Aesthetic image evaluation system and detection method thereof

Abstract

The invention relates to the technical field of image recognition, and discloses an aesthetic image evaluation system and a detection method thereof. The aesthetic image evaluation system comprises four modules, namely a personnel management module, a task management module, a credit management module, and a data management module. Through an online evaluation system based on the coordinated cooperation of the four modules, real-time multi-personnel evaluation in different locations, synchronous statistical evaluation information, and the problem of solving regional difference evaluation can be realized; the detection method of the aesthetic image evaluation system comprises a random repetition detection algorithm, an anchor library sampling detection algorithm, and an inter-personnel detection algorithm. Through these three algorithms, the reliable, accurate, and stable operation of the aesthetic image evaluation system can be guaranteed, and the evaluation cost can be greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, in particular to an aesthetic image evaluation system and a detection method thereof. Background Art

[0002] Aesthetic image evaluation is an interdisciplinary subject that includes cognitive psychology, computer vision, machine learning, etc. Computer evaluation systems based on machine learning usually require high-quality manually evaluated data for model training. However, at present, there are very few aesthetic image libraries that have been evaluated, and the data scale and quality of most image libraries cannot meet the needs of high-quality business models.

[0003] Currently, there are two existing methods for image aesthetics evaluation: evaluation systems based on machine learning and stand-alone evaluation software that is only suitable for small teams. The two commonly used methods have the following problems:

[0004] Computer evaluation system based on machine learning: The evaluation of image aesthetics is subjective. The computer evaluation system based on machine learning uses computers to completely replace manual evaluation, so the accuracy is low. The evaluation of image aesthetics is affected by geographical distribution, which is not abnormal. The individual's cognition of aesthetics is related to the aesthetic evaluation system of the group environment in which the individual is located. There are differences in the cognition of aesthetics between individuals with different aesthetic evaluation systems. At the same time, due to different experiences, individuals have certain aesthetic preferences for different objects. Therefore, it is difficult for a computer evaluation system based on machine learning to guarantee a more ideal evaluation level. At the same time, due to the differences between humans and machines, the evaluation accuracy is also difficult to use as a reference.

[0005] The stand-alone evaluation software: Although it improves the efficiency and accuracy of the evaluation and can realize some project management functions, it is only suitable for small teams: that is, 1 to 2 evaluators collaborate to complete it. It is not competent for team collaboration that requires more than a dozen or even more evaluators. For example, during the evaluation process, it is necessary to jointly evaluate an image. Since the stand-alone evaluation program cannot be connected to the Internet, to achieve this function, it is necessary to copy the evaluation information of two people to the same computer for comparison. Secondly, the cause of the abnormal samples in the evaluation process is whether it is an evaluation error or the personal preference of the evaluator. In addition, the problem of reduced evaluation quality caused by long-term evaluation cannot be avoided. Therefore, it is difficult to distinguish whether the evaluator has truly and effectively evaluated the image, and thus it is impossible to obtain better training data; this method lacks effective exception handling and abnormal avoidance means, resulting in high maintenance and management costs.

[0006] In summary, the current data evaluation system dedicated to image aesthetics is difficult to meet both business needs and business scale. Therefore, in order to make up for this shortcoming, it is necessary to provide a system that can be used by a large number of people to evaluate the same aesthetic image at the same time, and can obtain more accurate evaluation results, avoid the problem of unsatisfactory evaluation results of computer evaluation systems based on machine learning, and at the same time ensure the accuracy of the evaluators' evaluations through detection methods, and can separately process and store the evaluation results of differentiated groups of people for aesthetic analysis. Summary of the invention

[0007] In view of the shortcomings of the above-mentioned existing production technologies, the present application provides an aesthetic image evaluation system and a detection method thereof, so that multiple people can evaluate the same image at different locations at the same time. At the same time, the evaluation information can be synchronized in real time through the database, which can solve the problem of regional difference evaluation.

[0008] The technical solution adopted by the present invention is as follows:

[0009] An aesthetic image evaluation system, characterized in that it comprises four modules, namely a personnel management module, a task management module, a credit management module, and a data management module;

[0010] Personnel management module: used to provide registration, login and logout functions; collect basic information of users after logging into the system; store the corresponding data generated by users in the system; assign different usage permissions to users with different login identities;

[0011] Task management module: including task submission module, task collection module and task verification module; the task submission module is used by the evaluation task provider to provide basic task information; the task collection module is for the evaluators to provide a way to collect tasks; the task verification module is used to verify the completion of the evaluation task;

[0012] Credit management module: including credit detection module, credit rating module and credit warning module; the credit detection module is used to detect in real time whether the evaluation results of the evaluator are reliable, and adjust the credit rating of the evaluator in real time according to the detection results; the credit rating module is used to assess the evaluation quality of the evaluator; the credit warning module is used to remind the evaluator that the quality of the evaluated data is not up to standard.

[0013] Data management module: includes data processing module, data storage module and data transmission module; the data processing module is used to generate real-time data storage format for the data generated by the task management module; the data storage module is used to solidify the evaluation data and use it for query; the data transmission module is used for synchronous use of data between different terminal modules.

[0014] Furthermore, in the personnel management module, basic information includes age, gender, and region; stored information includes credit status and assessment progress; for logged-in users, the same user should be able to correspond to the assessment task provider, evaluator, and manager at the same time.

[0015] Furthermore, the task submission module provides basic information about the task, including: images to be evaluated, evaluation content, evaluation methods, and evaluation standards; the task collection module provides task collection methods, including: system allocation and free collection; the task verification module verifies the completion of the evaluation task, including: verifying whether the evaluation result is consistent with the number and ID of the evaluation task, verifying whether the evaluation data conforms to the format and is stored intact, and verifying whether the evaluator has the evaluation authority;

[0016] Furthermore, the credit rating module can dynamically allocate evaluation weights according to the credit rating of the evaluator; the credit warning module can refuse to evaluate an evaluator who has failed multiple times in a short period of time or whose credit rating is too low. Methods for restoring the normal evaluation status include evaluation examinations and application approval.

[0017] Furthermore, the data processing module generates data based on the task management module, including evaluation tasks and evaluation results of evaluators; the solidification method of the data storage module includes database storage and file storage; the data transmission module includes evaluation data when using data synchronously between different terminals, transmits the evaluation data to the same storage terminal, and retrieves the evaluation data for training.

[0018] A detection method for an aesthetic image evaluation system, comprising a random repetition detection algorithm, an anchor library sampling detection algorithm, and an inter-person detection algorithm;

[0019] The specific steps of the random repetition detection algorithm are:

[0020] Step A1: The evaluator completes the evaluation of N images;

[0021] Step A2: randomly select the images evaluated by the evaluators during this evaluation process for re-evaluation;

[0022] Step A3: Compare the two evaluation results to see if they are consistent;

[0023] Step A4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point;

[0024] The specific steps of the anchor library sampling detection algorithm are as follows:

[0025] Step B1: The evaluator completes the evaluation of N images;

[0026] Step B2: Extract the two images with the best and worst evaluation results from the anchor point library to re-evaluate the evaluators;

[0027] Step B3: Determine whether the re-evaluated result is consistent with the evaluation results of the two images extracted from the anchor point library;

[0028] Step B4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point;

[0029] The specific steps of the inter-person detection algorithm are:

[0030] Step C1: The evaluator completes the evaluation of N images;

[0031] Step C2: randomly select an image that has been evaluated by M evaluators for re-evaluation;

[0032] Step C3: record the difference between the evaluator and the M evaluators who have been evaluated as β. If β exceeds the preset difference set by the back-end staff, mark the evaluation result as abnormal and jump to step C4; if β does not exceed the preset value set by the back-end staff, the test is passed, and the batch credit is increased by 1 point and the total credit is increased by 1 point.

[0033] Step C4: the evaluator continues to evaluate K images, and the images extracted in step 2 are re-evaluated by the evaluator, and the process goes to step C5;

[0034] Step C5: Determine whether the results of the two re-evaluations in step C2 and step C4 are consistent. If they are consistent, the test passes. If they are inconsistent, record the difference m between the result of the re-evaluation in step C4 and the results of the M evaluators who have been evaluated in step C2, and compare m with the evaluation abnormality value γ set in the background. If m is greater than γ, reduce the batch credit by 1 point and the total credit by 1 point; if m is less than γ, reduce the batch credit by 0.5 points and the total credit by 0.5 points.

[0035] Furthermore, in the random repetition detection algorithm, anchor library sampling detection algorithm, and inter-personnel detection algorithm, the re-evaluation of the evaluators is performed by the background, and the evaluators are not aware of the re-evaluation. The system interface during the re-evaluation is consistent with the normal evaluation.

[0036] Furthermore, when the batch credit and total credit are lower than the warning line set by the background, the background forces the user to log off or issue a warning.

[0037] Furthermore, the evaluation scores detected by any of the random repetition detection algorithm, the anchor library sampling detection algorithm, and the inter-person detection algorithm will be entered into the scoring database.

[0038] Furthermore, the numbers N, K, and γ are all adjusted and set by the backend staff.

[0039] The beneficial effects of the present invention are as follows:

[0040] The advantages of an aesthetic image evaluation system provided by the present invention are as follows:

[0041] 1) By setting up personnel management module, task management module, credit management module and data management module, it is possible to realize multi-person evaluation in different locations, and at the same time, the evaluation information is synchronized in real time through the database, thereby reducing the time and manpower costs of the evaluation, and avoiding the human errors that may occur when manually counting the evaluation data of multiple machines, thereby ensuring the validity of the data.

[0042] 2) It can solve the problem of regional difference assessment. Through the registration information of the assessors and the storage logic design of the assessment results, it can retain the assessors' original information such as region, interests and hobbies that can be used for regional analysis.

[0043] The advantages of the aesthetic image evaluation system detection method provided by the present invention are as follows:

[0044] Based on the above-mentioned online evaluation system, a detection method is provided for detecting the evaluation quality of evaluators, including a random repetition detection algorithm, an anchor library sampling detection algorithm and an inter-personnel detection algorithm, thereby ensuring the reliability of the evaluation data set to the greatest extent. The data set evaluated by the aesthetic evaluation system provided by the present invention can well support subsequent related work. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the aesthetic evaluation system of the present invention;

[0046] Figure 2 This is an evaluation flow chart based on the random repeated sampling detection method in the present invention;

[0047] Figure 3 It is an evaluation flow chart of the anchor library sampling detection method in the present invention;

[0048] Figure 4 This is a flow chart of the evaluation based on the inter-person sampling detection method in the present invention. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention will be described below in conjunction with the accompanying drawings.

[0050] An aesthetic image evaluation system, characterized in that it comprises four modules, namely a personnel management module, a task management module, a credit management module, and a data management module;

[0051] Personnel management module: used to provide registration, login and logout functions; collect basic information of users after logging into the system; store the corresponding data generated by users in the system; assign different usage permissions to users with different login identities;

[0052] Task management module: including task submission module, task collection module and task verification module; the task submission module is used by the evaluation task provider to provide basic task information; the task collection module is for the evaluators to provide a way to collect tasks; the task verification module is used to verify the completion of the evaluation task;

[0053] Credit management module: including credit detection module, credit rating module and credit warning module; the credit detection module is used to detect in real time whether the evaluation results of the evaluator are reliable, and adjust the credit rating of the evaluator in real time according to the detection results; the credit rating module is used to assess the evaluation quality of the evaluator; the credit warning module is used to remind the evaluator that the quality of the evaluated data is not up to standard.

[0054] Data management module: includes data processing module, data storage module and data transmission module; the data processing module is used to generate real-time data storage format for the data generated by the task management module; the data storage module is used to solidify the evaluation data and use it for query; the data transmission module is used for synchronous use of data between different terminal modules.

[0055] Furthermore, in the personnel management module, basic information includes age, gender, and region; stored information includes credit status and assessment progress; for logged-in users, the same user should be able to correspond to the assessment task provider, evaluator, and manager at the same time.

[0056] Furthermore, the task submission module provides basic information about the task, including: images to be evaluated, evaluation content, evaluation methods, and evaluation standards; the task collection module provides task collection methods, including: system allocation and free collection; the task verification module verifies the completion of the evaluation task, including: verifying whether the evaluation result is consistent with the number and ID of the evaluation task, verifying whether the evaluation data conforms to the format and is stored intact, and verifying whether the evaluator has the evaluation authority;

[0057] Furthermore, the credit rating module can dynamically allocate evaluation weights according to the credit rating of the evaluator; the credit warning module can refuse to evaluate an evaluator who has failed multiple times in a short period of time or whose credit rating is too low. Methods for restoring the normal evaluation status include evaluation examinations and application approval.

[0058] Furthermore, the data processing module generates data based on the task management module, including evaluation tasks and evaluation results of evaluators; the solidification method of the data storage module includes database storage and file storage; the data transmission module includes evaluation data when using data synchronously between different terminals, transmits the evaluation data to the same storage terminal, and retrieves the evaluation data for training.

[0059] like Figure 1 In the embodiment shown, the workflow of an aesthetic image evaluation system in the present invention is illustrated: the personnel management module enables the evaluator to register or log in to the evaluation system, the evaluator can freely receive or passively assign evaluation tasks from the task management module, after receiving the evaluation task, switch to the evaluation interface, the evaluator evaluates the image and makes the result of this evaluation; during the evaluation process, the credit management module detects the result of the evaluator's evaluation through the detection algorithm, and dynamically adjusts the credit rating. When the credit score is lower than the preset value, such as 60 points, it is forced to go offline, and when it is lower than 80 points, a warning is given; after a whole batch of evaluations is completed, the evaluation results are summarized and stored through the data management module. In particular, during the login and registration process, the evaluator can access the URL of this evaluation system on any terminal; in the process of assigning evaluation tasks, based on the pre-built evaluation server, the evaluation tasks can be assigned to the evaluator in real time, and the evaluation results can be fed back in real time; in the credit detection process, based on the real-time interactive evaluation server and the evaluation terminal, multiple algorithms can be used to perform credit detection on the abnormal evaluation results of the evaluator separately or simultaneously. First, based on the data evaluated by the existing appraisers, determine whether there are any abnormalities in the evaluation results of the appraisers, and conduct credit checks on the evaluation results with abnormalities.

[0060] It should be noted that, according to the characteristics of aesthetic image evaluation, one of the data storage formats provided in the embodiment of the present invention is as follows:

[0061] According to the evaluator's registration information, the evaluated image content and the evaluated image aspects, the naming format of the csv is: evaluated factors + evaluation method + country and region + portrait / landscape + day / night + evaluator ID.csv, and is stored under "evaluated factors / / evaluation method / / country and region / / portrait / landscape / / day / night", named in English, where the evaluation methods include: 1-5 points evaluation (1to5) and comparative evaluation (0or1), evaluated factors (proposed factors), country and region (proposed some regions), portrait or landscape (human, scene), day or night (day, night).

[0062] A detection method for an aesthetic image evaluation system, comprising a random repetition detection algorithm, an anchor library sampling detection algorithm, and an inter-person detection algorithm;

[0063] The specific steps of the random duplicate detection algorithm are:

[0064] Step A1: The evaluator completes the evaluation of N images;

[0065] Step A2: randomly select the images evaluated by the evaluators during this evaluation process for re-evaluation;

[0066] Step A3: Compare the two evaluation results to see if they are consistent;

[0067] Step A4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point;

[0068] like Figure 2 In the illustrated embodiment, during the evaluation of a batch of images, after the evaluator has completed the evaluation of N images (N is the number set by the background administrator), images that have been evaluated in the batch are randomly selected according to the evaluator ID and presented again for the evaluator to evaluate (completed in the background, without prompting the evaluator, and the display effect is a normal evaluation screen), and the evaluator's evaluation result this time is compared with the previous evaluation result. If the two results are consistent, the test is passed, the total credit is +1, and the credit of this batch is +1. If the two results are inconsistent, the credit test is judged to have failed, and the credit of this batch is -1, and the total credit is -1.

[0069] The specific steps of the anchor library sampling detection algorithm are as follows:

[0070] Step B1: The evaluator completes the evaluation of N images;

[0071] Step B2: Extract the two images with the best and worst evaluation results from the anchor point library to re-evaluate the evaluators;

[0072] Step B3: Determine whether the re-evaluated result is consistent with the evaluation results of the two images extracted from the anchor point library;

[0073] Step B4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point;

[0074] like Figure 3In the illustrated embodiment, based on the image evaluation process of the anchor library sampling detection algorithm, after the evaluator completes the evaluation of N images (N is the number set by the background administrator), the worst and best images are randomly selected from the anchor library and disguised as normal images to be evaluated, and the evaluator is asked to evaluate (completed in the background, without prompting the evaluator, and the display effect is a normal evaluation screen), and the consistency of the evaluator's evaluation results is compared with the evaluated results in the anchor library. If the two results are consistent, the verification is passed, the total credit is +1, and the credit of this batch is +1. If the two results are inconsistent, it is determined that the credit test fails, the credit of this batch is -1, and the total credit is -1.

[0075] The specific steps of the inter-person detection algorithm are as follows:

[0076] Step C1: The evaluator completes the evaluation of N images;

[0077] Step C2: randomly select an image that has been evaluated by M evaluators for re-evaluation;

[0078] Step C3: record the difference between the evaluator and the M evaluators who have been evaluated as β. If β exceeds the preset difference set by the back-end staff, mark the evaluation result as abnormal and jump to step C4; if β does not exceed the preset value set by the back-end staff, the test is passed, and the batch credit is increased by 1 point and the total credit is increased by 1 point.

[0079] Step C4: the evaluator continues to evaluate K images, and the images extracted in step 2 are re-evaluated by the evaluator, and the process goes to step C5;

[0080] Step C5: Determine whether the results of the two re-evaluations in step C2 and step C4 are consistent. If they are consistent, the test passes. If they are inconsistent, record the difference m between the result of the re-evaluation in step C4 and the results of the M evaluators who have been evaluated in step C2, and compare m with the evaluation abnormality value γ set in the background. If m is greater than γ, reduce the batch credit by 1 point and the total credit by 1 point; if m is less than γ, reduce the batch credit by 0.5 points and the total credit by 0.5 points.

[0081] like Figure 4In the embodiment shown, based on the evaluation process of the inter-personnel detection algorithm, after the evaluator has completed the evaluation of N images (N is the number set by the backend administrator), an image that has been evaluated by at least M evaluators is selected (M is the number set by the backend administrator), and it is detected whether there is a significant difference between the scores of the new evaluator and the previous M evaluators (by comparing the difference β with the backend evaluation abnormality γ, and exceeding this interval is considered to be a significant difference, such as: the difference β = 5 points, the backend preset evaluation abnormality γ is ≤ 4 points, then it is determined to be a significant difference), then the evaluation result is marked as abnormal, and in the random number K After the images of the batch are received (K is the number set by the backend administrator), the evaluator is asked to evaluate again and the two evaluation results are compared. If the two evaluation results are consistent, the evaluator's evaluation quality is considered qualified, and this result is added to the existing evaluation database, the total credit is +1, and the credit of this batch is +1; if the two evaluation results are inconsistent, and the difference m of the second evaluation does not exceed the evaluation outlier value γ (the evaluation outlier value γ is set by the backend adjustment), then the credit of this batch is -0.5, and the total credit is -0.5; if the two evaluation results are inconsistent and the difference m of the second evaluation exceeds the evaluation outlier value γ, then the credit of this batch is -1, and the total credit is -1.

[0082] Furthermore, in the random repetition detection algorithm, anchor library sampling detection algorithm, and inter-personnel detection algorithm, the re-evaluation of the evaluators is performed by the background, and the evaluators are not aware of the re-evaluation. The system interface during the re-evaluation is consistent with the normal evaluation.

[0083] Furthermore, when the batch credit and total credit are lower than the warning line set by the background, the background forces the user to log off or issue a warning.

[0084] Furthermore, the evaluation scores detected by any of the random repetition detection algorithm, the anchor library sampling detection algorithm, and the inter-person detection algorithm will be entered into the scoring database.

[0085] Furthermore, the numbers N, K, and γ are all adjusted and set by the backend staff.

[0086] The above description is an explanation of the present invention, not a limitation of the present invention. The scope of the present invention is defined in the claims. Any form of modification may be made within the scope of protection of the present invention.

Claims

1. An aesthetic image evaluation system, characterized in that: It includes four modules, namely personnel management module, task management module, credit management module and data management module; Personnel management module: used to provide registration, login and logout functions; collect basic information of users after logging into the system; store the corresponding data generated by users in the system; assign different usage permissions to users with different login identities; Task management module: including task submission module, task collection module and task verification module; the task submission module is used by the assessment task provider to provide basic task information; the task collection module is for the assessors to provide a way to collect the task; The task verification module is used to verify the completion of the evaluation task; Credit management module: including credit detection module, credit rating module and credit warning module; the credit detection module is used to detect in real time whether the evaluation results of the evaluator are reliable, and adjust the credit rating of the evaluator in real time according to the detection results; The credit rating module is used to assess the quality of the assessment by the assessor; the credit warning module is used to remind the assessor that the quality of the assessed data is not up to standard; Data management module: includes data processing module, data storage module and data transmission module; the data processing module is used to generate real-time data storage format for the data generated by the task management module; the data storage module is used to solidify the evaluation data and use it for query; the data transmission module is used for synchronous use of data between different terminal modules.

2. An aesthetic image evaluation system as claimed in claim 1, characterized in that: In the personnel management module, basic information includes age, gender, and region; stored information includes credit status and assessment progress; for logged-in users, the same user should be able to correspond to the assessment task provider, assessor, and manager at the same time.

3. An aesthetic image evaluation system as claimed in claim 1, characterized in that: The task submission module provides basic task information including: images to be evaluated, evaluation content, evaluation methods, and evaluation standards; the task collection module includes: system allocation and free collection; the task verification module verifies the completion of the evaluation task including: verifying whether the evaluation result is consistent with the number and ID of the evaluation task, verifying whether the evaluation data conforms to the format and is stored intact, and verifying whether the evaluator has the evaluation authority.

4. The aesthetic image evaluation system according to claim 1, characterized in that: The credit rating module can dynamically allocate evaluation weights according to the credit rating of the evaluator; the credit warning module can refuse to evaluate an evaluator who has failed multiple times in a short period of time or whose credit rating is too low. Methods for restoring the normal evaluation status include evaluation examinations and application approval.

5. The aesthetic image evaluation system according to claim 1, characterized in that: The data processing module generates data based on the task management module, including assessment tasks and assessment results of assessors; the solidification method of the data storage module includes database storage and file storage; the data transmission module includes assessment data when data is used synchronously between different terminals, transmits the assessment data to the same storage terminal, and retrieves the assessment data for training.

6. A detection method for an aesthetic image evaluation system as claimed in any one of claims 1 to 5, comprising a random repetition detection algorithm, an anchor library sampling detection algorithm and an inter-person detection algorithm; The specific steps of the random repetition detection algorithm are: Step A1: The evaluator completes the evaluation of N images; Step A2: randomly select the images evaluated by the evaluators during this evaluation process for re-evaluation; Step A3: Compare the two evaluation results to see if they are consistent; Step A4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point; The specific steps of the anchor library sampling detection algorithm are as follows: Step B1: The evaluator completes the evaluation of N images; Step B2: Extract the two images with the best and worst evaluation results from the anchor point library to re-evaluate the evaluators; Step B3: Determine whether the re-evaluated result is consistent with the evaluation results of the two images extracted from the anchor point library; Step B4: If the evaluation results are consistent, the test is passed, and the batch credit is +1 point, and the total credit is +1 point; if the evaluation results are inconsistent, the test is not passed, and the batch credit is -1 point, and the total credit is -1 point; The specific steps of the inter-person detection algorithm are: Step C1: The evaluator completes the evaluation of N images; Step C2: randomly select an image that has been evaluated by M evaluators for re-evaluation; Step C3: record the difference between the evaluator and the M evaluators who have been evaluated as β. If β exceeds the preset difference set by the back-end staff, mark the evaluation result as abnormal and jump to step C4; if β does not exceed the preset value set by the back-end staff, the test is passed, and the batch credit is increased by 1 point and the total credit is increased by 1 point. Step C4: the evaluator continues to evaluate K images, and the images extracted in step 2 are re-evaluated by the evaluator, and the process goes to step C5; Step C5 : Determine whether the results of the two re-evaluations in step C2 and step C4 are consistent. If they are consistent, the test is passed. If they are inconsistent, record the difference m between the result of the re-evaluation in step C4 and the results of the M evaluators who have been evaluated in step C2, and compare m with the evaluation abnormality value γ set in the background. If m is greater than γ, reduce the batch credit by 1 point and the total credit by 1 point; if m is less than γ, reduce the batch credit by 0.5 points and the total credit by 0.5 points.

7. The detection method of an aesthetic image evaluation system according to claim 6, characterized in that: In the random repetition detection algorithm, anchor library sampling detection algorithm and inter-personnel detection algorithm, the re-evaluation of the evaluators is performed by the background, and the evaluators are not aware of the re-evaluation. The system interface during the re-evaluation is consistent with the normal evaluation.

8. The detection method of an aesthetic image evaluation system according to claim 6, characterized in that: When the batch credit and total credit are lower than the warning line set by the background, the background will force the user to log off or issue a warning.

9. The detection method of an aesthetic image evaluation system according to claim 6, characterized in that: The evaluation scores detected by any of the random repetition detection algorithm, anchor library sampling detection algorithm, and inter-person detection algorithm will be entered into the scoring database.

10. The detection method of an aesthetic image evaluation system according to claim 6, characterized in that: The numbers N, K, and γ are all adjusted and set by the backend staff.

Images