Commodity image retrieval method and device, electronic equipment and storage medium

Abstract

The invention discloses a commodity image retrieval method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining a first historical retrieval record of each first commodity identifier; inputting all the first image feature information corresponding to the first commodity identifier into a preset classification model to obtain retrieval attribute information corresponding to each first commodity identifier; selecting a target warehousing strategy from a warehousing strategy set according to target attribute information determined from all the retrieval attribute information on the basis of warehousing conditions, and storing first image feature information of the corresponding first commodity identifier in an image retrieval database on the basis of the target warehousing strategy; and performing commodity image retrieval based on the image retrieval database. According to the method and the device, the quantity of the image feature information stored in the image retrieval database is adjusted by utilizing the retrieval attribute information corresponding to each commodity identifier and the storage strategy, and commodity image retrieval is performed on the basis of the image retrieval database adjusted by the technical scheme of the invention, so that the commodity image retrieval efficiency can be effectively improved.

Description

Technical Field

[0001] The present application relates to but is not limited to the field of image retrieval technology, and in particular to a commodity image retrieval method, device, electronic device, and storage medium. Background Art

[0002] In the task of product image retrieval, the image is first located and cropped, and then the image features of the cropped image are extracted and compared with the product feature map in the image retrieval database, and the product category with the highest similarity is output. However, in actual use, as the number of product categories increases, the image retrieval database will become larger and larger, thus affecting the accuracy of the search results of the retrieval system. Based on this, the prior art combines multiple algorithms or models to perform product image retrieval by adopting hybrid methods and integrated models; and when there are multiple retrieval results, scores are performed based on multiple factors such as similarity measurement, relevance score, user feedback, etc. to adjust the order of the retrieval results, thereby improving the accuracy of product image retrieval. However, due to the introduction of multiple algorithms and models, the product image retrieval efficiency of the retrieval system is low. Summary of the invention

[0003] The embodiments of the present application provide a product image retrieval method, device, electronic device, and storage medium, which can effectively improve the efficiency of product image retrieval.

[0004] In a first aspect, an embodiment of the present application provides a product image retrieval method, comprising:

[0005] Acquire a first historical search record for each first commodity identifier, wherein the first historical search record includes first image feature information corresponding to the first commodity identifier, and the number of the first historical search record is at least 1;

[0006] Inputting all the first image feature information corresponding to the first commodity identification into a preset classification model to obtain search attribute information corresponding to each of the first commodity identifications, wherein the search attribute information is used to indicate the supply status of the commodity corresponding to the first commodity identification;

[0007] Determining target attribute information from all the search attribute information based on a preset storage condition, wherein the target attribute information is used to indicate the search attribute information that satisfies the storage condition;

[0008] Selecting a target warehousing strategy from a preset warehousing strategy set based on the target attribute information, and saving the first image feature information of the corresponding first commodity identification to an image retrieval database based on the target warehousing strategy, wherein the warehousing strategy set includes a plurality of candidate warehousing strategies, and different candidate warehousing strategies correspond to different commodity categories;

[0009] Product image retrieval is performed based on the image retrieval database.

[0010] In some embodiments, the warehousing condition is that the commodity category corresponding to the first commodity identifier is the first category or the second category, the first category is used to indicate seasonal commodities, the second category is used to indicate non-seasonal commodities, the first category corresponds to a first category identifier, and the second category corresponds to a second category identifier, and the determining of target attribute information from all the search attribute information based on the preset warehousing condition includes:

[0011] Acquire a third category identifier from the current search attribute information, wherein the third category identifier is used to indicate the commodity category of the first commodity identifier corresponding to the current search attribute information;

[0012] Matching the third category identifier with the first category identifier and the second category identifier respectively to obtain corresponding first matching results and second matching results;

[0013] When the first matching result indicates that the third category identifier matches the first category identifier, or the second matching result indicates that the third category identifier matches the second category identifier, the currently retrieved attribute information is determined as the target attribute information.

[0014] In some embodiments, based on the target warehousing strategy, saving the first image feature information of the corresponding first commodity identifier to an image retrieval database includes:

[0015] When the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the second category, filtering all of the first image feature information to obtain a plurality of filtered second image feature information, wherein each of the second image feature information is different from each other;

[0016] Performing clustering processing on the second image feature information based on a preset clustering algorithm to obtain a plurality of cluster clusters;

[0017] The average feature value of each of the clusters is calculated, and all of the average feature values ​​are saved in the image retrieval database.

[0018] In some embodiments, clustering is performed on the first image feature information corresponding to the first commodity identifier based on a preset clustering algorithm to obtain a plurality of clusters, including:

[0019] Determining a reference quantity of the first historical search record corresponding to the first commodity identifier;

[0020] Determine a search frequency level of the first commodity identifier based on the reference quantity, and determine a target cluster number based on the search frequency level, wherein a higher search frequency level corresponds to a larger value of the target cluster number;

[0021] Based on the clustering algorithm, clustering processing is performed on the second image feature information corresponding to the first commodity identification to obtain the cluster clusters of the target cluster number.

[0022] In some embodiments, based on the target warehousing strategy, saving the first image feature information of the corresponding first commodity identifier to an image retrieval database includes:

[0023] When the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the first category, filtering all the first image feature information to obtain a plurality of filtered second image feature information, wherein each of the second image feature information is different from each other;

[0024] All of the second image features are saved in the image retrieval database.

[0025] In some embodiments, after the first image feature information of the first commodity identifier corresponding to the first commodity identifier is saved to an image retrieval database based on the target warehousing strategy, the method further includes:

[0026] When receiving warehousing demand information, the warehousing demand information includes a second commodity identifier, obtaining a second historical search record corresponding to the second commodity identifier, wherein the commodity category corresponding to the second commodity identifier is different from the first category and the second category, and the second historical search record includes third image feature information corresponding to the commodity identifier;

[0027] All the third image feature information is saved in the image retrieval database.

[0028] In some embodiments, the first category corresponds to a reference category identifier and a reference timestamp range, the reference timestamp range changes continuously according to a preset period, and after the first image feature information corresponding to the first commodity identifier is saved to an image retrieval database based on the target warehousing strategy, the method further includes:

[0029] Establishing an index record between the first commodity identifier, the corresponding commodity category identifier and the first image feature information in the image retrieval database;

[0030] Determine a first target identifier from the first commodity identifiers in the image retrieval database, and determine a first timestamp range of the first target identifier, wherein the commodity category identifier corresponding to the first target identifier is the reference category identifier;

[0031] Determine a second target identifier from the first commodity identifier in the image retrieval database, and determine the number of retrievals corresponding to the second target identifier;

[0032] When the number of intersections between the first timestamp range and the reference timestamp range is 0, deleting the index record corresponding to the first target identifier in the image retrieval database;

[0033] When the number of retrieval times is less than a preset threshold, the index record corresponding to the second target identifier in the image retrieval database is deleted.

[0034] In a second aspect, an embodiment of the present application provides a control device, comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the product image retrieval method as described in the first aspect.

[0035] In a third aspect, an embodiment of the present application further provides an electronic device, comprising the control device of the second aspect.

[0036] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the product image retrieval method as described in the first aspect.

[0037] The embodiment of the present application provides a commodity image retrieval method, device, electronic device, and storage medium, the method comprising: obtaining a first historical retrieval record for each first commodity identification, wherein the first historical retrieval record includes first image feature information corresponding to the first commodity identification, and the number of the first historical retrieval records is at least 1; inputting all the first image feature information corresponding to the first commodity identification into a preset classification model to obtain retrieval attribute information corresponding to each first commodity identification, wherein the retrieval attribute information is used to indicate the supply status of the commodity corresponding to the first commodity identification; determining target attribute information from all the retrieval attribute information based on a preset warehousing condition, wherein the target attribute information is used to indicate the retrieval attribute information that satisfies the warehousing condition; selecting a target warehousing strategy from a preset warehousing strategy set based on the target attribute information, and based on the target warehousing strategy, saving the first image feature information of the corresponding first commodity identification to an image retrieval database, wherein the warehousing strategy set includes multiple candidate warehousing strategies, and different candidate warehousing strategies correspond to different commodity categories, and commodity image retrieval is performed based on the image retrieval database. This application uses the retrieval attribute information and warehousing strategy corresponding to each product identification to adjust the amount of image feature information stored in the image retrieval database, and performs product image retrieval based on the image retrieval database adjusted based on the technical solution of this application, which can effectively improve the efficiency of product image retrieval. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a flowchart of the steps of a product image retrieval method provided by an embodiment of the present application;

[0039] Figure 2 is a schematic diagram of a product image retrieval method provided by another embodiment of the present application;

[0040] Figure 3 It is a structural diagram of a control device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] It is understood that although the functional modules are divided in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flow chart. The terms "first", "second", etc. in the specification, claims or the above drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0043] In the task of product image retrieval, the image is first located and cropped, and then the image features of the cropped image are extracted and compared with the product feature map in the image retrieval database, and the product category with the highest similarity is output. However, in actual use, as the number of product categories increases, the image retrieval database will become larger and larger, thus affecting the accuracy of the search results of the retrieval system. Based on this, the prior art combines multiple algorithms or models to perform product image retrieval by adopting hybrid methods and integrated models; and when there are multiple retrieval results, scores are performed based on multiple factors such as similarity measurement, relevance score, user feedback, etc. to adjust the order of the retrieval results, thereby improving the accuracy of product image retrieval. However, due to the introduction of multiple algorithms and models, the product image retrieval efficiency of the retrieval system is low.

[0044] To solve the above-mentioned problems, an embodiment of the present application provides a commodity image retrieval method, device, electronic device, and storage medium, the method comprising: obtaining a first historical retrieval record of each first commodity identification, wherein the first historical retrieval record includes first image feature information corresponding to the first commodity identification, and the number of the first historical retrieval records is at least 1; inputting all the first image feature information corresponding to the first commodity identification into a preset classification model to obtain retrieval attribute information corresponding to each first commodity identification, wherein the retrieval attribute information is used to indicate the supply status of the commodity corresponding to the first commodity identification; determining target attribute information from all the retrieval attribute information based on preset warehousing conditions, wherein the target attribute information is used to indicate retrieval attribute information that satisfies the warehousing conditions; selecting a target warehousing strategy from a preset warehousing strategy set based on the target attribute information, and based on the target warehousing strategy, saving the first image feature information of the corresponding first commodity identification to an image retrieval database, wherein the warehousing strategy set includes multiple candidate warehousing strategies, and different candidate warehousing strategies correspond to different commodity categories, and commodity image retrieval is performed based on the image retrieval database. This application uses the retrieval attribute information and warehousing strategy corresponding to each product identification to adjust the amount of image feature information stored in the image retrieval database, and performs product image retrieval based on the image retrieval database adjusted based on the technical solution of this application, which can effectively improve the efficiency of product image retrieval.

[0045] The embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0046] refer to Figure 1 , Figure 1 1 is a flowchart of a method for retrieving product images provided by an embodiment of the present application. The present application embodiment provides a method for retrieving product images, which includes but is not limited to the following steps:

[0047] Step S10 : obtaining a first history search record for each first product identifier, wherein the first history search record includes first image feature information corresponding to the first product identifier, and the number of the first history search record is at least one.

[0048] Specifically, the first product identifier of this embodiment is an identification parameter that can uniquely identify a product. This embodiment does not limit the specific parameter format of the first product identifier, and technical personnel in this field can determine it according to actual needs. For example, the first product identifier corresponding to watermelon is A001, the first product identifier corresponding to apple is A003, and the first product identifier corresponding to shower gel is D010.

[0049] It should be noted that this embodiment does not limit the method for obtaining the first historical search record. Different search systems store the first historical search record in different ways, and those skilled in the art may select a method according to actual conditions.

[0050] Specifically, the first historical search record of this embodiment is used to indicate a search record of searching for a product image corresponding to the first product identifier based on the image search database, such as Figure 2 As shown, a first product identification corresponds to one or more first historical retrieval records, and the number of first historical retrieval records can reflect the retrieval frequency of the corresponding first product identification, wherein the first historical retrieval record includes first image feature information, and the first image feature information includes texture features, color features, shape features, etc., which is image feature information used to query and match image features in the corresponding index stored in the image retrieval database. Therefore, the first historical retrieval record can provide an effective data basis for subsequent warehousing.

[0051] Step S20, inputting all first image feature information corresponding to the first commodity identification into a preset classification model to obtain retrieval attribute information corresponding to each first commodity identification, wherein the retrieval attribute information is used to indicate the supply status of the commodity corresponding to the first commodity identification.

[0052] Specifically, the embodiment of the present application does not involve improvements to the preset classification model, but is only capable of classifying and identifying the input image feature information to obtain the retrieval attribute information corresponding to the first product identification.

[0053] Specifically, the retrieval attribute information of this embodiment is used to indicate the supply status of the commodity corresponding to the first commodity identification. For example, the retrieval attribute information may include a commodity category identification, and the commodity category identification is used to characterize seasonal commodity categories, non-seasonal commodity categories, non-seasonal commodity categories or rare commodity categories, etc., which can provide an effective data basis for matching warehousing conditions and screening the first image feature information that can be stored.

[0054] Step S30, determining target attribute information from all the search attribute information based on the preset storage conditions, wherein the target attribute information is used to indicate the search attribute information that meets the storage conditions.

[0055] Specifically, since the image retrieval database is used to perform product image retrieval, and during the product image retrieval process, the retrieval frequencies of the first product identifications corresponding to different retrieval attribute information are different. For example, within a preset time period, the retrieval frequencies corresponding to rare product categories and non-seasonal product categories will be less, or even 0. Since the retrieval frequencies corresponding to such products are low, the impact on the overall retrieval accuracy of product image retrieval is small. The warehousing conditions of this embodiment are used to filter out warehousing data that has little relevance to the retrieval accuracy of subsequent product image retrievals, such as filtering out the first image feature information corresponding to the first product identifications corresponding to the rare product categories and non-seasonal product categories, and retaining the first image feature information corresponding to the first product identifications with a larger retrieval frequency such as the seasonal product categories and non-seasonal product categories as the warehousing data.

[0056] Understandably, reference Figure 2 In this embodiment, target attribute information that meets the conditions is screened out from multiple search attribute information through storage conditions, and then image feature information that can be stored in the image retrieval database (i.e., the first image feature information corresponding to the first product identifier X) is screened out from the first image feature information corresponding to all first product identifiers, thereby effectively controlling the amount of storage data and screening out storage data that has little correlation with the retrieval accuracy of subsequent product image retrievals, thereby ensuring the retrieval accuracy of subsequent product image retrieval operations while reducing the storage performance requirements for the image retrieval database, and effectively improving the efficiency of product image retrieval.

[0057] In addition, in some embodiments, the warehousing condition is that the commodity category corresponding to the first commodity identification is the first category or the second category, the first category is used to indicate seasonal commodities, the second category is used to indicate non-seasonal commodities, the first category corresponds to the first category identification, and the second category corresponds to the second category identification. Figure 1 Step S30 includes but is not limited to the following steps:

[0058] Step S31, obtaining a third category identifier from the current search attribute information, wherein the third category identifier is used to indicate the commodity category of the first commodity identifier corresponding to the current search attribute information;

[0059] Step S32, matching the third category identifier with the first category identifier and the second category identifier respectively to obtain corresponding first matching results and second matching results;

[0060] Step S33: when the first matching result indicates that the third category identifier matches the first category identifier, or the second matching result indicates that the third category identifier matches the second category identifier, the currently searched attribute information is determined as the target attribute information.

[0061] Specifically, referring to the description of the above embodiment, the retrieval attribute information of this embodiment includes a commodity category identifier (i.e., a third category identifier), and the warehousing condition is that the commodity category corresponding to the first commodity identifier is the first category or the second category, the first category is used to indicate seasonal commodities, and the second category is used to indicate non-seasonal commodities. The first category corresponds to a first category identifier, and the second category corresponds to a second category identifier. By matching the third category identifiers in each retrieval attribute information with the first category identifier and the second category identifier in the warehousing condition in turn, corresponding matching results are obtained. When one of the matching results indicates a successful match, it means that the successfully matched retrieval attribute information meets the warehousing condition, and the retrieval attribute information is determined as the target attribute information, and the first image feature information corresponding to the target attribute information can be stored.

[0062] It should be noted that the present embodiment does not limit the specific identification matching method, and similarity calculation, string matching, etc. can be used, and those skilled in the art can select them according to actual conditions.

[0063] Step S40, based on the target attribute information, a target warehousing strategy is selected from a preset warehousing strategy set, and based on the target warehousing strategy, the first image feature information of the corresponding first commodity identification is saved to the image retrieval database, wherein the warehousing strategy set includes multiple candidate warehousing strategies, and different candidate warehousing strategies correspond to different commodity categories.

[0064] Specifically, the storage strategy set of this embodiment includes multiple candidate storage strategies. Different candidate storage strategies correspond to different commodity categories. The first image feature information of the same data volume has different corresponding storage data volumes after passing through different candidate storage strategies.

[0065] It can be understood that after filtering out the storage data with little correlation with the retrieval accuracy of subsequent product image retrieval through the storage conditions, there will be multiple product categories corresponding to the target attribute information that can be stored. This embodiment further distinguishes the storage methods of the first image feature information of the corresponding target attribute information according to different product categories, that is, for different product categories, respectively select the corresponding target storage strategy from multiple candidate storage strategies in the storage strategy set, and perform data processing on the corresponding first image feature information and store it in the warehouse with the target storage strategy. Referring to the description of the above embodiment, the product categories that meet the storage conditions are the image feature information corresponding to the non-seasonal product category (i.e., the first category) and the seasonal product category (i.e., the second category). Different storage strategies are respectively corresponding to the two different product categories. For example, the first category corresponds to the candidate storage strategy A, and the second category corresponds to the candidate storage strategy B. The first image feature information with the same data volume has different corresponding storage data volumes after different candidate storage strategies, thereby further reducing the number of features input into the image retrieval database, thereby reducing the storage performance requirements for the image retrieval database, and effectively improving the efficiency of product image retrieval.

[0066] Additionally, in some embodiments, Figure 1 In step S40, based on the target warehousing strategy, the first image feature information of the corresponding first commodity identifier is saved to the image retrieval database, including but not limited to the following steps:

[0067] Step S401, when the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the second category, all the first image feature information is filtered to obtain a plurality of filtered second image feature information, wherein each second image feature information is different from each other;

[0068] Step S402, clustering the second image feature information based on a preset clustering algorithm to obtain a plurality of clusters;

[0069] Step S403, calculating the average eigenvalue of each cluster, and saving all the average eigenvalues ​​to the image retrieval database.

[0070] It is understandable that since the first product identifier corresponds to at least one first historical retrieval record, when the number of first historical retrieval records corresponding to the first product identifier of the target attribute information is multiple, the first image feature information corresponding to the first product identifier will have repeated feature information. Data preprocessing is performed before clustering processing to filter out repeated feature information and reduce the computational cost of clustering.

[0071] Specifically, since the motivation for triggering the product image retrieval operation is that the product corresponding to a certain first product identifier is purchased, the retrieval frequency is used to indicate the purchase frequency of the product corresponding to the corresponding first product identifier. Since the second category is a non-seasonal product category, the retrieval frequency corresponding to this type of product will be affected by factors such as the corresponding popularity and user evaluation. That is to say, the retrieval frequency of the first product identifier corresponding to the second category may have multiple situations (for example, high frequency and low frequency). If the first image feature information of all first product identifiers corresponding to the second category is saved in the image retrieval database, the storage performance pressure of the image retrieval database will be increased. Based on this, the present application filters the first image feature information corresponding to such products, and uses a preset clustering algorithm to filter the different second image feature information obtained after filtering. The second image feature information is clustered to obtain multiple clusters, the average feature value of each cluster is calculated, and all the average feature values ​​are saved in the image retrieval database, that is, all the first image feature information corresponding to each first product identification is first clustered and then feature averaged. During the feature clustering process, some noise or outliers may be excluded or weakened, thereby improving the purity and accuracy of the image retrieval database, which helps to reduce false detections and missed detections during retrieval, and through clustering, image features with similar semantics are organized together, and further feature averaged, which reduces the data volume of image feature information while ensuring the strong correlation between the average feature and the clustering category, thereby providing effective support for improving the efficiency of product image retrieval and improving the accuracy of product image retrieval results.

[0072] It should be noted that the embodiments of the present application are not limited to a specific clustering algorithm, which may be a K-means clustering algorithm, a GMMs clustering algorithm, etc., and those skilled in the art may determine it according to actual needs.

[0073] In addition, in some embodiments, clustering the second image feature information based on a preset clustering algorithm in step S42 to obtain a plurality of clusters includes but is not limited to the following steps:

[0074] Step S4021, determining the reference quantity of the first historical search record corresponding to the first product identifier;

[0075] Step S4022, determining the search frequency level of the first product identifier based on the reference quantity, and determining the target cluster number based on the search frequency level, wherein a higher search frequency level corresponds to a larger value of the target cluster number;

[0076] Step S4023: clustering the second image feature information corresponding to the first product identifier based on a clustering algorithm to obtain a target number of clusters.

[0077] It can be understood that the more features the image retrieval database stores for a first product identification, the higher the detection rate and retrieval accuracy of the first product identification will be when the image retrieval database is used to retrieve product images. The higher the retrieval frequency level, the greater the demand for the corresponding product. This application needs to ensure a high detection rate and high retrieval accuracy for such higher-frequency products, so as to ensure the overall retrieval accuracy of product image retrieval.

[0078] Specifically, the number of clusters obtained after clustering processing, that is, the target cluster number is the same as the number of average eigenvalues, and the larger the value of the target cluster number is, the more detailed the clustering result is, and more details and differences in the data can be captured. In this embodiment, for the clustering processing of the first image feature information of the first product identification with a higher retrieval frequency level, a larger value of the target cluster number is configured, thereby effectively ensuring the retrieval accuracy of such products.

[0079] Exemplarily, the retrieval frequency levels include level 1 (retrieval frequency is too low) and level 2 (retrieval frequency is relatively low). The target cluster number corresponding to the first product identification corresponding to level 1 is 2, and 2 average feature values ​​are obtained after the corresponding feature averaging processing. The target cluster number corresponding to the first product identification corresponding to level 2 is 5, and 5 average feature values ​​are obtained after the corresponding feature averaging processing.

[0080] Additionally, in some embodiments, Figure 1 In step S40, based on the target warehousing strategy, the first image feature information of the corresponding first commodity identifier is saved to the image retrieval database, including but not limited to the following steps:

[0081] Step S404, when the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the first category, all the first image feature information is filtered to obtain a plurality of filtered second image feature information, wherein each second image feature information is different from each other;

[0082] Step S405: save all second image features to the image retrieval database.

[0083] It can be understood that, referring to the description of the above embodiment, the second category corresponds to the seasonal product category, and the corresponding retrieval frequency is relatively high. Therefore, in order to ensure a high detection rate and high retrieval accuracy for such higher-frequency products, this embodiment saves all second image features corresponding to the first product identification corresponding to the second category to the image retrieval database to ensure the overall retrieval accuracy of product image retrieval.

[0084] In addition, in some embodiments, when executing Figure 1 After step S40, the product image retrieval method further includes but is not limited to the following steps:

[0085] Step S406, when receiving the warehousing demand information, the warehousing demand information includes the second commodity identifier, obtaining a second historical search record corresponding to the second commodity identifier, wherein the commodity category corresponding to the second commodity identifier is different from the first category and the second category, and the second historical search record includes third image feature information corresponding to the commodity identifier;

[0086] Step S407: Save all the third image feature information to the image retrieval database.

[0087] Specifically, the storage request information in this embodiment is issued by a retrieval system administrator, or is generated according to user needs.

[0088] It can be understood that, referring to the description of the above embodiment, since the first image feature information corresponding to some first product identifications (such as non-seasonal product categories and rare product categories) is not stored in the warehouse due to the warehousing conditions, the retrieval accuracy of such filtered products will inevitably be affected when performing product image retrieval. Therefore, in actual applications, after receiving the warehousing demand information carrying the second product identification, the second historical retrieval record corresponding to the second product identification is obtained, wherein the product category corresponding to the second product identification is different from the first category and the second category (that is, it does not meet the general warehousing conditions), and the second historical retrieval record includes the third image feature information corresponding to the product identification, and all the third image feature information is saved in the image retrieval database. In this way, in the worst case, such products will only be affected to a certain extent during the first retrieval, and will not affect the subsequent image retrieval accuracy and detection rate. This approach not only ensures the storage quantity limit on the image retrieval database, but also ensures that the retrieval accuracy will not be greatly affected.

[0089] In addition, in some embodiments, the first category corresponds to a reference category identifier and a reference timestamp range, and the reference timestamp range changes continuously according to a preset period. Figure 1 After step S40, the product image retrieval method further includes but is not limited to the following steps:

[0090] Step S408, creating an index record between the first commodity identifier, the corresponding commodity category identifier and the first image feature information in the image retrieval database;

[0091] Step S409, determining a first target identifier from the first commodity identifier in the image retrieval database, and determining a first timestamp range of the first target identifier, wherein the commodity category identifier corresponding to the first target identifier is a reference category identifier;

[0092] Step S410, determining a second target identifier from the first commodity identifier in the image retrieval database, and determining a search count corresponding to the second target identifier;

[0093] Step S411, when the number of intersections between the first timestamp range and the reference timestamp range is 0, deleting the index record corresponding to the first target identifier in the image retrieval database;

[0094] Step S412: when the number of retrieval times is less than a preset threshold, the index record corresponding to the second target identifier in the image retrieval database is deleted.

[0095] Specifically, since the first category is a seasonal commodity category, and as time changes, the current seasonal attribute is also constantly changing, so the reference timestamp range corresponding to the first category is constantly changing according to a preset period.

[0096] Specifically, the number of searches corresponding to this embodiment is the number of searches for the first product identifier within a preset time period. The specific preset threshold can be determined by those skilled in the art according to actual needs.

[0097] It can be understood that after completing the storage of different types of first image feature information into the image retrieval database, this embodiment updates the index records and related feature data stored in the image retrieval database. Specifically, an index record between the first product identifier, the corresponding product category identifier and the first image feature information is established in the image retrieval database, the first target identifier is determined from the first product identifier in the image retrieval database, and the first timestamp range of the first target identifier is determined, wherein the product category identifier corresponding to the first target identifier is the reference category identifier, the second target identifier is determined from the first product identifier in the image retrieval database, and the number of retrievals corresponding to the second target identifier is determined, when the number of intersections between the first timestamp range and the reference timestamp range is 0, the index record corresponding to the first target identifier in the image retrieval database is deleted, and when the number of retrievals is less than a preset threshold, the index record corresponding to the second target identifier in the image retrieval database is deleted; that is, the data that is redundant or affects the efficiency of product image retrieval in the image retrieval database (data corresponding to out-of-season product categories and data with too low a retrieval number) is deleted, and product image retrieval is performed based on the updated image retrieval database, which can effectively improve the image retrieval efficiency.

[0098] Step S50: performing product image retrieval based on an image retrieval database.

[0099] Referring to the description of the above embodiments and Figure 2 This application uses the retrieval attribute information corresponding to each product identification and multiple candidate storage strategies in the storage strategy set to adjust the amount of image feature information stored in the image retrieval database, and performs product image retrieval based on the adjusted image retrieval database, which can effectively improve the efficiency of product image retrieval while ensuring the overall accuracy of product image retrieval.

[0100] like Figure 3 As shown, Figure 3 : is a structural diagram of a control device provided by an embodiment of the present application. The present invention also provides a control device 300, including:

[0101] The processor 310 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0102] The memory 320 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 320 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 320, and the processor 310 calls and executes the commodity image retrieval method of the embodiment of this application;

[0103] Input / output interface 330, used to implement information input and output;

[0104] Communication interface 340, used to realize communication interaction between the apparatus and other devices, which can be realized by wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0105] bus 350 , which transmits information between the various components of the device (e.g., processor 310 , memory 320 , input / output interface 330 , and communication interface 340 );

[0106] The processor 310 , the memory 320 , the input / output interface 330 and the communication interface 340 are connected to each other in communication within the device via the bus 350 .

[0107] In addition, an embodiment of the present application further provides an electronic device, including the control device 300 of the above embodiment.

[0108] In addition, an embodiment of the present application further provides a storage medium, which is a computer-readable storage medium and stores a computer program. When the computer program is executed by a processor, the above-mentioned product image retrieval method is implemented.

[0109] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof. The device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and are implemented to be located in one place, or may also be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0110] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically include computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0111] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions under the shared conditions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A product image retrieval method, characterized in that: include: Acquire a first historical search record for each first commodity identifier, wherein the first historical search record includes first image feature information corresponding to the first commodity identifier, and the number of the first historical search record is at least 1; Inputting all the first image feature information corresponding to the first commodity identification into a preset classification model to obtain search attribute information corresponding to each of the first commodity identifications, wherein the search attribute information is used to indicate the supply status of the commodity corresponding to the first commodity identification; Determining target attribute information from all the search attribute information based on a preset storage condition, wherein the target attribute information is used to indicate the search attribute information that satisfies the storage condition; Selecting a target warehousing strategy from a preset warehousing strategy set based on the target attribute information, and saving the first image feature information of the corresponding first commodity identification to an image retrieval database based on the target warehousing strategy, wherein the warehousing strategy set includes a plurality of candidate warehousing strategies, and different candidate warehousing strategies correspond to different commodity categories; Product image retrieval is performed based on the image retrieval database.

2. The product image retrieval method according to claim 1, characterized in that: The entry condition is that the commodity category corresponding to the first commodity identifier is the first category or the second category, the first category is used to indicate seasonal commodities, the second category is used to indicate non-seasonal commodities, the first category corresponds to a first category identifier, and the second category corresponds to a second category identifier. The target attribute information is determined from all the search attribute information based on the preset entry condition, including: Acquire a third category identifier from the current search attribute information, wherein the third category identifier is used to indicate the commodity category of the first commodity identifier corresponding to the current search attribute information; Matching the third category identifier with the first category identifier and the second category identifier respectively to obtain corresponding first matching results and second matching results; When the first matching result indicates that the third category identifier matches the first category identifier, or the second matching result indicates that the third category identifier matches the second category identifier, the currently retrieved attribute information is determined as the target attribute information.

3. The product image retrieval method according to claim 2, characterized in that: Based on the target warehousing strategy, the first image feature information corresponding to the first commodity identifier is saved to an image retrieval database, including: When the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the second category, filtering all of the first image feature information to obtain a plurality of filtered second image feature information, wherein each of the second image feature information is different from each other; Performing clustering processing on the second image feature information based on a preset clustering algorithm to obtain a plurality of cluster clusters; The average feature value of each of the clusters is calculated, and all of the average feature values ​​are saved in the image retrieval database.

4. The product image retrieval method according to claim 3, characterized in that: The first image feature information corresponding to the first commodity identifier is clustered based on a preset clustering algorithm to obtain a plurality of clusters, including: Determining a reference quantity of the first historical search record corresponding to the first commodity identifier; Determine a search frequency level of the first commodity identifier based on the reference quantity, and determine a target cluster number based on the search frequency level, wherein a higher search frequency level corresponds to a larger value of the target cluster number; Based on the clustering algorithm, clustering processing is performed on the second image feature information corresponding to the first commodity identification to obtain the cluster clusters of the target cluster number.

5. The commodity image retrieval method according to claim 2, characterized in that: Based on the target warehousing strategy, the first image feature information corresponding to the first commodity identifier is saved to an image retrieval database, including: When the target attribute information indicates that the commodity category corresponding to the first commodity identifier is the first category, filtering all the first image feature information to obtain a plurality of filtered second image feature information, wherein each of the second image feature information is different from each other; All of the second image features are saved in the image retrieval database.

6. The commodity image retrieval method according to claim 2, characterized in that: After the first image feature information of the first commodity identifier corresponding to the first commodity identifier is saved in an image retrieval database based on the target warehousing strategy, the method further includes: When receiving warehousing demand information, the warehousing demand information includes a second commodity identifier, obtaining a second historical search record corresponding to the second commodity identifier, wherein the commodity category corresponding to the second commodity identifier is different from the first category and the second category, and the second historical search record includes third image feature information corresponding to the commodity identifier; All the third image feature information is saved in the image retrieval database.

7. The commodity image retrieval method according to claim 2, characterized in that: The first category corresponds to a reference category identifier and a reference timestamp range, the reference timestamp range changes continuously according to a preset period, and after the first image feature information corresponding to the first commodity identifier is saved to an image retrieval database based on the target warehousing strategy, the method further includes: Establishing an index record between the first commodity identifier, the corresponding commodity category identifier and the first image feature information in the image retrieval database; Determine a first target identifier from the first commodity identifiers in the image retrieval database, and determine a first timestamp range of the first target identifier, wherein the commodity category identifier corresponding to the first target identifier is the reference category identifier; Determine a second target identifier from the first commodity identifier in the image retrieval database, and determine the number of retrievals corresponding to the second target identifier; When the number of intersections between the first timestamp range and the reference timestamp range is 0, deleting the index record corresponding to the first target identifier in the image retrieval database; When the number of retrieval times is less than a preset threshold, the index record corresponding to the second target identifier in the image retrieval database is deleted.

8. A control device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the product image retrieval method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that: Includes the control device as claimed in claim 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the product image retrieval method according to any one of claims 1 to 7.

Images