Evaluation method and system for image intelligent processing algorithm

Abstract

The invention discloses an evaluation method and system for an image intelligent processing algorithm. The method comprises the following steps: establishing an image processing intelligent algorithm evaluation index system; determining a subjective index weight by using an analytic hierarchy process; determining an objective index weight by using an improved entropy weight method; writing subjective index weights and corresponding objective index weights into a linear combination form by using a dynamic weighting method, and dynamically adjusting weight coefficients of the subjective indexes and the corresponding objective indexes based on evaluation requirements and targets to finally determine comprehensive weights of the indexes; and carrying out actual score evaluation on the second-level indexes, determining actual scores of the second-level indexes and actual scores of the first-level indexes in combination with the comprehensive weights of the second-level indexes, finally obtaining comprehensive scores of the evaluation objects, and giving final grade comprehensive evaluation. According to the method, a complete intelligent algorithm evaluation index system is established, so that comprehensive and systematic evaluation of an image intelligent processing algorithm is realized.

Description

Technical Field

[0001] The present invention relates to an evaluation technology for image intelligent processing algorithms in the field of geographic information, and in particular to an evaluation method and system for image intelligent processing algorithms. Background Art

[0002] The rapid development and widespread application of artificial intelligence are changing human life. Artificial intelligence systems are playing an increasingly important role in the fields of image recognition, speech recognition, and natural language processing. For a comprehensive evaluation of such intelligent systems, it is necessary not only to evaluate whether they meet the set requirements from the perspectives of function and performance, but also to have an objective evaluation result in terms of reliability and resource utilization, so as to judge the comprehensive level of intelligent algorithms. However, since there is no unified standard for the understanding of intelligent algorithms in the industry, general intelligent algorithm evaluation indicators such as accuracy, precision, and recall are difficult to fully apply to specified application fields. The existing evaluation methods only evaluate from one aspect of algorithm performance, lacking a comprehensive and systematic evaluation of intelligent algorithm robustness, resource usage, etc., the evaluation elements are incomplete, the evaluation indicators are relatively single, and the evaluation system is not perfect; the evaluation process is usually carried out during the software testing process, and there is no fixed specification for the evaluation process, and the evaluation process is imperfect; it is difficult to have an accurate or standardized definition and calculation method for the evaluation indicators, and there is no accurate calculation result for the indicators at each stage, resulting in unclear evaluation content, and there is no specific calculation method for the weight of each indicator. Most of them use subjective judgment to determine the weight of different indicators; different evaluation indicators need to be selected in different scenarios. Traditional intelligent algorithm evaluation often uses the same evaluation indicators in different scenarios or simply deletes them, lacks scenario specificity, and may have redundancy of evaluation indicators and simple calculation problems of weights. Therefore, there is still a lack of evaluation methods for intelligent systems, and the ability of intelligent algorithms lacks objective evaluation results.

[0003] The evaluation of intelligent algorithms should be comprehensively evaluated using multiple indicators. The China Artificial Intelligence Open Source Software Development Alliance and 15 industry-university-research institutions jointly compiled the first Chinese artificial intelligence deep learning algorithm group standard "Artificial Intelligence Deep Learning Algorithm Evaluation Specification" (T / CESA 1026-2018), which fills the gap in the domestic deep learning algorithm evaluation standard. The reliability evaluation of deep learning algorithms is oriented to the algorithm demand stage, design stage, implementation stage, and operation stage. It proposes 7 first-level indicators and 20 second-level indicators for evaluation, and selects different indicators according to the algorithm reliability target. However, the standard does not have a clear calculation method for the proportion of each indicator, and lacks a comprehensive grade evaluation for the evaluation of multiple combined indicators; the standard is targeted at deep learning algorithms and has a wide range, and lacks targeted evaluation indicators for specific fields such as image processing algorithms. Regarding the determination of the weight of evaluation indicators, the current mainstream methods include subjective weighting method and objective weighting method. The subjective weighting method is mainly based on the subjective judgment of experts. The main methods include hierarchical analysis method and fuzzy evaluation method. For example, Li Jianbin and others from Wuhan University have constructed a multi-level fuzzy comprehensive evaluation model based on existing theories and national standards for complex and diverse evaluation indicators to evaluate the safety performance of hydraulic steel gates. Objective weighting methods include principal component analysis method, entropy weight method and other methods based on mathematical theory. The subjective weighting method does not require sample data. Experts make judgments based on their understanding and experience of the connotation and extension of the evaluation indicators. It is simple to use and has a wide range of applications. It is suitable for some qualitative fuzzy indicators. However, the judgment matrix of the subjective weighting method is obtained by experts based on their subjective judgment based on experience, which brings strong subjectivity to the evaluation results. Different experts have deviations in their understanding of the same indicator and do not have high credibility. The judgment results of the objective weighting method do not rely on the subjective judgment of experts, have a strong mathematical theoretical basis, and have high credibility. However, when applied, more sample data is required, the calculation process is more complicated, and it cannot reflect the importance of the judges to different attribute indicators. The universality and participation are poor. Therefore, the combined weighting method that combines the advantages of subjective and objective weighting methods has received more and more attention. The combined weighting method takes into account the evaluator's preference for indicator attributes, reduces the arbitrariness of subjective weighting, and the evaluation results have higher authenticity and credibility. The present invention introduces dynamic factors into the combined weighting method, which can flexibly adjust the proportion of objective weighting and subjective weighting, so as to better apply to the evaluation of intelligent processing algorithms in different fields. Summary of the invention

[0004] The purpose of this invention is to propose an evaluation framework and a dynamic weighting method for indicators for image intelligent processing algorithms. On the one hand, a systematic image processing intelligent algorithm evaluation indicator system and evaluation process are established to solve the problems of lack of intelligent algorithm evaluation system, imperfect evaluation process and insufficient evaluation in the field of image processing. On the other hand, according to different scenarios and different intelligent algorithm evaluation requirements, a dynamic weighting method is used to assign weights to indicators of intelligent algorithms in the field of image processing, and then the evaluation results are comprehensively scored to give a comprehensive evaluation conclusion.

[0005] The technical solution to achieve the purpose of the present invention is: an evaluation method for image intelligent processing algorithm, comprising the following steps:

[0006] Step 1: Establish an evaluation index system for image processing intelligent algorithms, including primary and secondary indicators. The primary indicators are determined according to the evaluation objectives of the image processing intelligent algorithms, and the secondary indicators are divided into the primary indicators using multiple measurable indicators.

[0007] Step 2, using the analytic hierarchy process to determine the subjective indicator weights, including the subjective indicator weights of the first-level indicators and the subjective indicator weights of the second-level indicators;

[0008] Step 3, using the improved entropy weight method to determine the objective indicator weights, including the objective indicator weights of the primary indicators and the objective indicator weights of the secondary indicators, the improved entropy weight method adds a correction coefficient and an expert confidence index, and weights the objective indicator weights with the correction coefficient and the objective indicator weights with the expert confidence index to obtain the final objective indicator weights;

[0009] Step 4: Use the dynamic weighting method to write the subjective indicator weights and the corresponding objective indicator weights into a linear combination form, and dynamically adjust the weight coefficients of the subjective indicators and the corresponding objective indicators based on the evaluation needs and goals to finally determine the comprehensive weights of the indicators, including the comprehensive weights of the first-level indicators and the comprehensive weights of the second-level indicators;

[0010] Step 5: evaluate the actual scores of the secondary indicators, combine the comprehensive weights of the secondary indicators, determine the actual scores of the secondary indicators and the actual scores of the primary indicators, and finally obtain the comprehensive scores of the evaluation objects;

[0011] Step 6: Carry out comprehensive evaluation grading. According to whether the indicator system meets the algorithm and the evaluation target requirements, the comprehensive evaluation of the indicators is divided into four levels: excellent, good, poor, and very poor. According to the comprehensive score results, the final comprehensive evaluation is given.

[0012] Furthermore, in step 1, an evaluation index system for image processing intelligent algorithms is established, including primary and secondary indicators, where:

[0013] The first-level indicators include: algorithm correctness evaluation, code correctness evaluation, algorithm model performance evaluation, algorithm resource utilization evaluation, algorithm robustness evaluation, data set evaluation, image quality evaluation and tracking capability evaluation;

[0014] Algorithm correctness assessment is used to evaluate whether the functions implemented by the image processing intelligent algorithm meet the requirements. Its secondary indicators include correctness, response time, time complexity and space complexity;

[0015] Code correctness assessment is used to evaluate the correctness of the algorithm code to implement the function. Its secondary indicators include code standardization, code readability, formula correctness and code vulnerabilities.

[0016] Algorithm model performance evaluation is used to evaluate whether the algorithm meets the performance indicators required by the evaluation requirements. Its secondary indicators include precision, accuracy, recall, F1 Score, mAP, IoU, mean square error, mean absolute error, root mean square error, determination coefficient, Rand index, mutual information and silhouette coefficient;

[0017] Algorithm resource utilization evaluation is used to evaluate the system resources consumed during algorithm execution. Its secondary indicators include CPU utilization, memory utilization, I / O utilization, and GPU utilization.

[0018] Algorithm robustness evaluation is used to evaluate the performance of the algorithm when it encounters adversarial samples, dependent platforms, and changes in environmental data to ensure the stability and reliability of the algorithm. Its secondary indicators include adversarial sample attack capability evaluation, adversarial sample noise amplitude evaluation, learning framework evaluation, operating system evaluation, hardware architecture evaluation, interference data evaluation, data set distribution migration evaluation, and outlier data evaluation.

[0019] Dataset evaluation is used to evaluate the quality of intelligent system datasets. Its secondary indicators include dataset size, dataset balance, dataset diversity, and dataset annotation quality.

[0020] Image quality assessment is used to evaluate the quality of images from image processing intelligent algorithms. Its secondary indicators include peak signal-to-noise ratio, structural similarity, pixel accuracy, and Dice coefficient.

[0021] Tracking capability evaluation is used to evaluate the target tracking capability of image processing intelligent algorithms. Its secondary indicators include multi-target tracking accuracy, number of tracking interruptions, ID Scores, and Precisionplot.

[0022] Furthermore, in step 2, the weights of subjective indicators are determined using the analytic hierarchy process. The specific method is as follows:

[0023] 2.1) Score the relative importance of each pair of first-level indicators to construct a first-level indicator judgment matrix; score the relative importance of each pair of second-level indicators under the first-level indicators to construct a second-level indicator judgment matrix;

[0024] 2.2) The judgment matrix is ​​expressed as:

[0025]

[0026] where a ij It indicates the relative importance of the i-th indicator relative to the j-th indicator, and n indicates the number of indicators;

[0027] Using the square root method, calculate the nth root of the product of each row in the judgment matrix, that is:

[0028]

[0029] 2.3) The vector Standardization, get the subjective index weight W i :

[0030]

[0031] Furthermore, in step 3, the objective indicator weights are determined using the improved entropy weight method. The specific method is as follows:

[0032] 4.1) m experts will give refined scores to the primary indicators and secondary indicators respectively. The score of each indicator is between 0 and 100. The higher the importance of the indicator, the higher the score. Based on this, an expert scoring table for the primary indicators and the secondary indicators under the primary indicators is constructed;

[0033] 4.2) Establish the expert judgment matrix X ij , expressed as:

[0034]

[0035] Where n represents the number of indicators, m represents the number of experts, i = 1, ..., n, j = 1, ..., m, T nm It represents the score of the mth expert on the nth indicator;

[0036] 4.3) Standardize the expert judgment matrix, the formula is as follows:

[0037]

[0038] Among them, max(X ij ) is X ij The maximum value in min(X ij ) is X ij The minimum value in ;

[0039] 4.4) Calculate the information entropy value E i , the formula is as follows:

[0040]

[0041] in,

[0042] 4.5) Calculate the objective indicator weight F with the correction factor i , the formula is as follows:

[0043]

[0044] in, represents the correction factor;

[0045] 4.6) Calculate the confidence index score of each expert using the following formula:

[0046] k i =a i +b i +c i (8)

[0047] Among them, a i 、b i 、c i They represent the index assessment confidence score, professional field familiarity score, and years of work experience score of the i-th expert respectively;

[0048] 4.7) Construct the expert confidence index score matrix as:

[0049] K=[k1…k m ](9)

[0050] 4.8) The expert confidence index score matrix is ​​normalized, and the formula is as follows:

[0051]

[0052] 4.9) Expert judgment matrix X ij Normalization is performed, and the formula is as follows:

[0053]

[0054] 4.10) Introducing the objective indicator weight F of the expert confidence index i '

[0055]

[0056] 4.11) F i With F i ′ performs coupling calculation to obtain the final objective indicator weight Wi ′, the formula is as follows:

[0057]

[0058] Furthermore, in step 4, the dynamic weighting method is used to write the subjective indicator weight and the corresponding objective indicator weight into a linear combination form, and the weight coefficients of the subjective indicator and the corresponding objective indicator are dynamically adjusted based on the evaluation needs and goals, and the comprehensive weight is finally determined. The specific method is:

[0059] The subjective indicator weight and the corresponding objective indicator weight are written in the form of a linear combination, and the formula is as follows:

[0060] W=λ1W i +λ2W i '(14)

[0061] The objective function is set as follows:

[0062]

[0063] Among them, W represents the comprehensive weight, λ1 and λ2 are W i and W i The weight of ′ is called the adjustment coefficient, which is dynamically adjusted according to the evaluation needs and goals to meet the following relationship:

[0064] λ1+λ2=1(0≤λ1≤1,0≤λ2≤1)(16).

[0065] Furthermore, in step 5, the actual score of the secondary indicators is evaluated, and the actual score of the secondary indicators and the actual score of the primary indicators are determined by combining the comprehensive weights of the secondary indicators, and finally the comprehensive score of the evaluation object is obtained. The specific method is:

[0066] 5.1) Assume that the total score of all first-level indicators is 100 points. According to the comprehensive weight of the first-level indicators, calculate the theoretical score S of the i-th first-level indicator. 1i ′;

[0067] S 1i ′=100×W 1i (17)

[0068] Among them, W 1i It represents the comprehensive weight of the i-th first-level indicator;

[0069] 5.2) According to the comprehensive weight of the secondary indicators, calculate the theoretical score S2′ of the secondary indicator corresponding to each primary indicator j ;

[0070] S′ 2j =S 1i ′×W 2j(18)

[0071] Among them, W 2j represents the comprehensive weight of the j-th secondary indicator;

[0072] 5.3) Experts use a three-level scaling method of 0, 0.5, and 1 to evaluate the theoretical scores of the secondary indicators. If the indicators are fully in line with the expected target, the score is 1; if they are partially in line with the expected target, the score is 0.5; if they are completely unsatisfactory, the score is 0. Finally, the actual score S corresponding to each secondary indicator is obtained. 2j =S' 2j * Secondary indicator score evaluation;

[0073] 5.4) Add up the actual scores of all secondary indicators to obtain the actual score S of each primary indicator 1j ;

[0074]

[0075] 5.5) Add up the actual scores of all first-level indicators to obtain the comprehensive score S of the evaluation object;

[0076]

[0077] Further, in step 6, the comprehensive evaluation level is divided into four levels: excellent, good, poor, and very poor, according to whether the indicator system meets the algorithm and the evaluation target requirements. According to the comprehensive score results, the final comprehensive evaluation level is given, where:

[0078] Excellent: 85<S≤100, the algorithm has excellent performance, excellent image processing ability and robustness;

[0079] Good: 70<S≤85, the algorithm has good performance, good image processing ability and robustness;

[0080] Poor: 55<S≤70, the algorithm performance, image processing ability and robustness are poor;

[0081] Very poor: S≤55. The algorithm performance, image processing capability and robustness are very poor. Optimization is recommended.

[0082] An evaluation system for an image intelligent processing algorithm implements the evaluation method for an image intelligent processing algorithm to achieve evaluation of the image intelligent processing algorithm.

[0083] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the evaluation method for image intelligent processing algorithms is implemented to achieve evaluation of image intelligent processing algorithms.

[0084] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the evaluation method for an image intelligent processing algorithm is implemented to achieve the evaluation of the image intelligent processing algorithm.

[0085] Compared with the prior art, the present invention has the following significant advantages: (1) By establishing a complete intelligent algorithm evaluation index system, a comprehensive and systematic evaluation of the image intelligent processing algorithm is achieved. (2) The intelligent algorithm evaluation process specification is established to clarify the overall algorithm evaluation process, which can provide clearer guidance for the evaluation process. (3) The weights of the intelligent system evaluation indexes are determined by a dynamic weighting method combining the hierarchical analysis method and the improved entropy weight method, which can dynamically weight the indexes according to the test objectives and requirements, thereby achieving a more accurate evaluation of the intelligent system. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is the image processing intelligent algorithm evaluation index system of the present invention;

[0087] Figure 2 is an evaluation flow chart of the present invention;

[0088] Figure 3 It is a schematic diagram of the indicator dynamic weighting process of the present invention. DETAILED DESCRIPTION

[0089] 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.

[0090] The present invention proposes an evaluation framework and an index dynamic weighting method for image intelligent processing algorithms, comprising the following steps:

[0091] Step 1: Establish an image processing intelligent algorithm evaluation index system, which includes 8 primary indicators and 45 secondary indicators. Figure 1 shown.

[0092] The first-level indicators include: algorithm correctness evaluation, code correctness evaluation, algorithm model performance evaluation, algorithm resource utilization evaluation, algorithm robustness evaluation, data set evaluation, image quality evaluation and tracking capability evaluation;

[0093] Algorithm correctness assessment is used to evaluate whether the functions implemented by the image processing intelligent algorithm meet the requirements. Its secondary indicators include correctness, response time, time complexity and space complexity;

[0094] Code correctness assessment is used to evaluate the correctness of the algorithm code to implement the function. Its secondary indicators include code standardization, code readability, formula correctness and code vulnerabilities.

[0095] Algorithm model performance evaluation is used to evaluate whether the algorithm meets the performance indicators required by the evaluation requirements. Its secondary indicators include precision, accuracy, recall, F1 Score, mAP, IoU, mean square error, mean absolute error, root mean square error, determination coefficient, Rand index, mutual information and silhouette coefficient;

[0096] Algorithm resource utilization evaluation is used to evaluate the system resources consumed during algorithm execution. Its secondary indicators include CPU utilization, memory utilization, I / O utilization, and GPU utilization.

[0097] Algorithm robustness evaluation is used to evaluate the performance of the algorithm when it encounters adversarial samples, dependent platforms, and changes in environmental data to ensure the stability and reliability of the algorithm. Its secondary indicators include adversarial sample attack capability evaluation, adversarial sample noise amplitude evaluation, learning framework evaluation, operating system evaluation, hardware architecture evaluation, interference data evaluation, data set distribution migration evaluation, and outlier data evaluation.

[0098] Dataset evaluation is used to evaluate the quality of intelligent system datasets. Its secondary indicators include dataset size, dataset balance, dataset diversity, and dataset annotation quality.

[0099] Image quality assessment is used to evaluate the quality of images from image processing intelligent algorithms. Its secondary indicators include peak signal-to-noise ratio, structural similarity, pixel accuracy, and Dice coefficient.

[0100] Tracking capability evaluation is used to evaluate the target tracking capability of image processing intelligent algorithms. Its secondary indicators include multi-target tracking accuracy, number of tracking interruptions, ID Scores, and Precisionplot.

[0101] Step 2: Establish an image processing intelligent algorithm evaluation process and perform algorithm testing. First, determine the evaluation target based on the evaluation requirements, then build an intelligent system evaluation indicator system, select the evaluation indicators corresponding to the evaluation target, then prepare the evaluation data set, and perform algorithm testing. Obtain the execution results of each selected indicator. The evaluation process is as follows: Figure 2 shown.

[0102] Step 3: Use the analytic hierarchy process to determine the weights of subjective indicators. The processing methods for primary indicators and secondary indicators are the same. The specific steps are as follows:

[0103] (1) Experts score the relative importance of the first-level indicators and the second-level indicators, respectively, to form a first-level indicator judgment matrix and a second-level indicator judgment matrix. The scoring basis for constructing the judgment matrix is ​​shown in Table 1.

[0104] Table 1 The basis for constructing the judgment matrix by the analytic hierarchy process

[0105] Scale meaning 1 Both elements have equal importance 3 The former is slightly more important than the latter. 5 The former of the two elements is obviously more important than the latter 7 The former is more important than the latter. 9 The former of the two elements is more important than the latter 2,4，6，8 The middle value of the above judgment Count down from 1 to 9 Indicates the importance of swapping order comparisons

[0106] The judgment matrix construction method is as follows:

[0107] Let the index be P1,…,P n , a total of n indicators, and then list the judgment matrix construction table with n rows and n columns. Finally, according to the rules of constructing the judgment matrix in Table 1, the importance of the nth element of each row and the nth element of each column are compared in turn, and finally a judgment matrix is ​​formed. The judgment matrix is ​​usually in the following form:

[0108]

[0109]

[0110] (2) Using the square root method, calculate the nth root of the product of each row in the judgment matrix, that is:

[0111]

[0112] in, represents the nth root of the product of all the judgment scoring elements in the judgment matrix of the i-th indicator, j represents the j-th column, n represents a total of n indicators, a ij Represents the element in the i-th row and j-th column of the judgment matrix.

[0113] (3) The vector After standardization, the weight W of the hierarchical analysis method is obtained i , that is, the subjective indicator weight.

[0114]

[0115] Step 4: Use the improved entropy weight method to determine the objective indicator weights. The processing methods for primary indicators and secondary indicators are the same. The specific steps are as follows:

[0116] (1) m experts conduct a detailed evaluation of the first-level indicators and the second-level indicators, where the score of each indicator ranges from 0 to 100 points. The higher the importance of the indicator, the higher the score, thereby obtaining the expert scoring table of the first-level indicators and the second-level indicators.

[0117] (2) Based on the expert scores, establish the indicator expert judgment matrix X ij .

[0118]

[0119] Where n represents the number of indicators, m represents the number of experts, i = 1, ..., n, j = 1, ..., m, T nm It represents the score of the mth expert on the nth indicator.

[0120] (3) Standardize the expert judgment matrix to obtain Y ij , the formula is as follows:

[0121]

[0122] Among them, max(X ij ) is X ij The maximum value in min(X ij ) is X ij The minimum value in .

[0123] (4) Calculate the information entropy value E of the indicator i , the formula is as follows:

[0124]

[0125] in, is the variability index, and n represents the number of indicators.

[0126] Therefore, the objective indicator weight F corresponding to the original entropy weight method i The calculation formula is as follows:

[0127]

[0128] Among them, F i ″ represents the original objective weight of the ith indicator, and n represents the number of indicators.

[0129] (5) The original entropy weight method has an inherent shortcoming, that is, when the information entropy E i When it is close to 1, the molecule of the above formula will become very small, and E i The slight difference between them will cause the corresponding entropy weight F i ″The change of multiples leads to a large error in the indicator weight. Therefore, a correction coefficient is used To correct the weight coefficient of the original entropy weight method, the correction coefficient takes the information entropy value E i The average of the differences from 1. Since a large positive number is added to both the numerator and the denominator, the inherent deficiencies in the original entropy weight method can be corrected.

[0130] Therefore, the objective indicator weight F with the correction coefficient is introduced i The calculation formula is as follows:

[0131]

[0132] in, represents the correction coefficient, and n represents the number of indicators.

[0133] (6) Use the expert confidence index to reduce the subjective preference of experts and improve the accuracy of objective evaluation. The expert confidence index table is shown in Table 2.

[0134] Table 2 Expert confidence index

[0135]

[0136]

[0137] The confidence index scores of each expert are as follows:

[0138] k i =a i +b i +c i (8)

[0139] Among them, a i 、b i 、c i They represent the confidence in indicator assessment, familiarity with the professional field, and years of work experience of the i-th expert respectively.

[0140] Then the expert confidence index score matrix is:

[0141] K=[k1…k m ](9)

[0142] Where m represents the number of experts.

[0143] Normalization is performed, and the formula is as follows:

[0144]

[0145] Among them, m represents the number of experts, k i Represents the expert confidence index of the i-th expert.

[0146] For the expert judgment matrix X ij Normalization is performed, and the formula is as follows:

[0147]

[0148] Among them, n represents the number of indicators, X ij represents the expert judgment matrix.

[0149] The objective indicator weight F obtained by combining each indicator with the expert confidence index i '

[0150]

[0151] Among them, n represents the number of indicators.

[0152] (7) The objective indicator weight F obtained by the improved entropy weight method i The objective indicator weight F obtained by adding the expert confidence index i ′ performs coupling calculation to obtain the final objective indicator weight W of the improved entropy weight method i ′, the formula is as follows:

[0153]

[0154] Among them, n represents the number of indicators.

[0155] Step 5: Use the dynamic weighting method to process the primary and secondary indicators as follows: Write the subjective indicator weights obtained by the hierarchical analysis method and the objective indicator weights obtained by the final improved entropy weight method into a linear combination form, and use the comprehensive weight W to compare with W i and W i ′ is minimized as the objective function, and the optimal combination of λ1 and λ2 is found to finally determine the weight of the comprehensive index. The formula is as follows:

[0156] W=λ1W i +λ2W i '(14)

[0157] The objective function is set as follows:

[0158]

[0159] Among them, λ1 and λ2 are W i and W i The weight of ′ is called the adjustment coefficient, which can be dynamically adjusted according to the evaluation needs and goals to meet the following relationship:

[0160] λ1+λ2=1(0≤λ1≤1,0≤λ2≤1)(16)

[0161] Taking into account the subjective and objective factors of the image intelligent processing algorithm, the present invention adopts a weight coefficient λ1=λ2=0.5, that is, the subjective and objective factors each account for 50% of the weight.

[0162] At this point, the comprehensive weights of all primary and secondary indicators can be obtained.

[0163] Step 6: Perform algorithm testing to obtain algorithm testing results. Experts will score the test results and combine the comprehensive weights of the secondary indicators obtained in step 5 to obtain the actual scores of the secondary indicators and the primary indicators, and finally obtain the comprehensive score of the evaluation object. The specific steps are as follows:

[0164] (1) Assuming the total score of all first-level indicators is 100 points, calculate the theoretical score S of the i-th first-level indicator according to the comprehensive weight of the first-level indicators. 1i ′.

[0165] S 1i ′=100×W 1i (17)

[0166] Among them, W 1i Represents the comprehensive weight of the i-th first-level indicator.

[0167] (2) According to the weight of the secondary indicators, calculate the theoretical score S2′ of the secondary indicator corresponding to each primary indicator j .

[0168] S′ 2j =S 1i ′×W 2j (18)

[0169] Among them, W 2j Represents the comprehensive weight of the j-th secondary indicator.

[0170] (3) Based on the algorithm execution results, experts use a three-level scaling method of 0, 0.5, and 1 to evaluate the actual scores of the secondary indicator algorithm execution results. If the algorithm execution results are completely consistent with the expected goals, the score is 1; if they are partially consistent, the score is 0.5; if they are completely inconsistent, the score is 0. Finally, the actual score S corresponding to each secondary indicator is obtained. 2j .

[0171] (4) Add up the actual scores of all secondary indicators to obtain the actual score S of each primary indicator 1j .

[0172]

[0173] (5) Add up the actual scores of all first-level indicators to obtain the comprehensive score S of the evaluation object.

[0174]

[0175] Step 7: Classify the comprehensive evaluation level. According to whether the indicator system meets the algorithm and the evaluation target requirements, the comprehensive evaluation of the indicators is divided into four levels: excellent, good, poor, and very poor. According to the comprehensive evaluation results, the final comprehensive evaluation level is given.

[0176] Table 3 Comprehensive evaluation level classification

[0177]

[0178] The present invention proposes an evaluation system for an image intelligent processing algorithm, implements the evaluation method for an image intelligent processing algorithm, and realizes the evaluation of the image intelligent processing algorithm.

[0179] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the evaluation method for image intelligent processing algorithms is implemented to achieve evaluation of image intelligent processing algorithms.

[0180] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the evaluation method for an image intelligent processing algorithm is implemented to achieve the evaluation of the image intelligent processing algorithm.

[0181] Example

[0182] In order to verify the effectiveness of the scheme of the present invention, the following experiment was carried out.

[0183] The present invention provides an evaluation framework for image intelligent processing algorithms and a weight determination method for an indicator dynamic weighting method, so as to realize a comprehensive evaluation of image processing intelligent algorithms. For the convenience of description, we assume the following simplified application example.

[0184] According to the steps mentioned above, implement them in sequence:

[0185] Step 1) Establish an evaluation index system for intelligent image processing algorithms, and select the typical image processing target detection algorithm YOLOv5 for evaluation. YOLOv5 (You Only Look Once version 5) is a popular target detection algorithm that directly predicts the target's bounding box and category probability through a convolutional neural network. The target detection algorithm is related to algorithm correctness, algorithm performance, code correctness, resource utilization, and data set quality, but pays less attention to algorithm robustness, image quality, and tracking capabilities; the target detection algorithm belongs to a classification model, so the indicators selected for the evaluation of the algorithm include: 5 first-level indicators and 22 second-level indicators, as shown in the following table.

[0186] Table 4. Primary and secondary indicators selected by YOLOv5

[0187]

[0188] Step 2) Establish an image processing intelligent algorithm evaluation process and perform algorithm testing. First, determine the evaluation target based on the evaluation requirements, then select the evaluation indicators corresponding to the evaluation target based on the evaluation system established in step 1, then prepare the evaluation data set and perform algorithm testing. Obtain the execution results of each selected indicator. The evaluation process is as follows: Figure 2 shown.

[0189] Step 3) Use the analytic hierarchy process to determine the subjective weights of the five first-level indicators and 22 second-level indicators. The specific process is as follows:

[0190] (1) According to the comparison of the importance of each indicator, a first-level indicator judgment matrix is ​​constructed, as shown in Table 5.

[0191] Table 5 First-level indicator judgment matrix

[0192] Functional Correctness Performance Indicators Dataset Resource Utilization Code Correctness Functional Correctness 1 3 5 7 3 Performance Indicators 1 / 3 1 3 5 2 Dataset quality 1 / 5 1 / 3 1 3 1 / 5 Resource Utilization 1 / 7 1 / 5 1 / 3 1 1 / 5 Code Correctness 1 / 3 1 / 2 5 5 1

[0193] (2) Secondary indicator judgment matrix

[0194] Table 6 Functional correctness secondary index judgment matrix

[0195] Correctness Response time Time complexity Space complexity Correctness 1 3 5 5 Response time 1 / 3 1 2 3 Time complexity 1 / 5 1 / 2 1 2 Space complexity 1 / 5 1 / 3 1 / 2 1

[0196] Table 7 Performance index secondary index judgment matrix

[0197] Accuracy Accuracy Recall F1score mAP IoU Accuracy 1 1 1 3 1 2 Accuracy 1 1 1 2 1 2 Recall 1 1 1 2 1 3 F1score 1 / 3 1 / 2 1 / 2 1 1 / 3 1 / 2 mAP 1 1 1 3 1 2 IoU 1 / 2 1 / 2 1 / 2 2 1 / 2 1

[0198] Table 8 Dataset quality secondary indicator judgment matrix

[0199]

[0200] Table 9 Resource utilization secondary index judgment matrix

[0201] CPU usage Memory usage I / O Utilization GPU Utilization CPU usage 1 1 / 4 3 1 / 3 Memory usage 4 1 5 2 I / O Utilization 1 / 3 1 / 5 1 1 / 3 GPU Utilization 3 1 / 2 3 1

[0202] Table 10 Code correctness secondary index judgment matrix

[0203] Code Vulnerabilities Formula Correctness Code readability Code Standardization Code Vulnerabilities 1 1 / 3 3 3 Formula Correctness 3 1 3 5 Code readability 1 / 3 1 / 3 1 3 Code Standardization 1 / 3 1 / 5 1 / 3 1

[0204] (3) Determine the subjective weights of primary and secondary indicators

[0205] The weights are calculated using the square root method according to the judgment matrix. The calculation method is shown in formula (1). The nth square root of the product of each row in the judgment matrix of each indicator is calculated, and the standardized calculation is performed according to formula (2). Finally, the subjective indicator weight is obtained. The same calculation method is used for the first-level indicator and the second-level indicator. The results of the subjective indicator weight are shown in Table 11.

[0206] Table 11 Weights of primary and secondary indicators

[0207]

[0208]

[0209] Step 4) Use the improved entropy weight method to determine the objective indicator weights of the 5 first-level indicators and 22 second-level indicators. Among them, the expert assignment results invite 5 experts in this field to assign scores from 0 to 100 based on their expert experience and construct an expert scoring table. According to the expert scoring table and formula (3), the expert judgment matrix is ​​constructed. The expert judgment matrix is ​​the expert scoring table written in matrix form. The specific steps are as follows: (1) Expert scoring table for first-level indicators and second-level indicators

[0210] Table 12 Expert scoring table for first-level indicators

[0211] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Functional Correctness 100 90 95 90 95 Performance Indicators 95 80 90 90 85 Dataset 70 60 70 75 70 Resource Utilization 80 75 70 65 80 Code Correctness 90 85 85 80 80

[0212] Table 13 Expert scoring table for functional correctness

[0213] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Correctness 95 100 90 100 90 Response time 90 90 85 95 85 Time complexity 85 70 75 80 85 Space complexity 60 75 65 70 65

[0214] Table 14 Performance index expert scoring table

[0215]

[0216]

[0217] Table 15 Dataset quality expert rating table

[0218]

[0219] Table 16 Resource Utilization Expert Rating Table

[0220] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 CPU usage 95 90 80 90 95 Memory usage 100 95 100 100 95 I / O Utilization 50 65 60 65 70 GPU Utilization 85 100 90 100 90

[0221] Table 17 Code Correctness Expert Rating Table

[0222] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Code Vulnerabilities 100 95 95 100 100 Formula Correctness 95 100 100 95 90 Code readability 70 65 70 50 60 Code Standardization 65 70 70 75 70

[0223] (2) Judgment matrix standardization calculation

[0224] According to formula (4), the expert judgment matrix of the first-level index and the second-level index is standardized and calculated to obtain the standardized results.

[0225] Table 18 Standardized results of the first-level indicator judgment matrix

[0226] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Functional Correctness 1.0 0.0 0.5 0.0 0.5 Performance Indicators 1.0 0.0 0.67 0.67 0.33 Dataset 0.67 0.0 0.67 1.0 0.67 Resource Utilization 1.0 0.67 0.33 0.0 1.0 Code Correctness 1.0 0.5 0.5 0.0 0.0

[0227] Table 19 Standardized results of functional correctness judgment matrix

[0228] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Correctness 0.5 1.0 0.0 1.0 0.0 Response time 0.5 0.5 0.0 1.0 0.0 Time complexity 1.0 0.0 0.33 0.67 1,0 Space complexity 0.0 1.0 0.33 0.67 0.33

[0229] Table 20 Standardization results of performance index judgment matrix

[0230] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Accuracy 0.0 1.0 0.0 1.0 1.0 Accuracy 0.0 0.0 1.0 0.0 1.0 Recall 1.0 0.0 1.0 1.0 0.0 F1score 0.5 0.75 0.0 1.0 0.25 mAP 1.0 0.33 1.0 0.0 0.33 IoU 0.0 0.75 1.0 0.25 0.75

[0231] Table 21 Dataset judgment matrix standardization results

[0232]

[0233] Table 22 Standardized results of resource utilization judgment matrix

[0234] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 CPU usage 1.0 0.67 0.0 0.67 1.0 Memory usage 1.0 0.0 1.0 1.0 0.0 I / O Utilization 0.0 0.75 0.5 0.75 1.0 GPU Utilization 0.0 1.0 0.33 1.0 0.33

[0235] Table 23 Standardized results of code correctness judgment matrix

[0236] Expert 1 Expert 2 Expert 3 Expert 4 Expert 5 Code Vulnerabilities 1.0 0.0 0.0 1.0 1.0 Formula Correctness 0.5 1.0 1.0 0.5 0.0 Code readability 1.0 0.75 1.0 0.0 0.5 Code Standardization 0.0 0.5 0.5 1.0 0.5

[0237] (3) Calculation results of each weight coefficient

[0238] According to the above judgment matrix standardization results, the original objective indicator weight F corresponding to the original entropy weight method is calculated according to formulas (5) and (6): i ″, and calculate the objective indicator weight F with the correction coefficient added according to formula (7) i , see Table 24 and Table 25. According to formula (8) to formula (12), the objective indicator weight F of increasing the expert confidence index is calculated i ′, the confidence index weights of the five experts are shown in Tables 24 and 25.

[0239] According to formula (13), the objective indicator weight W is calculated i ′, see Table 24 and Table 25. The calculation methods of the primary and secondary indicators are consistent.

[0240] Table 24 Calculation results of objective weights of primary indicators

[0241]

[0242] Table 25 Calculation results of objective weights of secondary indicators

[0243]

[0244] Step 5) According to formulas (14), (15), and (16), the dynamic weighting method is used to write the subjective index weights obtained by the hierarchical analysis method and the objective index weights obtained by the improved entropy weight method into the form of linear combinations to determine the comprehensive index weights. The specific results are as follows:

[0245] Table 26 Comprehensive weights of primary indicators

[0246]

[0247] Table 27 Comprehensive weights of secondary indicators

[0248]

[0249] Step 6) Perform algorithm testing, obtain algorithm test results, invite experts to score the test results, combine the weights of various indicators, obtain the specific situation of the secondary indicator scores and the primary indicator scores, and finally obtain the comprehensive score of the indicator, refer to formula (17) to formula (20). The specific steps are as follows:

[0250] (1) Theoretical scores of indicators. According to formula (17) and formula (18), the theoretical scores of the first-level indicators and the second-level indicators are calculated.

[0251] (2) Actual scores of secondary indicators. The three-level scaling method of 0, 0.5, and 1 was used to evaluate the comprehensive scores of the secondary indicator test execution results. The specific evaluation results are shown in Table 28.

[0252] (3) According to formula (19), the scores of each secondary indicator are added together to obtain the score of the primary indicator. According to formula (20), the actual scores of the primary indicators are added together to obtain the comprehensive score, as shown in Table 28. The comprehensive score is 89.95.

[0253] Table 28 Comprehensive weights of indicators

[0254]

[0255] Step 7) Comprehensive evaluation is graded. According to whether the indicator system meets the algorithm and the evaluation target requirements, the comprehensive evaluation of indicators is divided into four levels: excellent, good, poor, and very poor. According to the comprehensive evaluation results, the final comprehensive evaluation is given.

[0256] The comprehensive score of the Yolov5 algorithm is 89.95 points. According to Table 3, the comprehensive evaluation of the intelligent system is excellent, that is, the algorithm has excellent performance, excellent data set quality, and excellent resource usage.

[0257] In summary, the present invention adopts an image processing intelligent algorithm evaluation index system and a weight determination method based on dynamic weighting, implements the image processing intelligent algorithm evaluation method based on dynamic weighting, and realizes a comprehensive evaluation of the image processing intelligent algorithm.

[0258] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0259] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An evaluation method for image intelligent processing algorithms, characterized in that: The steps include: Step 1: Establish an evaluation index system for image processing intelligent algorithms, including primary and secondary indicators. The primary indicators are determined according to the evaluation objectives of the image processing intelligent algorithms, and the secondary indicators are divided into the primary indicators using multiple measurable indicators. Step 2, using the analytic hierarchy process to determine the subjective indicator weights, including the subjective indicator weights of the first-level indicators and the subjective indicator weights of the second-level indicators; Step 3, using the improved entropy weight method to determine the objective indicator weights, including the objective indicator weights of the primary indicators and the objective indicator weights of the secondary indicators, the improved entropy weight method adds a correction coefficient and an expert confidence index, and weights the objective indicator weights with the correction coefficient and the objective indicator weights with the expert confidence index to obtain the final objective indicator weights; Step 4: Use the dynamic weighting method to write the subjective indicator weights and the corresponding objective indicator weights into a linear combination form, and dynamically adjust the weight coefficients of the subjective indicators and the corresponding objective indicators based on the evaluation needs and goals to finally determine the comprehensive weights of the indicators, including the comprehensive weights of the first-level indicators and the comprehensive weights of the second-level indicators; Step 5: evaluate the actual scores of the secondary indicators, combine the comprehensive weights of the secondary indicators, determine the actual scores of the secondary indicators and the actual scores of the primary indicators, and finally obtain the comprehensive scores of the evaluation objects; Step 6: Carry out comprehensive evaluation grading. According to whether the indicator system meets the algorithm and the evaluation target requirements, the comprehensive evaluation of the indicators is divided into four levels: excellent, good, poor, and very poor. According to the comprehensive score results, the final comprehensive evaluation is given.

2. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 1: Establish an evaluation index system for image processing intelligent algorithms, including primary and secondary indicators, where: The first-level indicators include: algorithm correctness evaluation, code correctness evaluation, algorithm model performance evaluation, algorithm resource utilization evaluation, algorithm robustness evaluation, data set evaluation, image quality evaluation and tracking capability evaluation; Algorithm correctness assessment is used to evaluate whether the functions implemented by the image processing intelligent algorithm meet the requirements. Its secondary indicators include correctness, response time, time complexity and space complexity; Code correctness assessment is used to evaluate the correctness of the algorithm code to implement the function. Its secondary indicators include code standardization, code readability, formula correctness and code vulnerabilities. Algorithm model performance evaluation is used to evaluate whether the algorithm meets the performance indicators required by the evaluation requirements. Its secondary indicators include precision, accuracy, recall, F1 Score, mAP, IoU, mean square error, mean absolute error, root mean square error, determination coefficient, Rand index, mutual information and silhouette coefficient; Algorithm resource utilization evaluation is used to evaluate the system resources consumed during algorithm execution. Its secondary indicators include CPU utilization, memory utilization, I / O utilization, and GPU utilization. Algorithm robustness evaluation is used to evaluate the performance of the algorithm when it encounters adversarial samples, dependent platforms, and changes in environmental data to ensure the stability and reliability of the algorithm. Its secondary indicators include adversarial sample attack capability evaluation, adversarial sample noise amplitude evaluation, learning framework evaluation, operating system evaluation, hardware architecture evaluation, interference data evaluation, data set distribution migration evaluation, and outlier data evaluation. Dataset evaluation is used to evaluate the quality of intelligent system datasets. Its secondary indicators include dataset size, dataset balance, dataset diversity, and dataset annotation quality. Image quality assessment is used to evaluate the quality of images from image processing intelligent algorithms. Its secondary indicators include peak signal-to-noise ratio, structural similarity, pixel accuracy, and Dice coefficient. Tracking capability evaluation is used to evaluate the target tracking capability of image processing intelligent algorithms. Its secondary indicators include multi-target tracking accuracy, number of tracking interruptions, ID Scores, and Precision plot.

3. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 2: Use the analytic hierarchy process to determine the weights of subjective indicators. The specific method is as follows: 2.1) Score the relative importance of each pair of first-level indicators to construct a first-level indicator judgment matrix; score the relative importance of each pair of second-level indicators under the first-level indicators to construct a second-level indicator judgment matrix; 2.2) The judgment matrix is ​​expressed as: where a ij It indicates the relative importance of the i-th indicator relative to the j-th indicator, and n indicates the number of indicators; Using the square root method, calculate the nth root of the product of each row in the judgment matrix, that is: 2.3) The vector Standardization, get the subjective indicator weight W i :

4. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 3: Use the improved entropy weight method to determine the objective indicator weights. The specific method is as follows: 4.1) m experts will give refined scores to the primary indicators and secondary indicators respectively. The score of each indicator is between 0 and 100. The higher the importance of the indicator, the higher the score. Based on this, an expert scoring table for the primary indicators and the secondary indicators under the primary indicators is constructed; 4.2) Establish the expert judgment matrix X ij , expressed as: Where n represents the number of indicators, m represents the number of experts, i = 1, ..., n, j = 1, ..., m, T nm It represents the score of the mth expert on the nth indicator; 4.3) Standardize the expert judgment matrix, the formula is as follows: Among them, max(X ij ) is X ij The maximum value in min(X ij ) is X ij The minimum value in ; 4.4) Calculate the information entropy value E i , the formula is as follows: in, 4.5) Calculate the objective indicator weight F with the correction factor i , the formula is as follows: in, represents the correction factor; 4.6) Calculate the confidence index score of each expert using the following formula: k i =a i +b i +c i (8) Among them, a i 、b i 、c i They represent the index assessment confidence score, professional field familiarity score, and years of work experience score of the i-th expert respectively; 4.7) Construct the expert confidence index score matrix as: K=[k1…k m ](9) 4.8) The expert confidence index score matrix is ​​normalized, and the formula is as follows: 4.9) Expert judgment matrix X ij Normalization is performed, and the formula is as follows: 4.10) Introducing the objective indicator weight F of the expert confidence index i ' 4.11) F i With F i ′ performs coupling calculation to obtain the final objective indicator weight W i ′, the formula is as follows:

5. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 4: Use the dynamic weighting method to write the subjective indicator weight and the corresponding objective indicator weight into a linear combination form, and dynamically adjust the weight coefficients of the subjective indicator and the corresponding objective indicator based on the evaluation needs and goals, and finally determine the comprehensive weight. The specific method is: The subjective indicator weight and the corresponding objective indicator weight are written in the form of a linear combination, and the formula is as follows: W=λ1W i +λ2W i ′(14)The objective function is set as follows: Among them, W represents the comprehensive weight, λ1 and λ2 are W i and W i The weight of ′ is called the adjustment coefficient, which is dynamically adjusted according to the evaluation needs and goals to meet the following relationship: λ1+λ2=1(0≤λ1≤1,0≤λ2≤1)(16).

6. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 5: evaluate the actual scores of the secondary indicators, combine the comprehensive weights of the secondary indicators, determine the actual scores of the secondary indicators and the actual scores of the primary indicators, and finally obtain the comprehensive scores of the evaluation objects. The specific method is as follows: 5.1) Assume that the total score of all first-level indicators is 100 points. According to the comprehensive weight of the first-level indicators, calculate the theoretical score S of the i-th first-level indicator. 1i ′; S 1i ′=100×W 1i (17) Among them, W 1i It represents the comprehensive weight of the i-th first-level indicator; 5.2) According to the comprehensive weight of the secondary indicators, calculate the theoretical score S′ of the secondary indicator corresponding to each primary indicator 2j ; S′ 2j =S 1i ′×W 2j (18) Among them, W 2j represents the comprehensive weight of the j-th secondary indicator; 5.3) Experts use a three-level scaling method of 0, 0.5, and 1 to evaluate the theoretical scores of the secondary indicators. If the indicators are fully in line with the expected target, the score is 1; if they are partially in line with the expected target, the score is 0.5; if they are completely unsatisfactory, the score is 0. Finally, the actual score S corresponding to each secondary indicator is obtained. 2j =S' 2j* Secondary indicator score evaluation; 5.4) Add up the actual scores of all secondary indicators to obtain the actual score S of each primary indicator 1j ; 5.5) Add up the actual scores of all first-level indicators to obtain the comprehensive score S of the evaluation object; 7. The evaluation method for image intelligent processing algorithm according to claim 1, characterized in that: Step 6: Perform comprehensive evaluation grading. According to whether the indicator system meets the algorithm and the evaluation target requirements, the comprehensive evaluation of the indicators is divided into four levels: excellent, good, poor, and very poor. According to the comprehensive score results, the final comprehensive evaluation is given, where: Excellent: 85 < S ≤ 100, the algorithm has excellent performance, excellent image processing ability and robustness; Good: 70 ≤ S ≤ 85, the algorithm has good performance, good image processing ability and robustness; Poor: 55 < S ≤ 70, the algorithm has poor performance, image processing ability and robustness; Very poor: S ≤ 55, the algorithm has very poor performance, image processing ability and robustness, and optimization is recommended.

8. An evaluation system for image intelligent processing algorithms, characterized in that: Implement the evaluation method of the image intelligent processing algorithm according to any one of claims 1-7 to realize the evaluation of the image intelligent processing algorithm.

9. A computer device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the evaluation method of the image intelligent processing algorithm according to any one of claims 1-7 is implemented to realize the evaluation of the image intelligent processing algorithm.

10. A computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the evaluation method of the image intelligent processing algorithm according to any one of claims 1-7 is implemented to realize the evaluation of the image intelligent processing algorithm.

Images