A digital product return identification method and system for an intelligent AI vending machine

Abstract

The present invention discloses a method and system for identifying digital product returns of an intelligent AI vending machine, the method comprising: determining product information of a returned product through RFID identification, obtaining a first detection image; retrieving a three-dimensional image model of the returned product from a database; selecting an abnormal area of ​​the first detection image through a preset product abnormality detection model, and determining the difference type, first difference score and three-dimensional coordinate data of the abnormal area; analyzing the three-dimensional coordinate data of the abnormal area under different shooting angles, determining the first abnormal area and the second abnormal area; calculating the final difference score of the first abnormal area, combining the first difference score of the second abnormal area, and determining the product difference score of the returned product; when the product difference score is greater than the preset difference score threshold, the returned product does not meet the return condition; otherwise, the returned product meets the return condition. The present invention can improve the efficiency of identifying packaging problems of returned products.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and more specifically, to a method and system for identifying digital product returns of an intelligent AI vending machine. Background Art

[0002] With the rapid development of e-commerce and unmanned retail, smart vending machines have become an important channel for consumers to purchase daily necessities and electronic products. However, in the process of returning digital products, how to efficiently and accurately determine whether the returned products are damaged, contaminated, etc. has always been a problem that plagues operators and consumers. The traditional return quality inspection method requires manual participation, which is not only inefficient, but also easily affected by human factors, resulting in inaccurate inspection results.

[0003] Therefore, the prior art has defects and is in urgent need of improvement. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a digital product return identification method and system for an intelligent AI vending machine. By presetting a product anomaly detection model, it can efficiently and accurately identify problems such as packaging damage and contamination of returned products, thereby improving the efficiency of digital product returns.

[0005] The first aspect of the present invention provides a digital product return identification method for an intelligent AI vending machine, comprising:

[0006] Perform RFID identification on the returned product to determine the product information of the returned product;

[0007] When the returned product is successfully identified, a first detection image at multiple shooting angles is acquired by a detection device;

[0008] Retrieving the three-dimensional image model of the returned product from the database according to the product information of the returned product;

[0009] Based on the three-dimensional image model, the first detection images of the multiple shooting angles are analyzed by a preset product abnormality detection model, an abnormal area of ​​the first detection image is framed by a first detection frame, and a difference type, a first difference score and three-dimensional coordinate data of the abnormal area are determined;

[0010] Analyze the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area; intercept the second detection image corresponding to the first abnormal area from the first detection image corresponding to the first abnormal area;

[0011] Calculate the final difference score of the first abnormal area, and determine the product difference score of the returned product by combining the first difference score of the second abnormal area;

[0012] When the product difference score is greater than the preset difference score threshold, the return product does not meet the return conditions;

[0013] When the product difference score is less than or equal to a preset difference score threshold, the returned product meets the return conditions and the returned product is recycled.

[0014] In this solution, the RFID identification of the returned product to determine the product information of the returned product includes:

[0015] Get the response data of the RFID tag;

[0016] Comparing the response data with the product information in the database, and determining the product information identical to the response data as the product information of the returned product;

[0017] When the product information identical to the response data does not exist in the database, the return product identification fails.

[0018] In this solution, based on the three-dimensional image model, the first detection images of the multiple shooting angles are analyzed by a preset product abnormality detection model, the abnormal area of ​​the first detection image is framed by a first detection frame, and the difference type, the first difference score and the three-dimensional coordinate data of the abnormal area are determined, including:

[0019] According to the shooting angle of the first detection image, a standard image with the same shooting angle is obtained through the three-dimensional image model of the returned product;

[0020] Analyze the first detection image and the corresponding standard image at the same shooting angle through a preset product abnormality detection model, select the abnormal area of ​​the first detection image, and determine the difference type and the first difference score of the abnormal area; determine the candidate frame of the abnormal area as the first detection frame;

[0021] The coordinate data of the first detection frame is determined based on the three-dimensional image model of the returned product, and the center coordinate data of the first detection frame is determined as the three-dimensional coordinate data of the corresponding abnormal area.

[0022] In this solution, the analysis of the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area includes:

[0023] Calculate the three-dimensional coordinate distance of abnormal areas of the same difference type under different shooting angles through the three-dimensional coordinate data of the abnormal area;

[0024] Determine abnormal areas of the same difference type whose three-dimensional coordinate distance is less than a preset coordinate distance threshold as the same abnormal area;

[0025] intercepting a second detection image of the same abnormal area from the first detection image of the same abnormal area;

[0026] An abnormal area where a plurality of second detection images exist is determined as a first abnormal area, and other abnormal areas other than the first abnormal area are determined as second abnormal areas.

[0027] This plan also includes:

[0028] Determine a recommended viewing angle of the first abnormal area by using the three-dimensional image model of the returned product;

[0029] Performing image changes on a plurality of second detection images of the first abnormal area based on the recommended viewing angle;

[0030] superimposing the first detection frame after image changes are performed on the multiple second detection images, and determining the formed detection frame as the second detection frame;

[0031] Segmenting the second detection frame into a plurality of segmentation areas based on a preset segmentation specification;

[0032] Each second detection image is divided into a plurality of sub-images based on the plurality of divided areas, and coordinate data of each sub-image is determined according to central coordinate data of the divided areas.

[0033] In this solution, calculating the final difference score of the first abnormal area includes:

[0034] According to the first abnormal area a m In each second detection image b n The first difference score P 1 (a m b n ) determines the first abnormal area a m In each second detection image b n The first influence weight k 1 (a m b n );

[0035] According to the first abnormal area a m Sub-image C i In the second detection image b n The segmented area c j The area ratio of the second detection image b is determined by n The segmented area c j The second influence weight k 2 (C i b n c j );

[0036] Determine the first abnormal area a by using the preset product abnormality detection model m The second detection image b n Sub-image C i The first difference score P 1 (a m b n C i );

[0037] According to the first abnormal area a m The second detection image b n Sub-image C i The first difference score P 1 (a m b n C i ), the first abnormal area a m In the second detection image b n The first influence weight k 1 (a m b n ) and the second detection image b n The segmented area c j The second influence weight k 2 (C i b n c j ) Calculate the second detection image b n The segmented area c j The second difference score P 2 (C i b n c j );

[0038] ;

[0039] Calculate each second detection image b according to the coordinate data of the sub-image n The segmented area c j The second difference score P 2 (C i b n c j ) to determine the segmentation area c j The third difference score P 3 (c j );

[0040] For all segmented regions c j The third difference score P 3 (c j ) are accumulated to determine the first abnormal area a m The final difference score P 4 (a m).

[0041] In this solution, the step of combining the first difference score of the second abnormal area to determine the product difference score of the returned product includes:

[0042] determining an influence weight of each first abnormal region and second abnormal region based on the difference type;

[0043] The final difference score of each first abnormal area and the first difference score of the second abnormal area are respectively multiplied by the corresponding influence weight, and the calculation results are accumulated to determine the product difference score of the returned product.

[0044] This plan also includes:

[0045] When the returned product is successfully returned, the three-dimensional image model of the returned product is updated based on the first detection images at the multiple shooting angles.

[0046] The second aspect of the present invention provides a digital product return identification system for an intelligent AI vending machine, comprising:

[0047] A product information acquisition module, used to perform RFID identification on the returned product and determine the product information of the returned product;

[0048] A detection image module, used for acquiring first detection images at multiple shooting angles through a detection device when the returned product is successfully identified;

[0049] A first analysis module is configured to retrieve a three-dimensional image model of the returned product from a database according to the product information of the returned product; based on the three-dimensional image model, analyze the first detection images of the multiple shooting angles by using a preset product anomaly detection model, select an abnormal area of ​​the first detection image by using a first detection frame, and determine a difference type, a first difference score, and three-dimensional coordinate data of the abnormal area;

[0050] A second analysis module is used to analyze the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area; intercept the second detection image corresponding to the first abnormal area from the first detection image corresponding to the first abnormal area; calculate the final difference score of the first abnormal area, and determine the product difference score of the returned product in combination with the first difference score of the second abnormal area;

[0051] The product verification module is used to determine that when the product difference score is greater than a preset difference score threshold, the returned product does not meet the return conditions; when the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions and the returned product is recycled.

[0052] A third aspect of the present invention provides a computer-readable storage medium, which includes a digital product return identification method program for an intelligent AI vending machine. When the digital product return identification method program for an intelligent AI vending machine is executed by a processor, the steps of the digital product return identification method for an intelligent AI vending machine as described above are implemented.

[0053] The present invention discloses a method and system for identifying digital product returns of an intelligent AI vending machine, the method comprising: determining product information of a returned product through RFID identification, obtaining a first detection image; retrieving a three-dimensional image model of the returned product from a database; selecting an abnormal area of ​​the first detection image through a preset product abnormality detection model, and determining the difference type, first difference score and three-dimensional coordinate data of the abnormal area; analyzing the three-dimensional coordinate data of the abnormal area under different shooting angles, determining the first abnormal area and the second abnormal area; calculating the final difference score of the first abnormal area, combining the first difference score of the second abnormal area, and determining the product difference score of the returned product; when the product difference score is greater than the preset difference score threshold, the returned product does not meet the return condition; otherwise, the returned product meets the return condition. The present invention can improve the efficiency of identifying packaging problems of returned products. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flow chart of a digital product return identification method of an intelligent AI vending machine provided by the present invention is shown;

[0055] Figure 2 A flow chart of the RFID identification method for returned products provided by the present invention is shown;

[0056] Figure 3 A flow chart showing a method for determining a first abnormal region and a second abnormal region provided by the present invention is shown;

[0057] Figure 4 A block diagram of a digital product return identification system for an intelligent AI vending machine provided by the present invention is shown. DETAILED DESCRIPTION

[0058] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0060] Figure 1 A flow chart of a digital product return identification method for an intelligent AI vending machine provided by the present invention is shown.

[0061] like Figure 1 As shown, the present invention discloses a digital product return identification method of an intelligent AI vending machine, comprising:

[0062] S102, performing RFID identification on the returned product to determine the product information of the returned product;

[0063] S104, when the returned product is successfully identified, a first detection image at multiple shooting angles is acquired by a detection device;

[0064] S106, retrieving a three-dimensional image model of the returned product from a database according to the product information of the returned product;

[0065] S108, based on the three-dimensional image model, analyzing the first detection images from multiple shooting angles by using a preset product abnormality detection model, selecting an abnormal area of ​​the first detection image by using a first detection frame, and determining a difference type, a first difference score, and three-dimensional coordinate data of the abnormal area;

[0066] S110, analyzing the three-dimensional coordinate data of the abnormal area at different shooting angles to determine a first abnormal area and a second abnormal area; intercepting a second detection image corresponding to the first abnormal area from a first detection image corresponding to the first abnormal area;

[0067] S112, calculating the final difference score of the first abnormal area, and combining it with the first difference score of the second abnormal area to determine the product difference score of the returned product;

[0068] S114, when the product difference score is greater than the preset difference score threshold, the returned product does not meet the return conditions;

[0069] S116, when the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions and the returned product is recycled.

[0070] According to an embodiment of the present invention, when the intelligent AI vending machine detects the existence of an object in the return window, the RFID tag of the object is identified by the RFID reader, and the RFID tag is compared with the product information of the digital product stored in the database to determine whether it is a digital product sold by the intelligent AI vending machine. When the identification is successful, the product information of the returned product is retrieved from the database through the RFID tag of the returned product, and the detection device in the intelligent AI vending machine is controlled to obtain the first detection image of the returned product at multiple shooting angles according to the specified shooting angle. The three-dimensional image model of the returned product is retrieved from the database according to the product information of the returned product, and the standard image of the same shooting angle is obtained through the three-dimensional image model of the returned product according to the shooting angle of the first detection image. The first detection image and the corresponding standard image at the same shooting angle are analyzed by the preset product abnormality detection model, and the area in the first detection image with different image features from the corresponding standard image is framed and determined as an abnormal area. The detection image of the abnormal area is compared with the historical detection image of the digital product, and the difference type and the first difference score of the abnormal area are determined according to the image features, and the center coordinate data of the corresponding first detection frame is determined as the three-dimensional coordinate data of the abnormal area. The abnormal areas of multiple shooting angles are classified according to the difference type, and the three-dimensional coordinate distances of the abnormal areas of the same difference type at different shooting angles are calculated. The abnormal areas of the same difference type whose three-dimensional coordinate distances are less than a preset coordinate distance threshold are marked as the same abnormal area, and the second detection image of the same abnormal area is intercepted from the first detection image at the corresponding shooting angle. When the analysis of all abnormal areas is completed, the abnormal area with multiple second detection images is determined as the first abnormal area, and the other abnormal areas other than the first abnormal area are determined as the second abnormal area.

[0071] Based on the difference type of the first abnormal area, the recommended viewing angle that is convenient for reflecting the image features of the first abnormal area is determined through the three-dimensional image model of the returned product. Based on the recommended viewing angle, multiple second detection images of the first abnormal area are subjected to image changes (such as stretching, rotation, etc.), and the first detection frames of the transformed multiple second detection images are superimposed to determine the second detection frame of the first abnormal area under the recommended viewing angle. The second detection frame is divided into multiple segmented areas according to the preset segmentation specification, and each second detection image of the first abnormal area is segmented to determine multiple sub-images. According to the second difference score of each segmented area in each second detection image of the first abnormal area, the average value of the second difference scores of the segmented areas of all the second detection images is calculated, and the third difference score of the segmented area is determined. The third difference scores of all the segmented areas are accumulated to determine the final difference score of the first abnormal area. The product difference score of the returned product is determined by weighted calculation of the final difference score of each first abnormal area in the returned product and the first difference score of the second abnormal area. The product difference score of the returned product is compared with the preset difference score threshold. When the product difference score is greater than the preset difference score threshold, the returned product does not meet the return conditions, and the user is reminded to take the returned product away through voice or other means; when the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions and the returned product is recycled.

[0072] Among them, the preset difference score threshold is determined by the system based on the product difference score between the detection image of the digital product when it enters the warehouse and the standard product. The larger the product difference score with the standard product, the smaller the corresponding preset difference score threshold.

[0073] In addition, the system will evaluate the product difference score of the digital product after each return, predict whether the degree of damage to the packaging of the digital product will affect the sale of the product when the customer takes it out for inspection normally during the next sale, and notify the vending machine manager in advance to handle digital products that pose a sales risk.

[0074] Figure 2 A flow chart of the method for performing RFID identification on returned products provided by the present invention is shown.

[0075] like Figure 2 As shown, according to an embodiment of the present invention, performing RFID identification on a returned product to determine product information of the returned product includes:

[0076] S202, obtaining response data from the RFID tag;

[0077] S204, comparing the response data with the product information in the database, and determining the product information identical to the response data as the product information of the returned product;

[0078] S206: When the product information identical to the response data does not exist in the database, the return product identification fails.

[0079] It should be noted that each digital product is affixed with an RFID tag, which contains the unique ID of the digital product and relevant product information (such as product name, model, price, etc.). The smart AI vending machine is equipped with an RFID reader to transmit radio frequency signals and receive response signals from RFID tags.

[0080] During the digital product return process, when a digital product with an RFID tag enters the range of the reader, the reader will send out a radio frequency signal to activate the RFID tag. The activated RFID tag will receive the radio frequency signal from the reader and send its unique ID and related data in response. After receiving the response signal from the RFID tag, the reader will convert it into digital data and transmit it to the backend management system through a communication interface (such as a wired network or a wireless network). After receiving the data, the backend management system will compare it with the product information of the digital product stored in the database to verify the legitimacy, name, price, etc. of the returned product and determine the product information of the returned product.

[0081] When the return product recognition fails, the return product does not meet the return conditions, and the user is reminded to take the return product away through voice or other means.

[0082] According to an embodiment of the present invention, based on a three-dimensional image model, a preset product abnormality detection model is used to analyze first detection images at multiple shooting angles, an abnormal area of ​​the first detection image is framed by a first detection frame, and a difference type, a first difference score, and three-dimensional coordinate data of the abnormal area are determined, including:

[0083] According to the shooting angle of the first detection image, a standard image with the same shooting angle is obtained through the three-dimensional image model of the returned product;

[0084] The first detection image and the corresponding standard image at the same shooting angle are analyzed by a preset product anomaly detection model, the abnormal area of ​​the first detection image is framed, and the difference type and the first difference score of the abnormal area are determined; and the candidate frame of the abnormal area is determined as the first detection frame;

[0085] The coordinate data of the first detection frame is determined based on the three-dimensional image model of the returned product, and the center coordinate data of the first detection frame is determined as the three-dimensional coordinate data of the corresponding abnormal area.

[0086] It should be noted that the preset product anomaly detection model is obtained by training the historical detection images of digital products. By analyzing the historical detection images of digital products, the image features of different difference types (such as outer packaging damage, scratches, stains, sealing label displacement or damage, etc.) at different abnormal degrees are determined, and the corresponding difference scores are determined, thereby establishing the preset product anomaly detection model. By comparing the first detection image with the corresponding standard image at the same shooting angle, the abnormal area with differences is framed, and the framed abnormal area is analyzed by the preset product anomaly detection model to determine the difference type and the first difference score of the abnormal area.

[0087] Then, according to the shooting angle of the first detection image, the first detection frame is input into the three-dimensional image model of the returned product, the three-dimensional coordinate data corresponding to the vertices of the first detection frame are determined as the coordinate data of the first detection frame, and the center coordinate data of the first detection frame is calculated through the three-dimensional coordinate data of the vertices of the first detection frame, thereby determining the three-dimensional coordinate data of the abnormal area.

[0088] Figure 3 A flow chart of the method for determining the first abnormal area and the second abnormal area provided by the present invention is shown.

[0089] like Figure 3 As shown, according to an embodiment of the present invention, the three-dimensional coordinate data of the abnormal area under different shooting angles are analyzed to determine the first abnormal area and the second abnormal area, including:

[0090] S302, calculating the three-dimensional coordinate distances of abnormal areas of the same difference type at different shooting angles through the three-dimensional coordinate data of the abnormal areas;

[0091] S304, determining abnormal regions of the same difference type whose three-dimensional coordinate distance is less than a preset coordinate distance threshold as the same abnormal region;

[0092] S306, intercepting a second detection image of the same abnormal area from the first detection image of the same abnormal area;

[0093] S308: determine the abnormal area where there are multiple second detection images as the first abnormal area, and determine the other abnormal areas other than the first abnormal area as the second abnormal area.

[0094] It should be noted that by analyzing the three-dimensional coordinate data of abnormal areas of the same difference type under different shooting angles, when the three-dimensional coordinate distance of abnormal areas of the same difference type under two or more different shooting angles is less than the preset coordinate distance threshold, the abnormal areas of the same difference type under the two or more different shooting angles are determined to be the same abnormal area, and the corresponding second detection image is intercepted according to the first detection frame of the same abnormal area in different first detection images. When the analysis of abnormal areas of the same difference type under all different shooting angles is completed, it is determined whether there is a corresponding second detection image for each abnormal area, and the abnormal area with the second detection image is determined as the first abnormal area, and the area other than the first abnormal area without the second detection image is determined as the second abnormal area. Among them, the preset coordinate distance threshold is set by those skilled in the art according to actual needs.

[0095] According to an embodiment of the present invention, it also includes:

[0096] Determine a recommended viewing angle of the first abnormal area through the three-dimensional image model of the returned product;

[0097] Performing image changes on a plurality of second detection images of the first abnormal area based on the recommended viewing angle;

[0098] Superimposing the first detection frames after image changes on the plurality of second detection images, and determining the formed detection frame as the second detection frame;

[0099] Segmenting the second detection frame into a plurality of segmentation areas based on a preset segmentation specification;

[0100] Each second detection image is divided into a plurality of sub-images based on a plurality of segmented areas, and coordinate data of each sub-image is determined according to central coordinate data of the segmented areas.

[0101] It should be noted that the recommended viewing angle of the first abnormal area is a recommended viewing angle for observing the abnormal type and degree of the first abnormal area, which is determined by the three-dimensional image model of the returned product according to the abnormal type of the first abnormal area, the first detection frame at different shooting angles and the three-dimensional coordinate data. Then, the viewing angles (i.e., shooting angles) of multiple second detection images of the first abnormal area are adjusted to the recommended viewing angle by image stretching, rotation and other image change methods. The first detection frame of each second detection image will change accordingly according to the image change. The second detection frame of the first abnormal area at the recommended viewing angle is determined by superimposing the final first detection frame of each second detection image of the first abnormal area. The length and width of each segmented area are determined by the preset segmentation specifications, and the second detection frame corresponding to the first abnormal area is divided into multiple segmented areas of the same size. Each segmented area corresponds to a sub-image, and the coordinate data of each sub-image is the center coordinate data of the corresponding segmented area.

[0102] The preset segmentation specifications are set by the system, and those skilled in the art may adjust the length and width of each segmentation area in the preset segmentation specifications according to actual needs.

[0103] According to an embodiment of the present invention, calculating the final difference score of the first abnormal area includes:

[0104] According to the first abnormal area a m In each second detection image b n The first difference score P 1 (a m b n ) determines the first abnormal area a m In each second detection image b n The first influence weight k 1 (a m b n );

[0105] According to the first abnormal area a m Sub-image C i In the second detection image b n The segmented area c j The area ratio of the second detection image b is determined by n The segmented area c j The second influence weight k 2 (C i b n c j );

[0106] Determine the first abnormal area a by using the preset product abnormality detection model m The second detection image b n Sub-image C i The first difference score P 1 (a m b n C i );

[0107] According to the first abnormal area a m The second detection image b n Sub-image C i The first difference score P 1 (a m b n C i ), the first abnormal area a m In the second detection image b n The first influence weight k 1 (a m b n ) and the second detection image bn The segmented area c j The second influence weight k 2 (C i b n c j ) Calculate the second detection image b n The segmented area c j The second difference score P 2 (C i b n c j );

[0108] ;

[0109] Calculate each second detection image b according to the coordinate data of the sub-image n The segmented area c j The second difference score P 2 (C i b n c j ) to determine the segmentation area c j The third difference score P 3 (c j );

[0110] For all segmented regions c j The third difference score P 3 (c j ) are accumulated to determine the first abnormal area a m The final difference score P 4 (a m ).

[0111] It should be noted that the first difference score P 1 (a m b n ) is the mth first abnormal area a m In the nth second detection image b n The first difference score is the first abnormal area a m The first difference score at the recommended viewing angle is determined as the standard difference score, and the first difference score P is calculated. 1 (a m b n ) and the standard difference score to determine the mth first abnormal region a m In the nth second detection image b n The first influence weight k 1 (a m b n ). The second influence weight k 2 (C i b n cj ) is the first abnormal area a m In the nth second detection image b n The i-th sub-image C in i The image area and the corresponding j-th segmentation area c j ratio of the area of ​​the region.

[0112] The standard image of the first abnormal area at the recommended viewing angle is obtained by using the preset product abnormality detection model, and each sub-image of the first abnormal area in each second detection image is compared with the corresponding standard image to determine the first difference score of each sub-image of the first abnormal area in each second detection image. 1 (a m b n C i ) is the first abnormal area a m In the nth second detection image b n The i-th sub-image C in i The first difference score.

[0113] The first difference score P 1 (a m b n C i ) are respectively related to the corresponding first influence weight k 1 (a m b n ) and the second influence weight k 2 (C i b n c j ) is multiplied to determine the second detection image b n The segmented area c j The second difference score P 2 (C i b n c j ). Calculate the average of the second difference scores of the segmented areas corresponding to the sub-images with the same coordinate data in each second detection image, and determine the third difference score of the segmented area at the recommended viewing angle. In the calculation process, only the second difference scores with numerical values ​​are calculated.

[0114] According to an embodiment of the present invention, determining the product difference score of the returned product in combination with the first difference score of the second abnormal area includes:

[0115] determining an influence weight of each first abnormal region and second abnormal region based on the difference type;

[0116] The final difference score of each first abnormal area and the first difference score of the second abnormal area are multiplied by the corresponding influence weights respectively, and the calculation results are accumulated to determine the product difference score of the returned product.

[0117] It should be noted that the system sets the impact weight of the corresponding abnormal area according to the impact of different difference types (such as outer packaging damage, scratches, stains, sealing label displacement or damage, etc.) on the sales of digital products. The product difference score of the returned product is determined by weighted calculation of the final difference score of each first abnormal area and the first difference score of the second abnormal area of ​​the returned product.

[0118] According to an embodiment of the present invention, it also includes:

[0119] When the returned product is successfully returned, the three-dimensional image model of the returned product is updated based on the first detection images taken at multiple shooting angles.

[0120] It should be noted that when the returned product is successfully returned, the three-dimensional image model of the returned product is updated according to the current status of the returned product, so as to effectively evaluate the outer packaging status of the returned product when it is returned for the second time.

[0121] Figure 4 A block diagram of a digital product return identification system for an intelligent AI vending machine provided by the present invention is shown.

[0122] like Figure 4 As shown, the second aspect of the present invention provides a digital product return identification system for an intelligent AI vending machine, comprising:

[0123] The product information acquisition module is used to perform RFID identification on the returned product and determine the product information of the returned product;

[0124] A detection image module, used to obtain a first detection image at multiple shooting angles through a detection device when the returned product is successfully identified;

[0125] A first analysis module is used to retrieve a three-dimensional image model of the returned product from a database according to the product information of the returned product; based on the three-dimensional image model, the first detection images of multiple shooting angles are analyzed by a preset product anomaly detection model, the abnormal area of ​​the first detection image is framed by a first detection frame, and the difference type, the first difference score and the three-dimensional coordinate data of the abnormal area are determined;

[0126] The second analysis module is used to analyze the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area; intercept the second detection image corresponding to the first abnormal area from the first detection image corresponding to the first abnormal area; calculate the final difference score of the first abnormal area, and determine the product difference score of the returned product in combination with the first difference score of the second abnormal area;

[0127] The product verification module is used to determine that when the product difference score is greater than a preset difference score threshold, the returned product does not meet the return conditions; when the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions and the returned product is recycled.

[0128] The third aspect of the present invention provides a computer-readable storage medium, which includes a digital product return identification method program for an intelligent AI vending machine. When the digital product return identification method program for an intelligent AI vending machine is executed by a processor, the steps of the digital product return identification method for an intelligent AI vending machine as described above are implemented.

[0129] The information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals (including but not limited to signals transmitted between user terminals and other devices, etc.) involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. For example, the "product information of returned products" and "three-dimensional image model of returned products" involved in this disclosure are all obtained with full authorization.

[0130] The present invention discloses a method and system for identifying digital product returns of an intelligent AI vending machine, the method comprising: determining product information of a returned product through RFID identification, obtaining a first detection image; retrieving a three-dimensional image model of the returned product from a database; selecting an abnormal area of ​​the first detection image through a preset product abnormality detection model, and determining the difference type, first difference score and three-dimensional coordinate data of the abnormal area; analyzing the three-dimensional coordinate data of the abnormal area under different shooting angles, determining the first abnormal area and the second abnormal area; calculating the final difference score of the first abnormal area, combining the first difference score of the second abnormal area, and determining the product difference score of the returned product; when the product difference score is greater than the preset difference score threshold, the returned product does not meet the return condition; otherwise, the returned product meets the return condition. The present invention can improve the efficiency of identifying packaging problems of returned products.

[0131] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0132] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0133] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0134] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0135] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. A digital product return identification method for an intelligent AI vending machine, characterized in that: include: Perform RFID identification on the returned product to determine the product information of the returned product; When the returned product is successfully identified, a first detection image at multiple shooting angles is acquired by a detection device; Retrieving the three-dimensional image model of the returned product from the database according to the product information of the returned product; Based on the three-dimensional image model, the first detection images of the multiple shooting angles are analyzed by a preset product abnormality detection model, an abnormal area of ​​the first detection image is framed by a first detection frame, and a difference type, a first difference score and three-dimensional coordinate data of the abnormal area are determined; Analyze the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area; intercept the second detection image corresponding to the first abnormal area from the first detection image corresponding to the first abnormal area; Calculate the final difference score of the first abnormal area, and determine the product difference score of the returned product by combining the first difference score of the second abnormal area; When the product difference score is greater than the preset difference score threshold, the return product does not meet the return conditions; When the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions, and the returned product is recycled; The step of analyzing the three-dimensional coordinate data of the abnormal area at different shooting angles to determine the first abnormal area and the second abnormal area includes: Calculate the three-dimensional coordinate distance of abnormal areas of the same difference type under different shooting angles through the three-dimensional coordinate data of the abnormal area; Determine abnormal areas of the same difference type whose three-dimensional coordinate distance is less than a preset coordinate distance threshold as the same abnormal area; intercepting a second detection image of the same abnormal area from the first detection image of the same abnormal area; An abnormal area where a plurality of second detection images exist is determined as a first abnormal area, and other abnormal areas other than the first abnormal area are determined as second abnormal areas.

2. The digital product return identification method of the intelligent AI vending machine according to claim 1 is characterized in that: The performing RFID identification on the returned product to determine the product information of the returned product includes: Get the response data of the RFID tag; Comparing the response data with the product information in the database, and determining the product information identical to the response data as the product information of the returned product; When the product information identical to the response data does not exist in the database, the return product identification fails.

3. The digital product return identification method of the intelligent AI vending machine according to claim 1 is characterized in that: The method of analyzing the first detection images of the multiple shooting angles by using a preset product abnormality detection model based on the three-dimensional image model, selecting an abnormal area of ​​the first detection image by using a first detection frame, and determining a difference type, a first difference score, and three-dimensional coordinate data of the abnormal area includes: According to the shooting angle of the first detection image, a standard image with the same shooting angle is obtained through the three-dimensional image model of the returned product; Analyze the first detection image and the corresponding standard image at the same shooting angle through a preset product abnormality detection model, select the abnormal area of ​​the first detection image, and determine the difference type and the first difference score of the abnormal area; determine the candidate frame of the abnormal area as the first detection frame; The coordinate data of the first detection frame is determined based on the three-dimensional image model of the returned product, and the center coordinate data of the first detection frame is determined as the three-dimensional coordinate data of the corresponding abnormal area.

4. The digital product return identification method of the intelligent AI vending machine according to claim 1 is characterized in that: Also includes: Determine a recommended viewing angle of the first abnormal area by using the three-dimensional image model of the returned product; Performing image changes on a plurality of second detection images of the first abnormal area based on the recommended viewing angle; superimposing the first detection frame after image changes are performed on the multiple second detection images, and determining the formed detection frame as the second detection frame; Segmenting the second detection frame into a plurality of segmentation areas based on a preset segmentation specification; Each second detection image is divided into a plurality of sub-images based on the plurality of divided areas, and coordinate data of each sub-image is determined according to central coordinate data of the divided areas.

5. The digital product return identification method of the intelligent AI vending machine according to claim 4 is characterized in that: The calculating the final difference score of the first abnormal area comprises: Determine a first influence weight k1(ambn) of the first abnormal area am in each second detection image bn according to a ratio of the first difference score P1(ambn) of the first abnormal area am in each second detection image bn; Determine a second influence weight k2 (Cibncj) of the segmented region cj of the second detection image bn according to the area ratio of the sub-image Ci of the first abnormal region am in the segmented region cj of the second detection image bn; Determine a first difference score P1 (ambnCi) of a sub-image Ci of a second detection image bn of a first abnormal area am by using a preset product abnormality detection model; Calculate the second difference score P2(Cibncj) of the segmented area cj of the second detection image bn according to the first difference score P1(ambnCi) of the sub-image Ci of the second detection image bn of the first abnormal area am, the first influence weight k1(ambn) of the first abnormal area am in the second detection image bn, and the second influence weight k2(Cibncj) of the segmented area cj of the second detection image bn; ; Calculate each second detection image b according to the coordinate data of the sub-image n The segmented area c j The second difference score P2(C i b n c j ) to determine the segmentation area c j The third difference score P3(c j ); For all segmented regions c j The third difference score P3(c j ) are accumulated to determine the first abnormal area a m The final difference score P4(a m ).

6. The digital product return identification method of the intelligent AI vending machine according to claim 1 is characterized in that: Determining the product difference score of the returned product by combining the first difference score of the second abnormal area includes: determining an influence weight of each first abnormal region and second abnormal region based on the difference type; The final difference score of each first abnormal area and the first difference score of the second abnormal area are respectively multiplied by the corresponding influence weight, and the calculation results are accumulated to determine the product difference score of the returned product.

7. The digital product return identification method of the intelligent AI vending machine according to claim 1 is characterized in that: Also includes: When the returned product is successfully returned, the three-dimensional image model of the returned product is updated based on the first detection images at the multiple shooting angles.

8. A digital product return identification system for an intelligent AI vending machine, used to implement the digital product return identification method for an intelligent AI vending machine as claimed in any one of claims 1 to 7, characterized in that: include: A product information acquisition module, used to perform RFID identification on the returned product and determine the product information of the returned product; A detection image module, used for acquiring first detection images at multiple shooting angles through a detection device when the returned product is successfully identified; A first analysis module is configured to retrieve a three-dimensional image model of the returned product from a database according to the product information of the returned product; based on the three-dimensional image model, analyze the first detection images of the multiple shooting angles by using a preset product anomaly detection model, select an abnormal area of ​​the first detection image by using a first detection frame, and determine a difference type, a first difference score, and three-dimensional coordinate data of the abnormal area; A second analysis module is used to analyze the three-dimensional coordinate data of the abnormal area under different shooting angles to determine the first abnormal area and the second abnormal area; intercepting a second detection image corresponding to the first abnormal area from the first detection image corresponding to the first abnormal area; Calculate the final difference score of the first abnormal area, and determine the product difference score of the returned product by combining the first difference score of the second abnormal area; A product verification module, configured to determine that when the product difference score is greater than a preset difference score threshold, the returned product does not meet the return conditions; when the product difference score is less than or equal to the preset difference score threshold, the returned product meets the return conditions and the returned product is recycled; The step of analyzing the three-dimensional coordinate data of the abnormal area at different shooting angles to determine the first abnormal area and the second abnormal area includes: Calculate the three-dimensional coordinate distance of abnormal areas of the same difference type under different shooting angles through the three-dimensional coordinate data of the abnormal area; Determine abnormal areas of the same difference type whose three-dimensional coordinate distance is less than a preset coordinate distance threshold as the same abnormal area; intercepting a second detection image of the same abnormal area from the first detection image of the same abnormal area; An abnormal area where a plurality of second detection images exist is determined as a first abnormal area, and other abnormal areas other than the first abnormal area are determined as second abnormal areas.

9. The digital product return identification system of the intelligent AI vending machine according to claim 8, characterized in that: The step of analyzing the three-dimensional coordinate data of the abnormal area at different shooting angles to determine the first abnormal area and the second abnormal area includes: Calculate the three-dimensional coordinate distance of abnormal areas of the same difference type under different shooting angles through the three-dimensional coordinate data of the abnormal area; Determine abnormal areas of the same difference type whose three-dimensional coordinate distance is less than a preset coordinate distance threshold as the same abnormal area; intercepting a second detection image of the same abnormal area from the first detection image of the same abnormal area; An abnormal area where a plurality of second detection images exist is determined as a first abnormal area, and other abnormal areas other than the first abnormal area are determined as second abnormal areas.

Images