A commodity identification method based on AI intelligent identification

Abstract

The application relates to a commodity identification method based on AI intelligent identification. On one hand, the similarity recognition view angle of a target commodity is preset, and multiple continuous shooting images are acquired under the preset view angle for fusion. Compared with the existing mean fusion mode, the clear area of the image can be more accurately represented, and the accuracy of the subsequent similarity result is ensured. On the other hand, not only is the similarity and weight index compared separately, but also the weight index is jointly analyzed when the similarity is not a problem (because the use of materials causes a certain weight gap when the appearance of the commodity is very similar to that of the genuine product), so that the detection rate of abnormal commodities is further improved. In summary, the application not only improves the abnormal commodity detection efficiency of returned commodities through automatic processing, but also improves the detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of e-commerce big data analysis, and particularly relates to a commodity identification method based on AI intelligent identification. BACKGROUND

[0002] When e-commerce goods are sold, after-sales problems of commodities often occur, and if the after-sales problems cannot be solved, users will generally choose to return the goods for refund. In the return and refund process, the returned goods of the user need to be verified to prevent the user from causing loss to the merchant by "switching packages".

[0003] The existing verification method of returned goods is performed by manual verification. This method can normally process commodities with large differences in details, but if the appearance of the commodity is close to the imitation, the detection rate is low, and the manual verification method has large workload and low efficiency. SUMMARY

[0004] The purpose of the present application is to at least solve one of the problems of the prior art, and provide a commodity identification method based on AI intelligent identification.

[0005] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0006] Specifically, a commodity identification method based on AI intelligent identification is provided, which comprises the following:

[0007] P continuous shooting images under each view angle of a plurality of view angles of a target commodity are obtained at a preset workbench;

[0008] For any view angle, the P continuous shooting images associated therewith are subjected to image fusion to obtain an associated detection image of the view angle;

[0009] A template image of the target commodity of the i-th view angle is obtained, and the template image is subjected to similarity matching with the associated detection image to obtain the commodity similarity under the i-th view angle, and then the commodity similarity under all view angles is obtained;

[0010] The commodity similarity under all view angles is traversed, and if the commodity similarity is less than a first threshold value, it is directly determined that the target commodity has a risk of fake goods, and if the commodity similarity is not less than the first threshold value, the following mode is executed,

[0011] The weight data of the target commodity is continuously obtained, and the weight data is subtracted from the preset standard weight to obtain a weight index, and then it is determined whether the weight index is greater than a second threshold value, if yes, it is determined that the target commodity has a risk of fake goods, and if no, the commodity similarity under all view angles and the weight index are analyzed to determine whether the target commodity has a risk of fake goods;

[0012] The target commodity with the risk of fake goods is informed to the relevant staff for subsequent review.

[0013] Further, specifically, for any view angle, the P associated continuous shooting images are image fused to obtain an associated detection image of the view angle, comprising,

[0014] First, the P continuous shooting images are pixel mean fused by a classical image fusion method to obtain a fused image, that is, the pixel value of any pixel point in the fused image is the average value of the corresponding pixel points of the P continuous shooting images;

[0015] Then, the clear regions of each of the P continuous shooting images are found based on the clarity index;

[0016] The clear regions are sequentially replaced in the corresponding regions of the fused image according to the shooting order of the continuous shooting images to obtain an associated detection image;

[0017] Wherein, the clear coefficient is calculated as follows,

[0018] For the p-th image in the P continuous shooting images, p∈[1, P], since it is a two-dimensional image, let its image matrix be Matrix_p(p), and the clarity index of the pixel point Matrix_p(p, hp, sp) in the hp-th row and sp-th column of Matrix_p(p) is calculated as follows:

[0019] ;

[0020] Wherein, , , respectively represent the mean value, maximum value and minimum value of the pixel points in the neighborhood range of Matrix_p(p, hp, sp).

[0021] Further, specifically, the clear regions of each of the P continuous shooting images are found based on the clarity index, comprising,

[0022] When p=1, the pixel points with the maximum clarity index clear_1 and clear_2 in the p-th image are found by traversing, a first straight line is formed based on clear_1 and clear_2, and the pixel point with the maximum clarity index clear_3 is found from the region outside the first straight line in the p-th image, and the closed region formed by taking clear_1, clear_2 and clear_3 as vertices is the clear region of the image;

[0023] When p>1, the average value of the sharpness index of the pixels in the clear area corresponding to the (p-1)th image is calculated to obtain mean_p-1, the pixels in the pth image whose sharpness index is greater than mean_p-1 are searched out, two pixels clear_4 and clear_5 are randomly selected from the pixels, a second straight line is formed based on clear_4 and clear_5, and the pixel clear_6 with the maximum sharpness index is searched out from the area outside the second straight line in the pth image, and the closed area formed by taking clear_4, clear_5 and clear_6 as the vertices is the clear area of the image.

[0024] Further, specifically, based on the similarity of the goods under all viewing angles and the weight index, it is judged whether the target goods have a fake risk, including,

[0025] The maximum value of the similarity of the goods under all viewing angles is searched out and recorded as a representative value of the similarity, and the representative value of the similarity and the weight index are uniformly dimensioned;

[0026] At this time, the representative value of the similarity after uniform dimensioning and the weight index are subtracted to obtain an absolute value to obtain a correlation balance coefficient;

[0027] It is judged whether the correlation balance coefficient is greater than a third threshold value, if yes, it is judged that the target goods have a fake risk.

[0028] Further, the method further includes,

[0029] The similarity of the goods under all viewing angles and the weight index are used as multiple dimensions, and the Euclidean distance is calculated with the same goods of the target goods that have been inspected and qualified in the historical data, and the average value of the Euclidean distance is taken as an evaluation standard, the target goods whose average value of the Euclidean distance is greater than a fourth threshold value are searched out, and it is judged that the target goods have a fake risk.

[0030] Further, specifically, the target goods with a fake risk are informed to relevant staff for subsequent review, including,

[0031] A unique ID is added to the target goods with a fake risk, and the unique ID is informed to the staff for subsequent compliance, if the subsequent review result is a fake, the unique ID is not revoked, if the subsequent review result is not a fake, the unique ID is revoked for subsequent processing.

[0032] The application also proposes a goods identification system based on AI intelligent identification, including the following,

[0033] An image data acquisition module is configured to acquire P continuous shooting images under each viewing angle of multiple viewing angles of target goods at a preset workbench.

[0034] An image fusion module is configured to fuse P continuous shooting images associated with any view angle to obtain an associated detection image of the view angle;

[0035] A similarity calculation module is configured to obtain a template image of the target commodity in the i-th view angle, and perform similarity matching between the template image and the associated detection image to obtain a commodity similarity in the i-th view angle, and further obtain commodity similarities in all view angles;

[0036] A first judgment module is configured to traverse the commodity similarities in all view angles, and if there is a case that the commodity similarity is less than a first threshold value, it is directly judged that the target commodity has a fake risk, and if there is no case that the commodity similarity is less than the first threshold value, the following manner is executed,

[0037] A second judgment module is configured to continue to obtain weight data of the target commodity, and perform absolute value difference between the weight data and a preset standard weight to obtain a weight index, and then judge whether the weight index is greater than a second threshold value, if yes, it is judged that the target commodity has a fake risk, and if no, it is based on the commodity similarities in all view angles and the weight index to analyze whether the target commodity has a fake risk;

[0038] An alarm module is configured to inform relevant staff of the target commodity with a fake risk for subsequent review.

[0039] The present application has the following advantages:

[0040] The present application provides a commodity identification method based on AI intelligent identification, on the one hand, the view angle of the target commodity which is easy to identify similarity is preset, and multiple continuous shooting images are obtained in the preset view angle for fusion, compared with the existing mean fusion method, the clear area of the image can be more accurately represented, which provides guarantee for the accuracy of the subsequent similarity result; on the other hand, not only the similarity and the weight index are compared with the single threshold value, but also the weight index is analyzed jointly when the similarity has no problem (because the use of material causes a certain gap in weight when the appearance of the commodity is very similar to the genuine product), which further improves the detection rate of abnormal commodities. In summary, the present application not only improves the detection efficiency of abnormal commodities of returned commodities through automatic processing, but also improves the detection accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0041] The above and other features of the present disclosure will become more apparent by describing in detail embodiments thereof with reference to the attached drawings in which like reference numerals refer to like elements throughout the several views, which are not necessarily to scale, and in which:

[0042] Figure 1 A flow chart of a commodity identification method based on AI intelligent identification is shown. DETAILED DESCRIPTION

[0043] The concept, specific structure and technical effects of the present application will be described clearly and completely below with reference to the embodiments and the accompanying drawings, so as to fully understand the purpose, scheme and effects of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The same reference numerals in the drawings indicate the same or similar parts.

[0044] Embodiment 1, refer to Figure 1 The present application proposes a commodity identification method based on AI intelligent identification, comprising the following:

[0045] Step 110, obtaining P continuous shooting images of each view angle of the target commodity in multiple view angles at a preset workbench;

[0046] Step 120, for any view angle, performing image fusion on the associated P continuous shooting images to obtain an associated detection image of the view angle;

[0047] Step 130, obtaining a template image of the target commodity in the i-th view angle, and performing similarity matching on the template image and the associated detection image to obtain the commodity similarity in the i-th view angle, and then obtaining the commodity similarity in all view angles;

[0048] Step 140, traversing the commodity similarity in all view angles, if there is a case that the commodity similarity is less than the first threshold value, directly judging that the target commodity has a risk of fake goods, if there is no case that the commodity similarity is less than the first threshold value, executing in the following way,

[0049] Step 150, continuously obtaining the weight data of the target commodity, and taking the absolute value of the difference between the weight data and the preset standard weight to obtain a weight index, and then judging whether the weight index is greater than the second threshold value, if yes, judging that the target commodity has a risk of fake goods, if no, based on the commodity similarity in all view angles and the weight index, analyzing whether the target commodity has a risk of fake goods;

[0050] Step 160, inform the relevant staff of the target goods with the risk of fake goods for subsequent review.

[0051] In the embodiment 1, on the one hand, the view angle suitable for similarity recognition of the target goods is preset, and multiple continuous shooting images are obtained at the preset view angle (the selection of the view angle can be pre-selected according to the view angle characteristics of the target goods, and the view angle suitable for similarity comparison is selected), and compared with the existing mean fusion method, the clear area of the image can be more accurately represented, and the accuracy of the subsequent similarity result is guaranteed; on the other hand, not only the similarity and weight indicators are compared separately, but also the weight indicator is analyzed jointly when the similarity is not a problem (because the appearance of the goods is very similar to the genuine goods, the weight will be different due to the use of the material), and the detection rate of abnormal goods is further improved. In summary, the present application not only improves the detection efficiency of abnormal goods of returned goods through automatic processing, but also improves the detection accuracy.

[0052] As a preferred embodiment of the present application, specifically, for any view angle, the P associated continuous shooting images are fused to obtain the associated detection image of the view angle, comprising,

[0053] First, the P continuous shooting images are fused by the classical image fusion method to obtain the fused image, that is, the pixel value of any pixel point in the fused image is the average value of the corresponding pixel points of the P continuous shooting images;

[0054] Then, the clear area of each P continuous shooting image is found based on the clear index;

[0055] The clear area is replaced in the corresponding area of the fused image in the order of shooting of the continuous shooting image to obtain the associated detection image;

[0056] Wherein, the calculation method of the clear coefficient is as follows,

[0057] For the pth image in the P continuous shooting images, p∈[1, P], since it is a two-dimensional image, let its image matrix be Matrix_p(p), and the clear index of the pixel point Matrix_p(p, hp, sp) in the hpth row and spth column of Matrix_p(p) is calculated as follows:

[0058]

[0059] Wherein, , , The mean value, maximum value and minimum value of the pixel points in the neighborhood range of Matrix_p(p, hp, sp) are represented respectively.

[0060] ​In the preferred embodiment, in view of the accuracy of subsequent similarity matching, a new image fusion method is proposed, which is based on the existing image fusion (i.e. mean fusion), and the clear area is found by calculating the sharpness coefficient of the pixel points, and then the clear area of each image to be fused is replaced by the fusion image obtained by the existing image fusion. In this way, the image for subsequent similarity analysis can better represent the image characteristics of the target commodity, and the result of subsequent similarity detection is more reasonable.

[0061] As a preferred embodiment of the present application, specifically, the clear area is found from P continuous shooting images based on the clear index, including,

[0062] When p=1, the pixel points clear_1 and clear_2 with the maximum clear index in the pth image are found by traversing and searching, and a first straight line is formed based on clear_1 and clear_2, and then the pixel point clear_3 with the maximum clear index is found from the area outside the first straight line in the pth image, and at this time the closed area formed by clear_1, clear_2 and clear_3 as the vertices is the clear area of the image;

[0063] When p>1, the average value of the clear index of the pixel points in the clear area corresponding to the (p-1)th image is calculated to obtain mean_p-1, the pixel points with a clear index greater than mean_p-1 in the pth image are found by traversing and searching, and then two pixel points clear_4 and clear_5 are randomly selected from these pixel points, a second straight line is formed based on clear_4 and clear_5, and then the pixel point clear_6 with the maximum clear index is found from the area outside the second straight line in the pth image, and at this time the closed area formed by clear_4, clear_5 and clear_6 as the vertices is the clear area of the image.

[0064] In the preferred embodiment, through the above image processing process, the next image can be in a more excellent iterative process relative to the previous image, and thus the clear area of the final replaced fusion image is optimal.

[0065] As a preferred embodiment of the present application, specifically, the similarity of the commodity under all viewing angles and the weight index are analyzed to determine whether the target commodity has a risk of fake goods, including,

[0066] The maximum value of the similarity of the commodity under all viewing angles is found by traversing all the similarities, and is recorded as the similarity representative value, and the similarity representative value and the weight index are uniformly dimensioned;

[0067] At this time, the similarity representative value and the weight index after uniform dimensioning are subtracted to obtain the correlation balance coefficient;

[0068] determining whether the correlation balance coefficient is greater than a third threshold value, and if so, determining that the target product has a fake risk.

[0069] In the preferred embodiment, considering that when a product is counterfeited, the appearance details are often pursued to be close, so that the similarity is high when the image similarity is detected (but the use of different materials will cause a certain gap in weight), so that although the similarity of all perspectives meets the threshold rule, we select the similarity value of the perspective with the highest similarity, and perform unified dimension processing with the weight index value, and then find the gap between them (i.e. the difference after unified dimension processing), if the gap is large enough, then we also consider that the target product has a fake risk.

[0070] As a preferred embodiment of the present application, the method further comprises,

[0071] The product similarity under all perspectives and the weight index are used as multiple dimensions, and the Euclidean distance is calculated with the same type of products of the target product that have been inspected and qualified in the historical data, and the Euclidean distance mean is taken as an evaluation standard, and the target product with a Euclidean distance mean greater than a fourth threshold value is found, and it is determined that the target product has a fake risk.

[0072] In the preferred embodiment, another preferred scheme is proposed, which calculates the Euclidean distance between the product similarity under all perspectives and the weight index and the historical qualified target products (considering the calculation amount, a preset number of samples can be selected from the historical qualified target products, which can be adjusted according to actual conditions), and then takes the mean of the Euclidean distance with all samples as a judgment standard, if the Euclidean distance is too large, then it is considered that the target product has a fake risk.

[0073] As a preferred embodiment of the present application, specifically, the target product with a fake risk is informed to the relevant staff for subsequent review, including,

[0074] A unique ID is added to the target product with a fake risk, and the unique ID is informed to the staff for subsequent compliance, if the subsequent review result is a fake, the unique ID is not revoked, if the subsequent review result is not a fake, the unique ID is revoked for subsequent processing.

[0075] In the preferred embodiment, a unique ID is added to the target product with a fake risk for subsequent unified processing and recording, which can optimize the process and improve efficiency.

[0076] In embodiment 2, the present application also proposes a product identification system based on AI intelligent identification, which comprises the following,

[0077] An image data acquisition module is configured to acquire P continuous shooting images of each view angle of multiple views of a target commodity at a preset workbench;

[0078] An image fusion module is configured to perform image fusion on the P continuous shooting images associated with any view angle to obtain an associated detection image of the view angle;

[0079] A similarity calculation module is configured to acquire a template image of the target commodity of the i-th view angle, and perform similarity matching between the template image and the associated detection image to obtain a commodity similarity under the i-th view angle, and further obtain commodity similarities under all view angles;

[0080] A first judgment module is configured to traverse the commodity similarities under all view angles, and if there is a case where the commodity similarity is less than a first threshold value, it is directly judged that the target commodity has a fake risk, and if there is no case where the commodity similarity is less than the first threshold value, the following manner is executed,

[0081] A second judgment module is configured to continue to acquire weight data of the target commodity, and perform absolute value difference between the weight data and a preset standard weight to obtain a weight index, and then judge whether the weight index is greater than a second threshold value, if yes, it is judged that the target commodity has a fake risk, and if no, it is analyzed based on the commodity similarities under all view angles and the weight index to judge whether the target commodity has a fake risk;

[0082] An alarm module is configured to inform relevant staff of the target commodity with a fake risk for subsequent review.

[0083] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically, or two or more modules can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0084] The integrated module, if implemented in the form of a software function module and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can include any entity or system capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0085] Although the description of the present application has been quite detailed and particularly described with respect to several described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, but should be considered to be a broad interpretation of the claims in view of the prior art, so as to effectively encompass the intended scope of the present application. In addition, the present application is described above in embodiments that the inventors can foresee, and the purpose is to provide a useful description, and those non-essential modifications to the present application that have not yet been foreseen can still represent equivalent modifications of the present application.

[0086] The above is only the preferred embodiment of the present application, and the present application is not limited to the above-mentioned embodiments, as long as the same means achieve the technical effect of the present application, it should belong to the protection scope of the present application. The technical solutions and / or embodiments within the protection scope of the present application can have various modifications and changes.

Claims

1. A commodity identification method based on AI intelligent identification, characterized in that, The method comprises the following steps: obtaining P continuous shooting images of each view angle of the target commodity at the preset workbench; for any view angle, performing image fusion on the P continuous shooting images associated therewith to obtain an associated detection image of the view angle; obtaining a template image of the target commodity of the i-th view angle, and performing similarity matching between the template image and the associated detection image to obtain the commodity similarity of the i-th view angle, and then obtaining the commodity similarity of all view angles; traversing the commodity similarity of all view angles, if there is a case where the commodity similarity is less than a first threshold value, it is directly judged that the target commodity has a fake risk, if there is no case where the commodity similarity is less than the first threshold value, the following mode is executed, continuously obtaining weight data of the target commodity, and taking the absolute value of the difference between the weight data and the preset standard weight to obtain a weight index, and then judging whether the weight index is greater than a second threshold value, if yes, it is judged that the target commodity has a fake risk, if no, based on the commodity similarity of all view angles and the weight index, it is analyzed whether the target commodity has a fake risk; informing the relevant staff of the target commodity with a fake risk for subsequent review; Specifically, for any view angle, the P continuous shooting images associated therewith are fused to obtain an associated detection image of the view angle, which comprises: firstly, the P continuous shooting images are fused by a classical image fusion method to obtain a fused image, that is, the pixel value of any pixel point in the fused image is the average value of the corresponding pixel points of the P continuous shooting images; then, based on the sharpness index, the clear regions of the P continuous shooting images are found out; the clear regions are replaced in the corresponding regions of the fused image in the order of shooting of the continuous shooting images to obtain the associated detection image; wherein the calculation method of the sharpness coefficient is as follows, for the p-th image of the P continuous shooting images, p∈[1, P], since it is a two-dimensional image, the image matrix thereof is denoted as Matrix_p(p), and the sharpness index of the pixel point Matrix_p(p, hp, sp) in the hp-th row and sp-th column of Matrix_p(p) is calculated as follows: ； wherein, , , respectively represent the mean value, the maximum value and the minimum value of the pixels within the neighborhood range of Matrix_p (p, hp, sp). Specifically, based on the sharpness index, the clear regions of the P continuous shooting images are found out, which comprises: when p=1, the pixel points clear_1 and clear_2 with the largest sharpness index in the p-th image are found out, a first straight line is formed based on clear_1 and clear_2, and the pixel point clear_3 with the largest sharpness index is found out from the region outside the first straight line in the p-th image, at this time, the closed region formed by taking clear_1, clear_2 and clear_3 as the vertices is the clear region of the image; When p>1, the average value of the sharpness index of the pixels in the clear area corresponding to the (p-1)th image is calculated to obtain mean_p-1, the pixels in the pth image whose sharpness index is greater than mean_p-1 are searched, two pixels clear_4 and clear_5 are randomly selected from the pixels, a second straight line is formed based on clear_4 and clear_5, and the pixel clear_6 with the maximum sharpness index is searched from the area outside the second straight line in the pth image. At this time, the closed area formed by taking clear_4, clear_5 and clear_6 as the vertices is the clear area of the image. 2.The AI intelligent identification-based commodity identification method of claim 1, wherein, Specifically, based on the similarity of the goods under all viewing angles and the weight index, it is judged whether the target goods have the risk of fake goods, including, The maximum value of the similarity of the goods under all viewing angles is searched and recorded as a representative value of the similarity, and the representative value of the similarity and the weight index are uniformly dimensioned; At this time, the representative value of the similarity after uniform dimensioning and the weight index are subtracted to obtain an absolute value to obtain a correlation balance coefficient; If the correlation balance coefficient is greater than the third threshold value, it is judged that the target goods have the risk of fake goods. 3.The AI intelligent identification-based commodity identification method of claim 1, wherein, The method further includes, The similarity of the goods under all viewing angles and the weight index are used as multiple dimensions, and the Euclidean distance is calculated with the same goods of the target goods that have been inspected and qualified in the historical data, and the mean value of the Euclidean distance is taken as an evaluation standard. The target goods with a Euclidean distance mean value greater than a fourth threshold value are searched, and it is judged that the target goods have the risk of fake goods. 4.The AI intelligent identification-based commodity identification method of claim 1, wherein, Specifically, the target goods with the risk of fake goods are informed to the relevant staff for subsequent review, including, A unique ID is added to the target goods with the risk of fake goods, and the unique ID is informed to the staff for subsequent review. If the subsequent review result is fake, the unique ID is not revoked. If the subsequent review result is not fake, the unique ID is revoked for subsequent processing.

5. A commodity identification system based on AI intelligent identification, characterized in that, The system includes the following steps of the method of any one of claims 1-4, An image data acquisition module is configured to acquire P continuous shooting images under each viewing angle of multiple viewing angles of a target good at a preset workbench; An image fusion module is configured to perform image fusion on the P continuous shooting images associated with any viewing angle to obtain an associated detection image of the viewing angle; A similarity calculation module is configured to acquire a template image of a target good under the ith viewing angle, and perform similarity matching on the template image and the associated detection image to obtain the similarity of the goods under the ith viewing angle, and further obtain the similarity of the goods under all viewing angles; A first judgment module is configured to traverse the similarity of the goods under all viewing angles. If there is a case where the similarity of the goods is less than a first threshold value, it is directly judged that the target goods have the risk of fake goods. If there is no case where the similarity of the goods is less than the first threshold value, the following method is executed, The second judging module is configured to continue to acquire weight data of the target commodity, and obtain a weight index by taking an absolute value of a difference between the weight data and a preset standard weight, and then judge whether the weight index is greater than a second threshold value. If the weight index is greater than the second threshold value, it is judged that the target commodity has a risk of fake goods. If the weight index is not greater than the second threshold value, the commodity similarity under all perspectives and the weight index are analyzed to judge whether the target commodity has a risk of fake goods. The alarm module is configured to inform relevant staff of the target commodity with a risk of fake goods for subsequent review.

Images