Image recognition management system and method based on data analysis

Abstract

The present invention discloses an image recognition management system and method based on data analysis, which relates to the field of image recognition technology, and comprises an initial classification module, a labeling information analysis and storage module, an effective feedback index analysis module, an optimal labeling algorithm determination module and a real-time target image storage reminder module; the initial classification module is used to perform initial classification of target images on any image data set consisting of a plurality of images; the labeling information analysis and storage module is used to store associated images in a database corresponding to identification keywords based on labeling information; the effective feedback index analysis module is used to analyze the effective feedback index of each category of target images corresponding to each category of labeling algorithms; the optimal labeling algorithm determination module is used to determine the optimal labeling algorithm for each category of target images; and the real-time target image storage reminder module is used to analyze the optimal image category of the real-time target image and perform storage reminders for the optimal labeling algorithm corresponding to the optimal image category.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition management, and in particular to an image recognition management system and method based on data analysis. Background Art

[0002] Early image management was primarily based on simple database systems, storing images in a file system. The database recorded the image's file name, storage path, and basic annotation information (such as shooting time and location). This management approach met basic needs when the number of images was small and the application scenario was simple, but it was inefficient for retrieving and utilizing image content. With the development of image recognition technology, image recognition systems emerged. They allow users to search for images based on content features (such as color, texture, and shape). For example, a user could upload a landscape image containing a blue sky and green grass. The system would extract features such as the image's color histogram, search the database for images with similar color features, and return them to the user. However, this requires the use of numerous identification algorithms to identify the image data. Different identification algorithms can produce different identification results for the same type of image. Furthermore, when applied to real-world image databases, user search results based on databases can vary significantly depending on the identification results. Therefore, it is worth studying how to effectively identify various types of images using identification algorithms to achieve efficient user retrieval. Summary of the Invention

[0003] The purpose of the present invention is to provide an image recognition management system and method based on data analysis to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an image recognition management method based on data analysis, characterized in that the management method includes the following specific steps:

[0005] S1: Initial classification of target images for any image dataset consisting of several images;

[0006] The purpose of initial classification is to facilitate image structure analysis after generating image annotation information using image identification methods, as well as the rapid correspondence of similarities between images, which can effectively solve the problem of efficient classification under massive image data;

[0007] S2: Based on the classified target images, different types of image recognition and annotation algorithms are executed on all target images in each category, and the annotation information of the corresponding associated images generated after the annotation is completed is recorded. Based on the annotation information, the associated images are stored in a database corresponding to the identification keywords;

[0008] S3: Apply the database to the user side, extract the user's image retrieval events based on the database during the monitoring period, and obtain the retrieval factors and retrieval results recorded in the image retrieval event. Corresponding to the image categories in the image retrieval events, analyze the effective feedback index of each category of target images corresponding to various annotation algorithms;

[0009] S4: Based on the effective feedback index, determine the best labeling algorithm for each category of target images, analyze the best image category of the real-time target image, and use the best labeling algorithm corresponding to the best image category to store the labeling algorithm reminder.

[0010] Furthermore, the initial classification of the target image for any image dataset consisting of several images includes the following specific steps:

[0011] The target image refers to the image data containing the number of identifiable scene elements greater than or equal to n. The purpose of screening out the target image is to avoid the error caused by the single scene of the image itself in the dataset analysis. In addition, the target image of a single scene element is relatively simple to apply to the annotation algorithm and does not bring different retrieval efficiency to the user search.

[0012] Implementing a tagging algorithm that can automatically generate tagging information based on image content on a target image, and obtaining identification keywords output by the tagging algorithm regarding the target image; there are at least two types of tagging algorithms; identification keywords refer to elements annotated by the tagging algorithm after identifying image content; and mapping each target image to the identification keywords in all tagging algorithms to form tagging information;

[0013] The annotation information containing the most identification keywords is selected as the valid annotation information of the corresponding target image; the purpose of implementing different annotation algorithms for each target image is to improve the accuracy of identifying the target image, and the annotation information that can better summarize the image content is used as the valid annotation information; each target image in the image dataset is traversed to obtain the corresponding valid annotation information;

[0014] Sort each target image in the image dataset in ascending order according to the total number of identification keywords in the valid annotation information, and preliminarily divide the target images containing the same total number of identification keywords into the same category of target images and mark them as the first category of target images;

[0015] The average similarity of the identification keywords recorded by any two target images in each first-category target image is traversed and calculated. The average similarity refers to the average similarity value obtained by performing a one-to-one matching analysis of all the identification keywords recorded by the two target images; all target images corresponding to the average similarity difference in each first-category target image is marked as key target images; all first-category target images are traversed until the division is completed to generate key target images of various categories.

[0016] Furthermore, storing the associated images in a database corresponding to the identified keywords based on the annotation information includes the following specific steps:

[0017] Extract identification keywords from the annotation information of each type of key target image record, and combine the identification keywords in any number to generate keyword pairs under all conditions; the keyword pairs contain at least two identification keywords;

[0018] The key target image is then stored in a database containing the keyword pairs under each condition, so that when a user searches with the keyword pairs, the stored key target image is extracted based on the labeled keyword pairs in the database.

[0019] Furthermore, step S3 includes the following specific steps:

[0020] Step S31: An image retrieval event refers to an event in which a user inputs image-related keywords based on a database and selects a target image; a retrieval factor refers to the keyword content recorded in the search box; a retrieval result refers to the user retrieving the desired target image based on the database and recording the result as a successful retrieval; when the user processes the target image and the keyword content in the search box is unidirectional, the corresponding target image is output as the desired target image;

[0021] Unidirectionality means that the changes in keyword content must satisfy the unique corresponding keyword pair in the annotation information;

[0022] Step S32: Extract the retrieval factors that record all key target images in the image retrieval event, match the retrieval factors to the annotation algorithm corresponding to the annotation information, obtain the number U1 of image retrieval events that record successful retrieval under each annotation algorithm, and the total number U2 of image retrieval events executed using the annotation algorithm; define the time period when all key target images of various types are recorded in the image retrieval event as the effective period; use the formula:

[0023] Z=k1×(U1 / U2)+k2×(R2 / R1);

[0024] Calculate the effective feedback index Z of each type of annotation algorithm corresponding to each type of key target image; where R1 represents the number of keywords contained in the keyword pair with the highest success rate under the same algorithm, R2 represents the maximum number of keywords contained in the keyword pair generated under the same algorithm, and the success rate refers to the ratio of the number of successful retrieval events to the total number of image retrieval events for the same keyword pair; where k1 and k2 represent the corresponding reference coefficients.

[0025] The larger the Z is, the higher the achievement rate and efficiency of this type of annotation algorithm in realizing user-based database image retrieval.

[0026] Furthermore, step S4 includes the following:

[0027] Sort the various annotation algorithms in descending order according to the effective feedback index Z value, and select the annotation algorithm with the highest ranking as the best annotation algorithm for the target image of the corresponding category;

[0028] Analyze real-time target images for optimal image categories, including the following:

[0029] Determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match target images having the same number of real-time identification keywords to form a first image set;

[0030] Based on the first image set, the keyword similarity between each target image in the first image set and the real-time target image is calculated, and the target image corresponding to the maximum keyword similarity is selected as the optimal image category.

[0031] An image recognition management system based on data analysis, the management system includes an initial classification module, a labeling information analysis and storage module, an effective feedback index analysis module, an optimal labeling algorithm determination module and a real-time target image storage reminder module;

[0032] The initial classification module is used to perform initial classification of target images for any image dataset consisting of several images;

[0033] The tag information analysis and storage module is used to store the associated images in a database corresponding to the identification keywords based on the tag information;

[0034] The effective feedback index analysis module is used to analyze the effective feedback index of each category of target images corresponding to various types of labeling algorithms;

[0035] The optimal labeling algorithm determination module is used to determine the optimal labeling algorithm for each category of target images.

[0036] The real-time target image storage reminder module is used to analyze the best image category of the real-time target image and perform storage reminder of the annotation algorithm using the best annotation algorithm corresponding to the best image category.

[0037] Furthermore, the initial classification module includes a labeling algorithm implementation unit, an identification keyword generation unit, a valid labeling information analysis unit, a first-category target image marking unit, and a key target image division unit;

[0038] The annotation algorithm implementation unit is used to implement an annotation algorithm for the target image that can automatically generate annotation information based on the image content.

[0039] The identification keyword generation unit is used to obtain the identification keywords output by the annotation algorithm regarding the target image;

[0040] The effective annotation information analysis unit is used to form annotation information by corresponding to each target image with identification keywords in all annotation algorithms; and select the annotation information containing the most identification keywords as the effective annotation information corresponding to the target image;

[0041] The first-category target image marking unit is used to preliminarily classify target images containing the same total number of identification keywords into the same category of target images and mark them as first-category target images;

[0042] The key target image division unit is used to mark all target images corresponding to the average similarity difference value less than or equal to the difference threshold value in each first-category target image as key target images.

[0043] Furthermore, the effective feedback index analysis module includes a retrieval success event analysis unit and an effective feedback index calculation unit;

[0044] The retrieval success event analysis unit is used to extract the event that the user retrieves the target image required based on the database and records it as a retrieval success;

[0045] The effective feedback index calculation unit is used to calculate the effective feedback index of each type of key target image corresponding to each type of labeling algorithm.

[0046] Furthermore, the real-time target image storage reminder module includes a keyword number matching unit, a similarity analysis unit, and an optimal image category output unit;

[0047] The keyword number matching unit is used to determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match the target image with the same number of real-time identification keywords to form a first image set;

[0048] The similarity analysis unit is used to calculate the keyword similarity between each target image in the first image set and the real-time target image;

[0049] The best image category output unit is used to select the target image corresponding to the maximum keyword similarity as the best image category.

[0050] Compared with the prior art, the beneficial effects of the present invention are: the present invention identifies key target images that can be used as analysis objects by screening and analyzing images in the image database, and uses a multi-category identification algorithm to screen out identification keywords that maximize the representation of image features based on the key target images, and stores the images accordingly to achieve effective extraction when users search based on the image database. At the same time, the present invention analyzes the user's retrieval events based on the database to determine the best identification algorithm for each type of key target image. From the perspective of user retrieval efficiency, the data has a strong theoretical foundation and high data accuracy, thereby solving the problem of effectively giving the best identification algorithm for the corresponding image when new image data is input into the image library to achieve efficient retrieval by users. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a structural diagram of an image recognition management system based on data analysis in the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] Example: Figure 1 As shown, the present invention provides a data analysis-based image recognition management system and method technical solution, an image recognition management method based on data analysis, characterized in that the management method includes the following specific steps:

[0054] S1: Initial classification of target images for any image dataset consisting of several images;

[0055] The purpose of initial classification is to facilitate image structure analysis after generating image annotation information using image identification methods, as well as the rapid correspondence of similarities between images, which can effectively solve the problem of efficient classification under massive image data;

[0056] S2: Based on the classified target images, different types of image recognition and annotation algorithms are executed on all target images in each category, and the annotation information of the corresponding associated images generated after the annotation is completed is recorded. Based on the annotation information, the associated images are stored in a database corresponding to the identification keywords;

[0057] S3: Apply the database to the user side, extract the user's image retrieval events based on the database during the monitoring period, and obtain the retrieval factors and retrieval results recorded in the image retrieval event. Corresponding to the image categories in the image retrieval events, analyze the effective feedback index of each category of target images corresponding to various annotation algorithms;

[0058] S4: Based on the effective feedback index, determine the best labeling algorithm for each category of target images, analyze the best image category of the real-time target image, and use the best labeling algorithm corresponding to the best image category to store the labeling algorithm reminder.

[0059] The initial classification of the target image for any image dataset consisting of several images includes the following specific steps:

[0060] The target image refers to the image data containing identifiable scene elements greater than or equal to n. Generally, n is set to 3. The purpose of screening out target images is to avoid the error caused by the single scene of the image itself in the dataset analysis. In addition, the target image of a single scene element is relatively simple to apply to the annotation algorithm and does not bring different retrieval efficiency to the user search.

[0061] Implementing a tagging algorithm that can automatically generate tagging information based on image content on a target image, and obtaining identification keywords output by the tagging algorithm regarding the target image; there are at least two types of tagging algorithms; identification keywords refer to elements annotated by the tagging algorithm after identifying image content; and mapping each target image to the identification keywords in all tagging algorithms to form tagging information;

[0062] The annotation algorithms in this application can be target detection and recognition technology, scene understanding and semantic segmentation, and image feature extraction and label generation. Each type of algorithm can generate annotation information for the corresponding target image.

[0063] The annotation information containing the most identification keywords is selected as the valid annotation information of the corresponding target image; the purpose of implementing different annotation algorithms for each target image is to improve the accuracy of identifying the target image, and the annotation information that can better summarize the image content is used as the valid annotation information; each target image in the image dataset is traversed to obtain the corresponding valid annotation information;

[0064] Sort each target image in the image dataset in ascending order according to the total number of identification keywords in the valid annotation information, and preliminarily divide the target images containing the same total number of identification keywords into the same category of target images and mark them as the first category of target images;

[0065] The average similarity of the identification keywords recorded by any two target images in each first-category target image is traversed and calculated. The average similarity refers to the average similarity value obtained by performing a one-to-one matching analysis of all the identification keywords recorded by the two target images; all target images corresponding to the average similarity difference in each first-category target image is marked as key target images; all first-category target images are traversed until the division is completed to generate key target images of various categories.

[0066] Storing the associated images in a database corresponding to the identified keywords based on the annotation information includes the following specific steps:

[0067] Extract identification keywords from the annotation information of each type of key target image record, and combine the identification keywords in any number to generate keyword pairs under all conditions; the keyword pairs contain at least two identification keywords;

[0068] The key target image is then stored in a database containing the keyword pairs under each condition, so that when a user searches with the keyword pairs, the stored key target image is extracted based on the labeled keyword pairs in the database.

[0069] As shown in the embodiment: there is a key target image 1, and the algorithms for image recognition and labeling of the key target image 1 are algorithm a and algorithm b;

[0070] Algorithm a generates annotation information based on the key target image 1: keyword 1 → keyword 2 → keyword 3; then, any number of combinations can be used to generate keyword pairs under all conditions, including the following:

[0071] Keyword 1 → Keyword 2, Keyword 1 → Keyword 3, Keyword 2 → Keyword 3, Keyword 1 → Keyword 2 → Keyword 3;

[0072] Then the key target image 1 is stored in the keyword pair marked as "keyword 1→keyword 2" in the database; similarly, the other three keyword pairs will also store the key target image 1, but the retrieval method is different.

[0073] Step S3 includes the following specific steps:

[0074] Step S31: An image retrieval event refers to an event in which a user inputs image-related keywords based on a database and selects a target image; a retrieval factor refers to the keyword content recorded in the search box; a retrieval result refers to the user retrieving the desired target image based on the database and recording the result as a successful retrieval; when the user performs a processing operation on the target image and the keyword content in the search box is unidirectional, the corresponding target image is output as the desired target image; processing operations include copying, downloading, etc.

[0075] Unidirectionality means that the changes in keyword content must satisfy the unique corresponding keyword pair in the annotation information;

[0076] Step S32: Extract the retrieval factors that record all key target images in the image retrieval event, match the retrieval factors to the annotation algorithm corresponding to the annotation information, obtain the number U1 of image retrieval events that record successful retrieval under each annotation algorithm, and the total number U2 of image retrieval events executed using the annotation algorithm; define the time period when all key target images of various types are recorded in the image retrieval event as the effective period; use the formula:

[0077] Z=k1×(U1 / U2)+k2×(R2 / R1);

[0078] Calculate the effective feedback index Z of each type of annotation algorithm corresponding to each type of key target image; where R1 represents the number of keywords contained in the keyword pair with the highest success rate under the same algorithm, R2 represents the maximum number of keywords contained in the keyword pair generated under the same algorithm, and the success rate refers to the ratio of the number of successful retrieval events to the total number of image retrieval events for the same keyword pair; where k1 and k2 represent the corresponding reference coefficients.

[0079] The larger the Z is, the higher the achievement rate and efficiency of this type of annotation algorithm in realizing user-based database image retrieval.

[0080] Step S4 includes the following:

[0081] Sort the various annotation algorithms in descending order according to the effective feedback index Z value, and select the annotation algorithm with the highest ranking as the best annotation algorithm for the target image of the corresponding category;

[0082] Analyze real-time target images for optimal image categories, including the following:

[0083] Determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match target images having the same number of real-time identification keywords to form a first image set;

[0084] Based on the first image set, the keyword similarity between each target image in the first image set and the real-time target image is calculated, and the target image corresponding to the maximum keyword similarity is selected as the optimal image category.

[0085] An image recognition management system based on data analysis, the management system includes an initial classification module, a labeling information analysis and storage module, an effective feedback index analysis module, an optimal labeling algorithm determination module and a real-time target image storage reminder module;

[0086] The initial classification module is used to perform initial classification of target images for any image dataset consisting of several images;

[0087] The tag information analysis and storage module is used to store the associated images in a database corresponding to the identification keywords based on the tag information;

[0088] The effective feedback index analysis module is used to analyze the effective feedback index of each category of target images corresponding to various types of labeling algorithms;

[0089] The optimal labeling algorithm determination module is used to determine the optimal labeling algorithm for each category of target images.

[0090] The real-time target image storage reminder module is used to analyze the best image category of the real-time target image and perform storage reminder of the annotation algorithm using the best annotation algorithm corresponding to the best image category.

[0091] The initial classification module includes a labeling algorithm implementation unit, an identification keyword generation unit, an effective labeling information analysis unit, a first-category target image marking unit, and a key target image division unit;

[0092] The annotation algorithm implementation unit is used to implement an annotation algorithm for the target image that can automatically generate annotation information based on the image content.

[0093] The identification keyword generation unit is used to obtain the identification keywords output by the annotation algorithm regarding the target image;

[0094] The effective annotation information analysis unit is used to form annotation information by corresponding to each target image with identification keywords in all annotation algorithms; and select the annotation information containing the most identification keywords as the effective annotation information corresponding to the target image;

[0095] The first-category target image marking unit is used to preliminarily classify target images containing the same total number of identification keywords into the same category of target images and mark them as first-category target images;

[0096] The key target image division unit is used to mark all target images corresponding to the average similarity difference value less than or equal to the difference threshold value in each first-category target image as key target images.

[0097] The effective feedback index analysis module includes a retrieval success event analysis unit and an effective feedback index calculation unit;

[0098] The retrieval success event analysis unit is used to extract the event that the user retrieves the target image required based on the database and records it as a retrieval success;

[0099] The effective feedback index calculation unit is used to calculate the effective feedback index of each type of key target image corresponding to each type of labeling algorithm.

[0100] The real-time target image storage reminder module includes a keyword number matching unit, a similarity analysis unit and an optimal image category output unit;

[0101] The keyword number matching unit is used to determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match the target image with the same number of real-time identification keywords to form a first image set;

[0102] The similarity analysis unit is used to calculate the keyword similarity between each target image in the first image set and the real-time target image;

[0103] The best image category output unit is used to select the target image corresponding to the maximum keyword similarity as the best image category.

[0104] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An image recognition management method based on data analysis, characterized by: The management method comprises the following specific steps: S1: Initial classification of target images for any image dataset consisting of several images; The initial classification of a target image for any image dataset consisting of a plurality of images comprises the following specific steps: The target image refers to the image corresponding to the case where the number of identifiable scene elements contained in the image data is greater than or equal to n; Implementing a tagging algorithm capable of automatically generating tagging information based on image content on a target image, and obtaining identification keywords output by the tagging algorithm regarding the target image; the tagging algorithms are of at least two types; the identification keywords are elements annotated by the tagging algorithm after identifying the image content; and composing tagging information corresponding to each target image by mapping the identification keywords in all tagging algorithms; Select the annotation information containing the most identification keywords as the valid annotation information of the corresponding target image; traverse each target image in the image data set to obtain the corresponding valid annotation information; Sort each target image in the image dataset in ascending order according to the total number of identification keywords in the valid annotation information, and preliminarily divide the target images containing the same total number of identification keywords into the same category of target images and mark them as the first category of target images; Calculate the average similarity of identification keywords recorded by any two target images in each first-category target image, where the average similarity refers to the average similarity obtained by performing a one-to-one matching analysis of all identification keywords recorded by the two target images; mark all target images corresponding to the average similarity difference less than or equal to a difference threshold in each first-category target image as key target images; traverse all first-category target images until the division is completed to generate key target images of each category; S2: Based on the classified target images, different types of image recognition and annotation algorithms are executed on all target images in each category, and the annotation information of the corresponding associated images generated after the annotation is completed is recorded. Based on the annotation information, the associated images are stored in a database corresponding to the identification keywords; S3: Apply the database to the user side, extract the user's image retrieval events based on the database during the monitoring period, and obtain the retrieval factors and retrieval results recorded in the image retrieval event. Corresponding to the image categories in the image retrieval events, analyze the effective feedback index of each category of target images corresponding to various annotation algorithms; S4: Based on the effective feedback index, determine the best labeling algorithm for each category of target images, analyze the best image category of the real-time target image, and use the best labeling algorithm corresponding to the best image category to store the labeling algorithm reminder.

2. The image recognition management method based on data analysis according to claim 1, characterized in that: The step of storing the associated images in a database corresponding to the identified keywords based on the annotation information includes the following specific steps: Extracting identification keywords from the annotation information of each type of key target image record, and combining the identification keywords in any number to generate keyword pairs under all conditions; the keyword pairs contain at least two identification keywords; The key target image is then stored in a database containing the keyword pairs under each condition, so that when a user searches with the keyword pairs, the stored key target image is extracted based on the labeled keyword pairs in the database.

3. The image recognition management method based on data analysis according to claim 1, characterized in that: The step S3 includes the following specific steps: Step S31: The image retrieval event refers to an event in which a user inputs image-related keywords based on a database and selects a target image; the retrieval factor refers to the keyword content recorded in the search box; the retrieval result refers to the user retrieving the desired target image based on the database and recording it as a successful retrieval; when the user processes the target image and the keyword content in the search box is unidirectional, the corresponding target image is output as the desired target image; The unidirectionality means that all changes in keyword content satisfy the unique corresponding keyword pair in the annotation information; Step S32: Extract the retrieval factors that record all key target images in the image retrieval event, match the retrieval factors to the annotation algorithm corresponding to the annotation information, obtain the number U1 of image retrieval events that record successful retrieval under each annotation algorithm, and the total number U2 of image retrieval events executed using the annotation algorithm; define the time period when all key target images of various types are recorded in the image retrieval event as the effective period; use the formula: Z=k1×(U1 / U2)+k2×(R2 / R1); Calculate the effective feedback index Z of each type of annotation algorithm for each type of key target image; where R1 represents the number of keywords contained in the keyword pair with the highest success rate under the same algorithm, and R2 represents the maximum number of keywords contained in the keyword pair generated under the same algorithm. The success rate refers to the ratio of the number of successful retrieval events to the total number of image retrieval events for the same keyword pair; Where k1 and k2 represent the corresponding reference coefficients.

4. The image recognition management method based on data analysis according to claim 3, characterized in that: The step S4 includes the following: Sort the various annotation algorithms in descending order according to the effective feedback index Z value, and select the annotation algorithm with the highest ranking as the best annotation algorithm for the target image of the corresponding category; The best image categories for real-time target image analysis include the following: Determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match target images having the same number of real-time identification keywords to form a first image set; Based on the first image set, the keyword similarity between each target image in the first image set and the real-time target image is calculated, and the target image corresponding to the maximum keyword similarity is selected as the optimal image category.

5. An image recognition management system based on data analysis, such as using the image recognition management method based on data analysis according to any one of claims 1 to 4, characterized in that: The management system includes an initial classification module, a labeling information analysis and storage module, an effective feedback index analysis module, an optimal labeling algorithm determination module and a real-time target image storage reminder module; The initial classification module is used to perform initial classification of target images on any image data set consisting of a plurality of images; The tag information analysis and storage module is used to store the associated images in a database corresponding to the identification keywords based on the tag information; The effective feedback index analysis module is used to analyze the effective feedback index of each category of target images corresponding to various types of labeling algorithms; The optimal labeling algorithm determination module is used to determine the optimal labeling algorithm for each category of target images. The real-time target image storage reminder module is used to analyze the best image category of the real-time target image and perform a storage reminder of the annotation algorithm using the best annotation algorithm corresponding to the best image category.

6. The image recognition management system based on data analysis according to claim 5, characterized in that: The initial classification module includes a labeling algorithm implementation unit, an identification keyword generation unit, an effective labeling information analysis unit, a first-category target image marking unit, and a key target image division unit; The annotation algorithm implementation unit is used to implement an annotation algorithm on the target image that can automatically generate annotation information based on the image content. The identification keyword generation unit is used to obtain identification keywords output by the annotation algorithm regarding the target image; The effective annotation information analysis unit is used to form annotation information by using the identification keywords in all the annotation algorithms corresponding to each target image; and select the annotation information containing the most identification keywords as the effective annotation information corresponding to the target image; The first-category target image marking unit is used to preliminarily classify target images containing the same total number of identification keywords into the same category of target images and mark them as first-category target images; The key target image division unit is used to mark all target images corresponding to the average similarity difference value less than or equal to the difference threshold value in each first-category target image as key target images.

7. The image recognition management system based on data analysis according to claim 6, characterized in that: The effective feedback index analysis module includes a retrieval success event analysis unit and an effective feedback index calculation unit; The retrieval success event analysis unit is used to extract the event that the user retrieves the required target image based on the database and records it as a retrieval success; The effective feedback index calculation unit is used to calculate the effective feedback index of each type of key target image corresponding to each type of labeling algorithm.

8. The image recognition management system based on data analysis according to claim 6, characterized in that: The real-time target image storage reminder module includes a keyword number matching unit, a similarity analysis unit and an optimal image category output unit; The keyword number matching unit is used to determine valid annotation information for the real-time target image; extract the number of identification keywords recorded in the valid annotation information, and match the target image with the same number of real-time identification keywords to form a first image set; The similarity analysis unit is used to calculate the keyword similarity between each target image in the first image set and the real-time target image; The optimal image category output unit is used to select the target image corresponding to the maximum keyword similarity value as the optimal image category.

Images