described in detail below in conjunction with the accompanying drawings.

[0069] Four data processing cycles are mentioned in this embodiment. In order to facilitate the explanation of the scheme, the four data processing cycles are explained here first. The first data processing cycle is the collection cycle of fresh commodity images, and the fresh commodity images are synchronously processed to analyze the corresponding overall freshness. The shorter the first data cycle, the higher the accuracy of monitoring the freshness of fresh commodities, and the greater the amount of calculation. The user balances the accuracy and amount of calculation according to his own needs to obtain the length of the first data processing cycle. The second data processing cycle is the price adjustment cycle. The price adjustment cycle has a minimum limit and is usually much larger than the first data processing cycle; this is because in supermarkets, if the price of commodities changes frequently, it is not conducive to the sale of commodities; users can set the price adjustment cycle in combination with the selling period, etc. The third data processing cycle is the update cycle of the electronic tag. This update cycle is different from the update of the electronic tag in the second data processing cycle. In the second data processing cycle, the update of the electronic tag is to update the price value of the specified commodity, while in the third data processing cycle, the display commodity type and commodity price display position of the electronic tag are updated, that is, the display content of the electronic tag is rearranged, but the price value is not updated. The fourth data processing cycle is the verification cycle, which generally occurs after each electronic tag update. The displayed data of the electronic tag is verified again to determine whether the update is successful.

[0070] 1. Freshness Judgment

[0071] See also Figure 2 ,The freshness evaluation model includes a feature extraction unit, a first feature fusion unit, a second feature fusion unit, an ,adaptive maximum pooling layer, a global average pooling layer and a ,fully connected layer in sequence.

[0072] The feature extraction unit includes three convolution modules, which perform layer-by-layer feature extraction on the imported fresh product images of the designated product placement area; the first feature fusion unit is used to extract the outputs of the second convolution module and the third convolution module, and perform feature fusion of different sizes; the second feature fusion unit is used to perform feature fusion again on the output of the first feature fusion unit and the output of the first convolution module; the features fused by the second feature fusion unit are subjected to adaptive maximum pooling and global average pooling processing by the adaptive maximum pooling layer and the global average pooling layer respectively, and the global features of the fresh product images are extracted; the fully connected layer is used to classify the extracted global features of the fresh product images, calculate the probability that the global features of the fresh product images belong to each freshness level, and output the freshness level of the current fresh product image.

[0073] The first feature fusion unit is used to fuse the middle-level feature values ​​output by the second convolution module and the deep-level feature values ​​output by the third convolution module to obtain the first fused feature, which is used to characterize the semantic information of fresh products. The second feature fusion unit is used to fuse the middle-level feature values ​​output by the second convolution module and the deep-level feature values ​​output by the third convolution module to obtain the first fused feature. The first convolution module focuses more on extracting the geometric detail features of fresh products. The second feature fusion unit is then used to perform feature fusion again on the first fused feature (which focuses more on the epidermal semantic features of fresh products) and the shallow feature values ​​output by the first convolution module (which focuses more on the geometric detail features of fresh products) to obtain the global features of the fresh product image. In addition, this embodiment judges the overall freshness of fresh products, while taking into account the accuracy and efficiency of the evaluation results, and only needs to use a three-layer convolutional network to execute, realizing the lightweight computing requirements of the fresh area.

[0074] The process of collecting training samples used by the freshness evaluation model includes:

[0075] For designated fresh products of different quantities and different freshness levels, stir them at intervals, and collect initial sample images in batches during the stirring intervals; enhance the initial sample images of each batch, and use multiple methods such as scaling, rotation, transformation, and flipping to expand the data set. For example, the freshness of a certain fresh product is divided into five levels: less than 20%, 20% to 40%, 40% to 60%, 60% to 80%, and 80% to 100%. For each freshness level, four quantities of 200, 100, 50, and 20 are set, totaling 20 kinds of original fresh product samples. Stir these 20 kinds of original fresh product samples, and collect images during the stirring intervals. This cycle is repeated to collect the initial sample image data. After enhancing the initial sample image data, it is expanded by scaling, rotation, transformation, and flipping, and finally 16,400 fresh image sample data of different resolutions can be obtained. In practical applications, the freshness level and quantity level can be set by yourself, and are not limited to the aforementioned one.

[0076] Considering the judgment efficiency, this embodiment proposes to judge the freshness of fresh products as a whole, rather than a single fresh product. In supermarkets, environmental factors such as lighting are usually consistent. Even if the ambient light of some products is reduced due to human body occlusion and other reasons, the influence of light intensity on freshness judgment can be ignored by just filtering the collected images accordingly. On this basis, the change in freshness of fresh products is related to two factors: the amount of water loss that changes over time, and the change in the freshness ratio of the remaining products caused by customer selection. For the former, the freshness change curve caused by water loss is relatively flat. For the latter, customers are more inclined to choose fresher products. Therefore, the behavior of customers buying products will cause the freshness of the remaining products to change suddenly. For example, for a fresh product, such as apples, since supermarket warehouses usually have a fresh-keeping area, we assume that the apples just placed in the product display area are fresh (from an overall perspective, not for individual apples). As time goes by, the moisture of individual apples will be lost, and since customers are more inclined to choose fruits with higher freshness, the overall freshness level of apples in the product display area continues to decline.

[0077] 2. Calculate and update prices with the goal of maximizing sales

[0078] The price updating method of this embodiment is based on the following shopping assumptions obtained from historical data analysis: the sales volume under the same freshness level tends to be stable, but is related to factors such as season, selling price and the number of similar commodities provided by the current supermarket, among which the season and the number of similar commodities provided by the current supermarket are uncontrollable external factors. In other words, when the external factors are stable, there is a relatively fixed relationship between the freshness level, price and sales volume of fresh commodities. The analysis process of external factors is quite complicated. If they are analyzed one by one, the amount of calculation will be very large. Therefore, in this embodiment, a relationship evaluation model based on a neural network is used to analyze the imported historical freshness level, price and sales volume, and the relationship function between the freshness level, price and sales volume of the corresponding fresh commodity is obtained through a classifier. Specifically, according to the preset price adjustment period T, the price J of fresh commodities in the current price adjustment period, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed, and the process of obtaining the relationship function S[J, X(t)] of the overall freshness and sales volume under the price J condition includes the following steps:

[0079] S41, according to different external factors, collect the overall freshness change curve, sales volume curve, and overall freshness-sales volume relationship curve of fresh products within the price range [J-ΔJ, J+ΔJ], generate several groups of sample data corresponding to price J, and construct a training set; ΔJ is the preset allowable price error of the fresh product.

[0080] S42, constructing a relationship evaluation model based on a neural network, and using a training set to train the relationship evaluation model until convergence.

[0081] S43, the price J of the fresh product in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are introduced into the relationship evaluation model, and the relationship function S[J, X(t)] between the overall freshness and the sales volume under the price J condition is obtained by classification, and the relationship function S[J, X(t)] satisfies the following basic conditions:

[0082]

[0083] Where: X T is the overall freshness at the beginning of historical period T, F1(t) is the freshness loss value caused by moisture loss, and F2[Y(t)] is the freshness loss value caused by sales volume Y(t).

[0084] This embodiment avoids the analysis of external factors, and obtains the relationship function between the overall freshness and sales volume under several groups of different price conditions by analyzing the historical data under different external factors. Then, the overall freshness, sales volume, and price collected in real time are analyzed to determine which external factor condition the relationship function under which the changing trend of the overall freshness and sales volume is more consistent with, thereby obtaining the corresponding relationship function.

[0085] After knowing the relationship function between the overall freshness and sales volume, the optimal price is calculated with the sales volume maximization as the optimization goal. Specifically, the relationship function S[J, X(t)] between the overall freshness and sales volume under the price J condition is imported into the trend analysis module, and the price range corresponding to the overall freshness X0 at the current moment is found. The process of estimating the change trend of the overall freshness and sales volume under different price conditions includes the following steps:

[0086] S51, a trend analysis module is constructed based on the GRU network model of the attention mechanism. The output of the trend analysis module is the historical freshness level data of the fresh products corresponding to the m sampling periods counted forward from the current moment, and the output is the estimated freshness level data of the m sampling periods counted backward from the current moment.

[0087] S52, searching and obtaining a corresponding price range according to the estimated freshness level data.

[0088] S53, calling the relationship evaluation model to estimate the change curve of the overall freshness and sales volume under different price conditions within the price range.

[0089] The ideal price J for the next price adjustment cycle is calculated according to the following formula: * :

[0090]

[0091] Y * (t) = S[J * , X * (t)]

[0092] X * (t) = X0-F1(t)-F1[Y * (t)]

[0093] Where: X0 is the overall freshness at the current moment, X * (t) is the estimated freshness change curve, Y * (t) is the estimated sales volume change curve.

[0094] Preferably, the trend analysis module is trained using the freshness grade sequence data of several fresh commodities corresponding to the relationship function S[J, X(t)] between the overall freshness and the sales volume and the freshness change sample data caused by different sales volumes under each freshness grade condition. According to the estimated freshness grade data of the m sampling periods counted backward from the current moment output by the trend analysis module, the corresponding price range is found, and an optimal price is selected from the price range to maximize the sales volume in the next price adjustment cycle. This is determined by the characteristics of the short sales cycle, fast new products and frequent replenishment of commodities in the fresh food area. Commodities in the fresh food area only need to focus on sales volume in the short term (for example, a price adjustment cycle). Different from the mode of price promotion based only on freshness, this embodiment comprehensively considers freshness and sales volume, uses sales volume as the judgment basis, and adjusts prices to maximize the merchant's profits without the need for frequent participation of professionals. For example, although the freshness level of a certain product has dropped to 40%, the freshness level will drop significantly in the next price adjustment cycle and the sales volume is in line with the merchant's psychological expectations. Customers' psychological acceptance of the product is still high. At this time, you can choose to maintain the original price or reduce the price slightly. On the contrary, if the freshness level of a certain product has dropped to 40% and the sales volume is very low and the freshness level continues to decline steadily in the next price adjustment cycle, it means that customers' psychological acceptance of the product is very low. In order to prevent the product from being unsalable due to the continuous decline in the freshness level, you can choose to reduce the price significantly.

[0095] For example, when the optimal price J is selected * When the corresponding sales volume is less than the preset sales volume, the staff is reminded to restock or adopt other manual promotion methods. This embodiment can reduce manual intervention, manage the commodity price with the optimization goal of maximizing sales volume by estimating the freshness level and commodity sales volume, and reduce the workload of manual inspection of commodities and manual updating of price tags (even updating electronic tags). The price range corresponding to the freshness level can be preset by the user.

[0096] 3. Display of price tags for multiple products

[0097] The method for controlling the display of multiple commodity price tags in this embodiment includes the following steps:

[0098] Step 1: Generate the correspondence between the electronic tag and the unit product area.

[0099] Due to the wide variety of goods in the fresh food area, a medium-height platform structure is usually adopted, and a dozen or even dozens of unit product areas are set up on one platform. The merchant sets an electronic tag of about 10 inches for each product and places it directly in front of the corresponding product to indicate the price. The user needs to match the product with the electronic tag one by one, and then move to view the price of the product. To this end, this embodiment proposes to display multiple product prices on one electronic tag at the same time. The display content of the electronic tag is affected by the user's visual range, that is, the electronic tag in each fresh food area must be able to display the prices of other products within a certain field of view of the user centered on the corresponding product. Compared with setting electronic tags corresponding to the number of products one by one according to the placement of the products, the effort required by the user to find the price is greatly reduced.

[0100] For example, the correspondence between the electronic tag and the unit product area can be preset, or the correspondence between the electronic tag and the unit product area can be automatically regenerated based on the product recognition result corresponding to the captured image. When the position of the electronic tag or the product moves, the consistency of the tag and the product can also be ensured. The staff only needs to pay attention to whether the distribution of the electronic tags can ensure that the total product types captured are complete (the viewing angle coverage of the electronic tags is complete enough).

[0101] Step 2: regularly capture images of shelves to which the unit product area corresponds to each electronic tag and within a preset field of view centered on the unit product area.

[0102] The collection of shelf images can be achieved through the shooting devices distributed in the fresh food area, or it can be combined with manual patrol shooting. If it is necessary to combine manual patrol shooting, the images taken by manual patrol can be calibrated again, or the correlation between them and the images taken at fixed points can be indirectly calculated, and then used for price update applications. In this embodiment, in addition to being executed when the goods are put on the shelves, price updates can also be reflected in the daily operation and maintenance of the supermarket. For example, if one of the goods on a shelf is sold out, the price tag can be adjusted in time through regular updates to avoid misunderstandings by users, which is especially suitable for the short-cycle sales characteristics of goods in the fresh food area.

[0103] Step 3, price update.

[0104] See also Figure 3 , the price update process includes the following steps:

[0105] The shelf image is preliminarily divided into unit product areas, and the similarity of the images of adjacent unit product areas is judged. The adjacent unit product areas with similarity higher than the similarity threshold are regarded as the same product placement area. The unit product areas on the shelf image are preliminarily integrated to generate several product placement areas, and each product placement area is marked. The product recognition model is then called to identify the product type in each product placement area.

[0106] Exemplarily, the process of identifying the commodity type in each commodity placement area is as follows:

[0107] Collect several product image samples contained in the supermarket to construct a training set. The product image samples include stacked image samples of multiple similar products. Construct a product recognition model based on a neural network, and use the training set to train the product recognition model. The product recognition model can adopt the image recognition technology in the prior art. For example, the invention with application number 202010874612.9 published on October 2, 2020 discloses a product recognition management method, which obtains the category and model information of the product by training and recognizing the image of the target product. Take the unit product area as the unit, preliminarily divide the shelf image, judge the similarity of the images of adjacent unit product areas, and regard the adjacent unit product areas with similarity higher than the similarity threshold as the same product placement area; preliminarily integrate the unit product areas on the shelf image, generate several product placement areas, and mark each product placement area. Import the image of each product placement area into the product recognition model, and identify the product type of each product placement area.

[0108] When one of the product placement areas contains more than two kinds of goods, the types of goods contained therein are matched with the types of goods in the adjacent product placement area. If all the types of goods are matched successfully, the product placement area is cleared, otherwise, the corresponding product placement area is generated according to the types of goods that are not matched successfully; the adjacent product placement area refers to the product placement area that belongs to the same electronic tag display coverage. This is because in a supermarket environment, due to the open shopping environment, it is often the case that users take the goods and put them in another area by mistake. In the prior art, for this situation, it is usually discovered through manual inspections or image recognition combined with database proofreading, and then the position of the goods is corrected by notifying the staff to re-place the goods. In this embodiment, it is not regarded as a simple misplaced product, but as a secondary product. Its price information is displayed simultaneously in the large-screen electronic tag, so as to increase the probability of users purchasing the product and reduce the workload of the staff to pick and place.

[0109] According to the frequency of occurrence of the product placement area, the product placement area corresponding to only one electronic tag is defined as the first placement area, and the remaining product placement areas are defined as the second placement areas. The main label of each second placement area is determined according to the image area ratio.

[0110] In this embodiment, in addition to the corresponding misplaced secondary selected commodities, the commodities on the second display area also include commodities in the same viewing direction that are less likely to attract the user's attention because of the small area occupied. These commodities occupy a small area in the viewing direction A due to the angle problem and can only be partially displayed, but they may be fully displayed in the viewing direction B (for example, an electronic tag is located directly in front of commodity 1. Therefore, when the shelf image is collected with the electronic tag orientation as the viewing center, commodity 1 will occupy a large area, making the electronic tag the main label of commodity 1). Of course, in some examples, the main label of the commodity can also be specified by binding the relationship between the label and the commodity in the background; but the confirmation process of the previous main label is more in line with the visual law of the user when shopping, especially when the commodity is displaced, the commodity is repeatedly placed in multiple unit product areas, or the label position is offset, the self-correction of the electronic label price display can still be achieved. Combined with the characteristics of multi-commodity display, on the basis of ensuring that the commodity price is fully displayed, it is ensured that customers can continuously and quickly obtain correct price information.

[0111] The product label image corresponding to each product placement area is retrieved, and the retrieved product label images are spliced ​​according to the positional relationship of each product placement area on the shelf image to generate an electronic label price tag; wherein, the label image of the second placement area other than the main label is scaled according to the area occupied by the second placement area on the shelf image.

[0112] The process of generating an electronic label price tag is as follows: for each electronic label, read the product placement area number and the product type on each product placement area contained in the current collection period, and retrieve the product label image data corresponding to each product placement area according to the product type; generate a label template according to the position information of the unit product area occupied by each product placement area, and the label template contains a number of label filling boxes, each label filling box corresponds to a product placement area; determine whether the electronic label corresponds to a second placement area other than the main label, and if so, scale the corresponding label filling box within a preset image size range according to the area occupied by the second placement area on the shelf image; fill the retrieved product label image data into the corresponding label filling box to generate an electronic label price tag within the current collection period.

[0113] After obtaining the correspondence between the electronic tag and the product placement area, the splicing process of the product price tag can be realized according to the positional relationship of the product placement area. The electronic tag price board of this embodiment is spliced ​​according to the positional relationship of the product, which can correctly guide customers to understand the product information; at the same time, the product label image in the electronic tag price board is related to the user's attention under the current viewing angle, which can reduce the effort required by the user to obtain price information and reduce the misjudgment rate.

[0114] In this embodiment, although the product label image is scaled, it is still within the reasonable range of human vision, and the size and number of the electronic labels corresponding to the shelf can be determined in combination with the type and quantity of the products on the shelf.

[0115] Finally, the spliced ​​electronic label price tag is sent to the corresponding electronic label for display.

[0116] In this embodiment, the purpose of setting the second display area is not only to enable the user to fully and quickly understand the prices of all commodities within the field of vision, but also has another important significance, that is, to serve as a verification standard for the update results.

[0117] At present, the verification of the price tag display results is still done manually. There are many types of goods in the fresh food area, and the probability of promotion and discount is also high. Even experienced staff find it difficult to remember the real-time prices of most goods. If the update is more frequent, the workload of the staff will increase exponentially. This is especially true for large-screen price tags that display prices of multiple goods. To this end, this embodiment proposes to perform consistency verification on the price tag image data of the second display area that is repeatedly displayed to achieve an estimation of the display results of the entire electronic tag.

[0118] See also Figure 4 The verification process includes: after the electronic tag is updated, the display image data of several electronic tags corresponding to any second placement area are retrieved, and the update result of the electronic tag is estimated by checking the similarity and scaling size of the product label image data corresponding to the second placement area in all retrieved display image data. Specifically, the following steps are included:

[0119] A71, combining the electronic tag and product placement area relationship data table, collects the display image data set of the electronic tags in the same second placement area a

[0120] A72, call the original collected shelf image data set corresponding to the above electronic tag According to the position information of the unit product area occupied by the second placement area a, the display image data set of the electronic tag is The corresponding image data set of the second placement area a is extracted from(k)}; k = 1, 2, ..., K.

[0121] A73, determine the image data set {a (k)} is complete. If complete, go to step A74. Otherwise, output the update failure result and end the process.

[0122] A74, calculate the second placement area a in the shelf image data set The scale ratio set of the label filling box of the second display area a is calculated by calculating the area occupied by each shelf image in

[0123] A75, according to the scale collection For the data image data set {a (k)} to obtain the scaled validation set {b (k)}, Then judge the validation set {b (k)} are consistent, if they are consistent, go to step A76, otherwise, output the update failure result and end the process.

[0124] A76, for the validation set {b (k)} and the similarity between the product price tag of the corresponding product on the second display area a. If the similarity is higher than the preset similarity threshold, the update success result is output; otherwise, the update failure result is output.

[0125] Take the four-sided cargo platform as an example. Assume that there are several commodities 1 on the cargo platform on the east side. Some commodities 1 can also be observed from the south and north aisles. The four-sided cargo platform includes three electronic tags, one of which faces the east and corresponds to commodity 1, while the other two face the southwest and northwest, corresponding to commodities 2 and 3. After analysis, electronic tag 1 facing the east is the main tag, and electronic tags 2 and 3 facing the southwest and northwest are secondary tags. Electronic tags 1, 2, and 3 process and display the commodity price image of commodity 1 according to the area occupied and position of commodity 1 on its corresponding shelf image. In electronic tag 2, the price image of commodity 1 is smaller and displayed to the right, and in electronic tag 3, the price image of commodity 1 is smaller and displayed to the left.

[0126] During the verification process, the complete display images of electronic tags 1, 2, and 3 are collected respectively, and the corresponding display area price images are intercepted from the set positions in the complete display images according to the electronic tag template. If the display area price image is incomplete, it means that the current electronic tag update has failed, and it is very likely that the updated image data has not been received; otherwise, the original shelf image is processed again, the scaling ratio is calculated, and the intercepted display area price image is scaled and then compared for consistency, and the calculation results, scaling ratio, and image data display of the original shelf image are verified. The original shelf image here refers to the data basis when the electronic tag is updated, not just the image when the product is put on the shelf. This is for the verification work when the electronic tag is updated during the product sales process.

[0127] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for controlling electronic tags in a fresh food area to maximize sales, which is used to automatically control the display content of electronic tags installed in the fresh food area, characterized in that: The control method comprises the following steps: S1, constructing a freshness evaluation model, wherein the freshness evaluation model is trained by using a number of stacked sample images of similar fresh products; S2, taking images of fresh products in the designated product placement area according to a preset collection cycle, importing the fresh product images into a freshness evaluation model, and evaluating the overall freshness of the fresh products; the fresh products in the designated product placement area are set with flexible prices, and different freshness levels correspond to different price ranges; S3, with time as the horizontal axis, fit the historical overall freshness data and corresponding sales data of fresh products evaluated in the n collection cycles from the current moment forward, and obtain the historical overall freshness change curve X(t) and sales volume curve Y(t) within the preset price adjustment cycle T; S4, according to the preset price adjustment cycle T, the price J of fresh products in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed to obtain the relationship function S[J,X(T)] between the overall freshness and sales volume under the price J condition: Where: X T is the overall freshness at the beginning of the historical period T, F1(t) is the freshness loss value caused by water loss, and F2[Y(t)] is the freshness loss value caused by sales volume Y(t); S5, import the relationship function S[J,X(t)] between the overall freshness and sales volume under the price J condition into the trend analysis module, find the price range corresponding to the next price adjustment cycle T, estimate the change curve of the overall freshness and sales volume under different price conditions within the price range, and calculate the ideal price J for the next price adjustment cycle according to the following formula * : Y * (T)=S[J * ,X * (t)] X * (t)=X0-f1(t)-F2[Y * (t)] Where: X0 is the overall freshness at the current moment, X * (t) is the estimated freshness change curve, Y * (t) is the estimated sales volume change curve; In step S5, the relationship function S[J, X(t)] between the overall freshness and the sales volume under the price J condition is imported into the trend analysis module, and the price range corresponding to the overall freshness X0 at the current moment is found. The process of estimating the change trend of the overall freshness and sales volume under different price conditions includes the following steps: The GRU network model based on the attention mechanism is used to construct a trend analysis module. The output of this trend analysis module is the historical freshness level data of fresh products corresponding to the m sampling periods counted forward from the current moment, and the output is the estimated freshness level data of the m sampling periods counted backward from the current moment; Find the corresponding price range based on the estimated freshness level data; The call relationship evaluation model is used to estimate the change curves of overall freshness and sales volume under different price conditions within the price range.

2. The method for managing and controlling the electronic tags in the fresh food area to maximize sales according to claim 1, characterized in that: In step S1, the freshness evaluation model includes a feature extraction unit, a first feature fusion unit, a second feature fusion unit, an adaptive maximum pooling layer, a global average pooling layer and a fully connected layer connected in sequence; The feature extraction unit includes three convolution modules, which perform layer-by-layer feature extraction on the imported fresh product images of the designated product placement area; the first feature fusion unit is used to extract the outputs of the second convolution module and the third convolution module, and perform feature fusion of different sizes; the second feature fusion unit is used to perform feature fusion again on the output of the first feature fusion unit and the output of the first convolution module; The features fused by the second feature fusion unit are subjected to adaptive maximum pooling and global average pooling processing by the adaptive maximum pooling layer and the global average pooling layer respectively, so as to extract the global features of the fresh product image; the fully connected layer is used to classify the extracted global features of the fresh product image, calculate the probability that the global features of the fresh product image belong to each freshness level, and output the freshness level of the current fresh product image.

3. The method for managing and controlling the electronic tags in the fresh food area to maximize sales according to claim 1, characterized in that: In step S4, the price J of fresh products in the current price adjustment period, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are analyzed according to the preset price adjustment period T to obtain the relationship function S[J,X(t)] between the overall freshness and the sales volume under the price J condition, which includes the following steps: According to different external factors, the overall freshness change curve, sales volume curve, and overall freshness-sales volume relationship curve of fresh products within the price range [J-ΔJ, J+ΔJ] are collected to generate several groups of sample data corresponding to price J to construct a training set; ΔJ is the preset allowable price error of the fresh product; A relationship evaluation model is constructed based on a neural network, and the relationship evaluation model is trained using a training set until convergence; The price J of fresh products in the current price adjustment cycle, the historical overall freshness change curve X(t) and the sales volume curve Y(t) are introduced into the relationship evaluation model, and the relationship function S[J,X(t)] between the overall freshness and sales volume under the price J condition is classified.

4. The method for managing and controlling the electronic tags in the fresh food area for maximizing sales according to any one of claims 1 to 3, characterized in that: The control method also The following steps are involved: A1, generate the correspondence between the electronic tag and the unit product area; A2, regularly photographing the shelf image of the unit product area corresponding to each electronic tag within a preset field of view centered on the unit product area of ​​the electronic tag; A3, preliminarily divide the shelf image into unit product areas, perform similarity judgment on the images of adjacent unit product areas, regard adjacent unit product areas with similarity higher than the similarity threshold as the same product placement area, preliminarily integrate the unit product areas on the shelf image, generate several product placement areas, and mark each product placement area; then call the product recognition model to identify the product type of each product placement area; A4, according to the frequency of occurrence of the product placement area, the product placement area corresponding to only one electronic tag is defined as the first placement area, and the remaining product placement areas are defined as the second placement areas, and the main label of each second placement area is determined according to the image area ratio; A5, retrieve the product label image corresponding to each product placement area, splice the retrieved product label images according to the positional relationship of each product placement area on the shelf image, and generate an electronic label price tag; wherein the label image of the second placement area other than the main label is scaled according to the area occupied by the second placement area on the shelf image; A6, sends the generated electronic label price tag to the corresponding electronic label for display.

5. The method for managing and controlling the electronic tags in the fresh food area to maximize sales according to claim 4, characterized in that: In step A4, when one of the product placement areas contains more than two products, the product types it contains are matched with the product types of the adjacent product placement areas. If all product types are matched successfully, the product placement area is cleared, otherwise, a corresponding product placement area is generated according to the product types that failed to match; adjacent product placement areas refer to product placement areas that are within the coverage of the same electronic tag display.

6. The method for managing and controlling the electronic tags in the fresh food area to maximize sales according to claim 4, characterized in that: In step A5, the process of generating the electronic price tag includes the following steps: A51, for each electronic tag, read the number of the product placement area and the product type on each product placement area in the current verification cycle, and retrieve the product label image data corresponding to each product placement area according to the product type; A52, generating a label template according to the position information of the unit product area occupied by each product placement area, wherein the label template includes a plurality of label filling boxes, and each label filling box corresponds to a product placement area; A53, determining whether the electronic label corresponds to a second placement area other than the main label, and if so, calling a scaling component to scale the corresponding label filling frame within a preset image size range according to the area occupied by the second placement area on the shelf image; A54, fill the retrieved product label image data into the corresponding label filling box to generate the electronic label price tag within the current collection cycle.

7. The method for managing and controlling the electronic tags in the fresh food area to maximize sales according to claim 4, characterized in that: The control method further comprises the following steps: After the electronic tag is updated, the display image data of several electronic tags corresponding to any second placement area are retrieved, and the update result of this electronic tag is estimated by checking the similarity and scaling size of the product label image data corresponding to the second placement area in all retrieved display image data.

Images