Commodity loss prevention method applied to shopping malls and supermarkets

Abstract

The invention relates to a commodity loss prevention method applied to shopping malls and supermarkets, which belongs to the technical field of commodity loss prevention and comprises the following operation steps of: 1, acquiring data required by data model construction and data model prediction by adopting a data interface; 2, constructing a data model by using the collected data; and 3, in the constructed data model, outputting a mark which corresponds to each identification record and indicates whether the identification record is a high-price and low-price mark, transmitting the records to a corresponding display module, and notifying a customer to check the records. And 4, updating the data model: after the business of the day is finished, taking the day as a benchmark, pushing forward for a week, and reconstructing the data model for detection and judgment of the next day. And 5, counting the proportion of each commodity according to the high-price and low-price record of the week, outputting an analysis report, and giving the analysis report to the customer. The problem of high price and low price in the supermarket non-standard product weighing process is solved, and meanwhile, supermarket customers are helped to check and use simply and conveniently.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity loss prevention, and particularly relates to a commodity loss prevention method applied to shopping malls and supermarkets. Background Art

[0002] During the weighing and pricing process of non-standard products in supermarkets, there is a general reliance on manual supervision and management, or on some surveillance cameras for deterrence, and the following problems exist: (1) The supervision efficiency is low, and it is impossible to cover all weighing operations in real time, and it is quite time-consuming to trace.

[0003] (2) It is difficult to distinguish whether it is a human operation error or an intentional low-price entry.

[0004] (3) The management cost is high, and it is easy to cause disputes between customers and supermarkets.

[0005] Currently, in the market, efforts have been made to address the problems of high loss costs and great supervision difficulties caused by unintentional or malicious high-price under-recording during the weighing process of non-standard products in supermarkets: (1) Video surveillance system, installing cameras in the weighing area to record and monitor the weighing process, and then manually reviewing the videos afterwards to discover problems; however, this method has the following defects. The monitoring is only a passive means, unable to detect and prevent abnormal behaviors in real time, and the manual review cost is high and the efficiency is low.

[0006] (2) Image recognition plus weighing solution, introducing image recognition technology into weighing equipment, collecting product images through cameras, and using machine learning algorithms to identify product categories. Its defects are that the functions are relatively single, mostly auxiliary tools, only used for recommending product labels, unable to judge whether the label selected by the customer matches the product, and at the same time lacking means for analyzing historical data, making it difficult to accurately identify malicious behaviors. Summary of the Invention

[0007] The present invention mainly addresses the deficiencies existing in the prior art, and provides a commodity loss prevention method applied to shopping malls and supermarkets. It solves the problem of high-price under-recording during the weighing process of non-standard products in supermarkets. Through data mining algorithms, it analyzes the weighing recognition behaviors of supermarkets, finds out abnormal weighing behaviors; and associates them with the equipment to locate the specific weighing personnel; at the same time, it provides data-based management to help supermarket customers view and use simply and conveniently.

[0008] The above technical problems of the present invention are mainly solved by the following technical solutions: A commodity loss prevention method applied to shopping malls and supermarkets includes the following operating steps: Step 1: The data for weighing and identification in the store is generated by the identification algorithm running on the front-end software and automatically uploaded. It is automatically stored on the company's server through the network, and the data required for data model construction and data model prediction is obtained using a data interface.

[0009] Step 2: In order to detect abnormal behaviors of overpricing and undercharging, a data model is constructed using the collected data.

[0010] Step 3: The relevant information of the identification records for the current day, including the goods and the identification results, and the output results are passed into the constructed data model to output whether each identification record is a sign of overpricing and undercharging. These records are then transmitted to the corresponding display module and the customer is notified to view.

[0011] Step 4: Update the data model. After the business for the current day ends, based on the current day as a benchmark, push back one week and reconstruct the data model for the detection and judgment of the next day.

[0012] Step 5: For the overpricing and undercharging records of this week every week, count the proportion of each commodity and output an analysis report to the customer.

[0013] Preferably, the data required for data model construction includes: (1) The commodity identification records of the recent week, each record including the commodity identified this time and the commodity category output by the deep learning model; (2) A mapping relationship table of fresh food keywords and commodity categories extracted based on the commodity names sold in all current supermarkets and the recognition results of the deep learning model; (3) Some filtering words statistically obtained based on the commodities sold in all current supermarkets, used to screen some abnormal commodity lists; (4) Commodity price correspondence.

[0014] Preferably, the data required for data model prediction includes the commodity identification records of the current day, each record including the commodity identified this time, the commodity category output by the deep learning, and the corresponding commodity output by the recommendation algorithm.

[0015] Preferably, based on the commodity identification records of one week, count the quantity and average score of each recognition result corresponding to each commodity; set a threshold for both the quantity and the average score. For each commodity, select all the commodity categories output by the deep learning model whose quantity is above the threshold or the average score is above the threshold as the label list of the current commodity, and finally obtain such a mapping relationship.

[0016] Preferably, use the commodity category mapping relationship table, the fresh food keyword category mapping table, the abnormal commodity filtering word list, and the commodity price mapping relationship to construct a loss prevention judgment logic; for each identification record to be predicted, first judge whether the commodity of this record is in the output result of the recommendation algorithm.

[0017] Preferably, if not, use the text matching algorithm to determine whether the name of the product in this record is in the filtering word list; if not, obtain the product category output by the deep learning model corresponding to this record; obtain the product category of the product selected in this record in the product category mapping relationship table, and use the text matching algorithm to obtain the keywords corresponding to the product name in this record in the fresh food keyword category mapping table, so as to determine the product category corresponding to its name.

[0018] Preferably, determine whether the product category of this record is the same as the product category of the selected product in the product category mapping relationship table, or whether it is the same as the product category corresponding to the selected product in the fresh food keyword category mapping table. If they are all inconsistent, it is considered an abnormal weighing; if it belongs to abnormal weighing, compare the price corresponding to the selected product in this record with the price of the first product in the output result of the recommendation algorithm. If the price of the selected product is less than the price of the first product output by the recommendation algorithm, it belongs to the behavior of high price with low beating, and an alarm is given.

[0019] The present invention can achieve the following effects: The present invention provides a product anti-loss method applied to shopping malls and supermarkets. Compared with the prior art, it solves the problem of high price with low beating in the weighing process of non-standard products in supermarkets. Through data mining algorithms, it analyzes the weighing recognition behaviors of supermarkets, finds out abnormal weighing behaviors; and associates them with equipment to locate specific weighing personnel; at the same time, it provides digital management to help supermarket customers view and use simply and conveniently. Detailed implementation manners

[0020] The following further specifically describes the technical solutions of the invention through embodiments.

[0021] Embodiment: A product anti-loss method applied to shopping malls and supermarkets includes the following operation steps: First step: The data for weighing recognition in the store is generated by the recognition algorithm running on the front-end software and automatically uploaded, and automatically stored on the company's server through the network. The data required for data model construction and data model prediction is obtained through a data interface.

[0022] The data required for data model construction includes (1) product recognition records in the recent week, each record containing the product recognized this time and the product category output by the deep learning model; (2) a mapping relationship table of fresh food keywords and product categories extracted based on the product names sold in all current supermarkets and the recognition results of the deep learning model; (3) some filtering words statistically obtained based on the products sold in all current supermarkets, used to screen some abnormal product lists; (4) product price correspondence.

[0023] The data required for data model prediction includes the commodity identification records of the current day. Each record contains the commodity identified this time, as well as the commodity categories output by the deep learning and the corresponding commodities output by the recommendation algorithm.

[0024] Step 2: To detect abnormal behavior of underpricing high-priced goods, use the collected data to build a data model.

[0025] Based on the commodity identification records of one week, count the quantity of each recognition result corresponding to each commodity and the average score; set a threshold for both the quantity and the average score. For each commodity, select all the commodity categories output by the deep learning model whose quantity is above the threshold or the average score is above the threshold as the label list of the current commodity. Finally, obtain such a mapping relationship.

[0026] Use the commodity category mapping relationship table, the fresh food keyword category mapping table, the list of abnormal commodity filtering words, and the commodity price mapping relationship to construct the anti-loss judgment logic; for each recognition record to be predicted, first judge whether the commodity in this record is in the output result of the recommendation algorithm.

[0027] If not, use the text matching algorithm to judge whether the name of the commodity in this record is in the filtering word list; if not, obtain the commodity category output by the deep learning model corresponding to this record; obtain the commodity category of the commodity selected in this record in the commodity category mapping relationship table, and use the text matching algorithm to obtain the keyword corresponding to the name of the commodity in this record in the fresh food keyword category mapping table, so as to determine the commodity category corresponding to its name.

[0028] Judge whether the commodity category of this record is consistent with the commodity category of the selected commodity in the commodity category mapping relationship table, or whether it is consistent with the commodity category corresponding to the selected commodity in the fresh food keyword category mapping table. If all are inconsistent, it is considered abnormal weighing; if it belongs to abnormal weighing, compare the price of the commodity selected in this record with the price of the first commodity in the output result of the recommendation algorithm through the commodity price mapping relationship. If the price of the selected commodity is less than the price of the first commodity output by the recommendation algorithm, it belongs to the behavior of underpricing high-priced goods and an alarm is issued.

[0029] Step 3: Input the relevant information of the recognition records of the current day, including the commodity and the recognition result, and the output result into the built data model, output the flag indicating whether each recognition record is underpricing high-priced goods, and transmit these records to the corresponding display module and notify the customer to view.

[0030] Step 4: Update the data model. After the business of the current day ends, based on the current day, push back one week and rebuild the data model for the detection and judgment of the next day.

[0031] Step 5: For the records of under-ringing at high prices in the current week, count the proportion of each SKU on a weekly basis, and output an analysis report to the customer.

[0032] Not only recommend products, but also judge whether there is price abnormality by comparing the image recognition results with the customer selection results. Continuously optimize the data model using historical data to improve the accuracy of loss prevention. Analyze which products are high-loss products and provide the basis for relevant operation decisions to the customer.

[0033] In summary, the commodity loss prevention method applied to shopping malls and supermarkets solves the problem of under-ringing at high prices during the weighing process of non-standard products in supermarkets. Through data mining algorithms, analyze the weighing and recognition behaviors in supermarkets to find abnormal weighing behaviors; associate them with the equipment to locate the specific weighing personnel; at the same time, provide data-based management to help supermarket customers view and use them simply and conveniently.

[0034] The above are only specific embodiments of the present invention, but the structural features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art within the scope of the present invention are covered by the patent scope of the present invention.

Claims

1. A method for preventing commodity loss in shopping malls and supermarkets, characterized in that The steps are as follows: Step 1: The store's weighing and identification data is generated by the identification algorithm running on the front-end software and automatically uploaded. It is automatically stored on the company's server through the network, and the data interface is used to obtain the data required for data model construction and data model prediction; Step 2: In order to detect abnormal behavior of high-price low-play, a data model is constructed using the collected data; Step 3: The information related to the day's recognition records, including the products and recognition results, is output and passed into the constructed data model. The flag of whether each recognition record is a high-price low-priced product is output, and these records are transmitted to the corresponding display module, and the customer is notified to view them; Step 4: Update the data model. After the business is over, take the current day as the benchmark, push forward one week, and rebuild the data model for the next day's detection and judgment. Step 5: Record the high and low prices of the week, calculate the proportion of each product, and output an analysis report to the customer.

2. The commodity loss prevention method used in shopping malls and supermarkets according to claim 1 is characterized in that: The data required for data model construction includes (1) product recognition records for the past week, where each record contains the product recognized this time and the product category output by the deep learning model; (2) A mapping table of fresh food keywords and product categories extracted based on the product names currently sold in all supermarkets and the recognition results of the deep learning model; (3) Some filter words based on the statistics of all the products currently sold in supermarkets are used to filter out some abnormal product lists; (4) The corresponding relationship between product prices.

3. The commodity loss prevention method used in shopping malls and supermarkets according to claim 1 is characterized in that: The data required for data model prediction includes the commodity recognition records of the day. Each record contains the commodity recognized this time, as well as the commodity category output by deep learning and the corresponding commodity output by the recommendation algorithm.

4. The commodity loss prevention method used in shopping malls and supermarkets according to claim 1 is characterized in that: Based on a week's worth of product recognition records, the number of each recognition result and the average score for each product are counted. A threshold is set for the number and the average score. For each product, all product categories output by the deep learning model whose number is above the threshold or whose average score is above the threshold are selected as the label list of the current product. Finally, a mapping relationship is obtained.

5. The commodity loss prevention method used in shopping malls and supermarkets according to claim 4 is characterized in that: The loss prevention judgment logic is constructed using the product category mapping relationship table, the fresh food keyword category mapping table, the abnormal product filtering word list, and the product price mapping relationship; for each identification record to be predicted, first determine whether the product in the record is in the output result of the recommendation algorithm.

6. The method for preventing loss of goods in shopping malls and supermarkets according to claim 5, characterized in that: If not, use the text matching algorithm to determine whether the name of the product in the record is in the filter word list; If not, obtain the product category output by the deep learning model corresponding to the record; obtain the product category of the product selected in the record in the product category mapping relationship table, and use the text matching algorithm to obtain the keyword corresponding to the product name of the record in the fresh food keyword category mapping table, so as to determine the product category corresponding to its name.

7. The commodity loss prevention method used in shopping malls and supermarkets according to claim 6 is characterized in that: Determine whether the product category of the record is consistent with the product category of the selected product in the product category mapping relationship table, or whether it is consistent with the corresponding product category of the selected product in the fresh food keyword category mapping table. If all are inconsistent, it is considered to be abnormal weighing; If it is an abnormal weighing, the price corresponding to the selected product in the record and the price corresponding to the first product in the recommendation algorithm output result are compared through the product price mapping relationship. If the price of the selected product is lower than the price of the first product output by the recommendation algorithm, it is a high-price-low-price behavior and an alarm is issued.