Test coverage measurement method for deep learning intelligent software

Abstract

The invention relates to a test coverage measurement method for deep learning intelligent software, and belongs to the field of computers, and the method comprises the steps: 1, building an intelligent software test coverage measurement model, 2, determining a measurement index selection strategy S, and 3, adjusting a first weight Vuj and a second weight Vtk; 4, selecting a measurement index set; step 5, obtaining an index value of the bottom layer index; step 6, obtaining a test coverage rate when the test coverage rate is Resultt; if u00%, adding the test case and returning to the step 5, otherwise, ending the test of the intelligent software. According to the method, a three-layer intelligent software test coverage index system is established from data, algorithm, model and system layers according to deep learning intelligent software quality attributes, and the effectiveness and accuracy of intelligent software test coverage measurement can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of computers, and in particular relates to a test coverage measurement method for deep learning intelligent software. Background Art

[0002] As a research hotspot and mainstream direction in the field of artificial intelligence, deep learning technology has powerful data fitting capabilities and end-to-end modeling advantages without relying on feature engineering, which promotes intelligent software (software based on artificial intelligence technology) to play an increasingly important role in situational awareness, intelligent decision-making, human-computer interaction and other fields. While intelligent software brings great convenience to humans, due to its own characteristics such as uncertainty, complexity and intelligence, it has encountered challenges in software reliability and security that traditional software has never faced. In recent years, failures of intelligent software systems have become common.

[0003] Software testing is an important means to ensure and improve the quality of intelligent software. The adequacy of intelligent software testing will directly affect the quality of intelligent software. Therefore, it is necessary to measure the test coverage of intelligent software to evaluate the adequacy of testing. The test coverage measurement of intelligent software mainly has the following problems: 1) Since there are great differences between intelligent software implemented based on deep learning technology and traditional software implemented by direct programming, the test coverage measurement methods based on source code structure and software functional requirements applicable to traditional software are mostly not applicable to it; 2) The quality attributes of intelligent software are quite different from those of traditional software, and traditional software test coverage measurement is difficult to effectively evaluate the test adequacy of intelligent software quality attributes. Therefore, there is a need for a method to effectively measure the test coverage measurement of intelligent software to ensure the adequacy of intelligent software testing. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a test coverage measurement method for deep learning intelligent software. The core content of this technology is to design test coverage measurement indicators based on the quality attributes of intelligent software, with its data, algorithms, models and systems as objects, to form a software test coverage measurement indicator system for data, algorithm and system level decomposition, and to select indicators and test coverage in combination with indicator selection strategies, which can effectively ensure the test adequacy of deep learning intelligent software.

[0005] To achieve the above object, the present invention discloses a test coverage measurement method for deep learning intelligent software, which includes:

[0006] Step 1: Establish an intelligent software test coverage measurement model;

[0007] Step 1.1: Define the intelligent software test coverage measurement indicator set TC, TC = {TU, TM, TD}, where the top-level indicator set TU = {TU1, TU2, TU3, TU4}, the second-level indicator set TM = {TM1, TM2, ..., TM 10} and the underlying indicator set TD = {TD1, TD2, ..., TD 25};

[0008] Step 1.2: Establish a mapping relationship;

[0009] Establish a primary mapping relationship f1 (u->m) between the top-level indicator TU and the second-level indicator TM;

[0010] Step 1.3: j establishes a secondary mapping relationship;

[0011] Establish a secondary mapping relationship f2 (m->d) between the second-layer indicator TM and the bottom-layer indicator TD;

[0012] Step 1.4: Construct the intelligent software test coverage measurement model AITC = {TC, f1, f2};

[0013] Step 2: Determine the metric selection strategy S;

[0014] Determine the user requirement metric s1 according to the user requirement, determine the test target metric s2 according to the test target, determine the selection standard metric s3 according to the selection standard, determine the software level metric s4 according to the software level, determine the algorithm type metric s5 according to the algorithm type, and determine the metric selection strategy S; S = {s1, s2, s3, s4, s5};

[0015] Step 3: Adjust the first weight V1 j and the second weight V2 k ;

[0016] Step 4: Select the metric set;

[0017] According to the metric selection strategy S, the specific metric set STC is determined from the intelligent software test coverage metric set:

[0018] STC={if(V2 k ≠0)TD k , k=1,...,25} (10);

[0019] Step 5: Get the indicator value of the underlying indicator;

[0020] According to the metric selected in step 4, perform intelligent software testing to obtain the indicator value of the underlying indicator.

[0021] Step 6: Get test coverage;

[0022] According to the metric set STC and the primary mapping relationship and secondary mapping relationship in the intelligent software test coverage metric model, after obtaining the indicator value of the top-level indicator, the test coverage is obtained according to formula (11):

[0023]

[0024] Among them, Result is the test coverage, TU u is the top-level indicator;

[0025] If the test coverage Result is <100%, add test cases and return to step 5. Otherwise, the intelligent software test ends. Specifically:

[0026] When the test coverage result Result < 100%, it means that the test coverage does not meet the requirements. At this time, you need to add test cases and return to step 5 to re-execute the test;

[0027] When the test coverage result Result ≥ 100%, it means that the test coverage meets the requirements, and the intelligent software test can be ended at this time.

[0028] Preferably, in step 1.1, the intelligent software test coverage measurement indicator set TC, TC = {TU, TM, TD}, wherein the top-level indicator set TU = {TU1, TU2, TU3, TU4}, the second-level indicator set TM = {TM1, TM2, ..., TM 10} and the underlying indicator set TD = {TD1, TD2, ..., TD 25}, specifically:

[0029] The top-level indicators TU1, TU2, TU3, and TU4 are data coverage, algorithm coverage, model coverage, and system coverage, respectively. The second-level indicators TM1 to TM 10 They are data integrity, data correctness, algorithm adaptability, algorithm reliability, model stability, model functionality, model adaptability, system function, system performance, and other system attributes. The underlying indicator set TD1 to TD 25 They are dataset coverage, data distribution coverage, label space coverage, feature space coverage, time coverage, data type coverage, time complexity and space complexity, code coverage, regression task evaluation indicators, classification task evaluation indicators, response surface analysis, abnormal situation coverage, feature coverage, edge case coverage, generalization ability, scenario coverage, user group coverage, system function coverage, function implementation quality, system compliance, performance indicator coverage, resource usage coverage, fault recovery coverage, system reliability, and system security.

[0030] Preferably, the primary mapping relationship f1(u->m) in step 1.2 is specifically:

[0031]

[0032] Among them, V1 j Indicates the second-level indicator TM j At the top level indicator TU i Calculate weight, V1 j The default value is

[0033] The secondary mapping relationship f2 (m->d) in step 1.3 is specifically:

[0034]

[0035] Among them, V2 k Represents the underlying indicator TD k In the second-level indicator TM j Calculate weights, V2 k The default value is

[0036] Preferably, the step 2: determining the metric selection strategy S specifically includes the following steps:

[0037] Step 2.1: Determine s1 according to user requirements;

[0038]

[0039] Among them, M1 k For the indicator TD k The measurement value in the user requirement; s1 is the user requirement measurement index;

[0040] Step 2.2: Determine s2 according to the test objectives;

[0041]

[0042] Among them, M2 k For the indicator TD k The measurement value in the test target; s2 is the test target measurement indicator;

[0043] Step 2.3: Determine s3 according to the selection criteria;

[0044]

[0045] Among them, M3 k For the indicator TD k The measurement value in the selection standard; s3 is the measurement index of the selection standard;

[0046] Step 2.4: Determine s4 according to the software level;

[0047]

[0048] Among them, M4 k For the indicator TD k The measurement value at the software level; s4 is the software level measurement indicator;

[0049] Step 2.5: Determine s5 according to the algorithm type;

[0050]

[0051] Among them, M5 k is the measurement value in the algorithm type; s5 is the algorithm type measurement indicator;

[0052] Step 2.6: Determine the metric selection strategy. Select the metric selection strategy according to steps 2.1-2.5 and determine the metric selection strategy S, where S = {s1, s2, s3, s4, s5}.

[0053] Preferably, the step 3 adjusts the first weight V1 j and the second weight V2 k , specifically:

[0054]

[0055] Preferably, the step 6 is specifically:

[0056] Step 6.1: Calculate the value of the selected second-layer indicator. According to the value of the bottom-layer indicator and the secondary mapping relationship f2(m->d), calculate the value of the selected second-layer indicator.

[0057] Step 6.2: Calculate the value of the selected top-level indicator. According to the value of the second-level indicator and the first-order mapping relationship f1(u->m), calculate the value of the selected top-level indicator.

[0058] Step 6.3: Calculate the test coverage results. According to the top-level indicator values ​​calculated in the previous step, calculate the test coverage rate:

[0059]

[0060] Step 6.4: Determine test coverage;

[0061] When Result<100%, add test cases and return to step 5;

[0062] When the test coverage result Result ≥ 100%, it means that the test coverage meets the requirements and the intelligent software test is ended.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. According to the quality attributes of deep learning intelligent software, a three-layer intelligent software test coverage indicator system is established from the data, algorithm, model and system layers, which is more suitable for the intelligent software test coverage measurement based on deep learning technology;

[0065] 2. Based on the characteristics of deep learning intelligent software testing, establishing a closed-loop process for test coverage measurement for deep learning intelligent software can effectively improve the effectiveness and accuracy of intelligent software test coverage measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of the test coverage measurement method for deep learning intelligent software of the present invention. DETAILED DESCRIPTION

[0067] The exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0068] A certain image target recognition software uses YOL0V8 as the framework and selects RSVG data sets with moderate data volume, clear annotations, and regular data for training and verification. It can effectively identify dynamic and static targets such as airports, ports, bridges, airplanes, ships, and vehicles.

[0069] The software-related information includes: user requirements must include model coverage and system coverage related indicators; according to the test objective requirements, algorithm coverage related requirements must be included; the standard I S0 / I EC 23053-2022 is selected; the software level is important software, and the algorithm type is classification task.

[0070] To carry out test coverage measurement for the intelligent software, a three-layer intelligent software test coverage indicator system should be established from four aspects: data, algorithm, model and system. It is more suitable for intelligent software test coverage measurement based on deep learning technology. Establishing a closed-loop process for test coverage measurement for deep learning intelligent software can effectively improve the effectiveness and accuracy of intelligent software test coverage measurement. In combination with this work, the following will explain how to use the test coverage measurement method for deep learning intelligent software proposed in the present invention. The main steps are shown in the attached Figure 1 .

[0071] Step 1: Establish an intelligent software test coverage measurement model.

[0072] With the goal of covering the intelligent software quality measurement attribute indicator set, the intelligent software test coverage measurement model is determined; the specific description is as follows:

[0073] Step 1.1: Establish an indicator system.

[0074] Define the top-level indicator set TU of the intelligent software test coverage measurement indicator set u ={TU1,TU2,…,TU n}、The second-level indicator set TM={TM1,TM2,…,TM k} and the underlying indicator set TD={TD1,TD2,…,TD l}, intelligent software test coverage measurement indicator set TC = {TU, TM, TD}; and determine the relationship between indicators according to the indicator definition.

[0075] In this embodiment, the specific top-level indicator set TU u ={TU1,TU2,TU3,TU4}, the second-layer indicator set TM = {TM1,TM2,…,TM 10} and the underlying indicator set TD={TD1,TD2,…,TD 25}, intelligent software test coverage measurement index system TC = {TU, TM, TD}. Among them, the top-level indicators TU1, TU2, TU3, TU4 are data coverage, algorithm coverage, model coverage, and system coverage, respectively, and the second-level indicators TM1, TM2, ..., TM 10 They are data integrity, data correctness, algorithm adaptability, algorithm reliability, model stability, model functionality, model adaptability, system function, system performance, and other system attributes. The underlying indicator set is TD1, TD2,…, TD 25 They are dataset coverage, data distribution coverage, label space coverage, feature space coverage, time coverage, data type coverage, time complexity and space complexity, code coverage, regression task evaluation indicators, classification task evaluation indicators, response surface analysis, abnormal situation coverage, feature coverage, edge case coverage, generalization ability, scenario coverage, user group coverage, system function coverage, function implementation quality, system compliance, performance indicator coverage, resource usage coverage, fault recovery coverage, system reliability, and system security.

[0076] According to the definition of each indicator, the relationship between indicators at each level is determined, as shown in Table 1.

[0077] Table 1

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] Step 1.2: Establish a mapping relationship.

[0084] A mapping relationship f1(u->m) is established between the top-level indicator TU and the second-level indicator TM. According to Table 1, the mapping relationship f1(u->m) is established as follows:

[0085]

[0086] Among them, V1 j Indicates the second-level indicator TM j At the top level indicator TU i Calculate weight, V1 j The default value is The weight V1 j Adjustments can be made based on the selection strategy in step 2; see step 3 for specific adjustments.

[0087] Step 1.3: Establish a secondary mapping relationship.

[0088] A secondary mapping relationship f2(m->d) is established between the second-layer indicator TM and the bottom-layer indicator TD. According to Table 1, the mapping relationship f2(m->d) is established as follows:

[0089]

[0090] Among them, V2 k Represents the underlying indicator TD k In the second-level indicator TM j Calculate weights, V2 k The default value is Weight V2 k Adjustments will be made later. For details, see step 3.

[0091] Step 1.4: Construct the intelligent software test coverage measurement model AITC = {TC, f1, f2}.

[0092] Step 2: Determine the metric selection strategy S.

[0093] Determine s1 according to user requirements (see step 2.1), determine s2 according to test objectives (see step 2.2), determine s3 according to selection criteria (see step 2.3), determine s4 according to software level (see step 2.4), determine s5 according to algorithm type (see step 2.5), and determine the selection strategy S.

[0094] Step 2.1: Determine according to user requirements. If the user requirements clearly include indicator TD k , then the element M1 in s1 k Value M1 k =1, if the user requirement explicitly does not include the indicator TD k , then s1 takes the value M1 k =0, otherwise s1 is a random value M1 k =rand():

[0095]

[0096] Among them, M1 k For the indicator TD k The measurement value in the user requirement; s1 is the user requirement measurement indicator.

[0097] It should be noted that when an indicator is explicitly required to be included, all subordinate indicators of the indicator are also explicitly required to be included by default.

[0098] In this embodiment, according to user requirements, the value of s1 is determined according to formula (3):

[0099]

[0100] Step 2.2: Determine according to the test objectives. If the test objectives clearly include the indicator TD k , then s2 takes the value M2 k =1, if the test objective explicitly does not include indicator TD k , then s2 takes the value M2 k =0, otherwise s2 is a random value M2 k =rand():

[0101]

[0102] In this embodiment, according to the test target, the value of s2 is determined according to formula (4):

[0103]

[0104] Among them, M2 k For the indicator TD k The measurement value in the test target; s2 is the test target measurement indicator.

[0105] Step 2.3: Determine according to the selection criteria, if the selection criteria clearly include indicator TD k , then s3 takes the value M3 k =1, if the selection standard does not contain the indicator TD k , then s3 takes the value M3 k =0, otherwise s3 is a random value M3 k =rand():

[0106]

[0107] Among them, M3 k For the indicator TD k The measurement value in the selection standard; s3 is the measurement indicator of the selection standard.

[0108] According to the selection criteria, the value of s3 is determined according to formula (5):

[0109] s3={M3 k =rand(),k=1,...,25}.

[0110] Step 2.4: Determine according to the software level. If the software level is critical software, s4 takes the value M4. k =1, the software level is important software, then s4 is the random value M4 k = rand(), otherwise s4 is half of the random value M4 k =rand() / 2:

[0111]

[0112] Among them, M4 k For the indicator TD k Measurement value at the software level; s4 is the software level measurement indicator.

[0113] According to the software level, the value of s4 is determined according to formula (6):

[0114] s4={M4 k =rand(),k=1,...,25}.

[0115] Step 2.5: Determine according to the algorithm type. If the algorithm type is a regression task, s5 takes the value M5 10 =1,M5 11 =0, if the algorithm type is a classification task, then s5 takes the value M5 11 =1,M5 10 =0, other s5 is a random value M5 k =rand(),k≠10,k≠11:

[0116]

[0117] According to the algorithm type, the value of s5 is determined according to formula (7):

[0118]

[0119] Among them, M5 k is the measurement value in the algorithm type; s5 is the algorithm type measurement indicator.

[0120] Step 2.6: Determine the metric selection strategy. Select the metric selection strategy according to steps 2.1-2.5 and determine the metric selection strategy S, where S = {s1, s2, s3, s4, s5}.

[0121] Step 3: Adjust the first weight V1 j and the second weight V2 k .

[0122] According to formula (8) and formula (9), the first weight V1 j and the second weight V2 k Make adjustments, and the adjustment algorithm is as follows:

[0123]

[0124]

[0125] Step 4: Select the metric set.

[0126] According to the determined metric selection strategy S, the specific metric set STC is determined from the intelligent software test coverage metric set:

[0127] STC={if(V2 k ≠0)TD k , k=1,...,25} (10).

[0128] In this embodiment, according to the determined metric selection strategy S, the metric set STC of the software is determined from the intelligent software test coverage metric set TC, as shown in Table 2 for details.

[0129] Table 2

[0130]

[0131]

[0132] Step 5: Get the indicator value of the underlying metric.

[0133] According to the metric selected in step 4, perform intelligent software testing, collect parameter information involved in the metric definition in STC, and use the metric definition (see Table 1 for details) to obtain the indicator value of the underlying metric.

[0134] Step 6: Get test coverage.

[0135] Step 6.1: Calculate the value of the selected second-layer indicator. According to the value of the bottom-layer indicator, refer to the secondary mapping relationship f2 (m->d) defined in step 1.3 to calculate the value of the selected second-layer indicator.

[0136] Step 6.2: Calculate the value of the selected top-level indicator. According to the value of the second-level indicator, refer to the first-order mapping relationship f1 (u->m) defined in step 1.2 to calculate the value of the selected top-level indicator.

[0137] In this embodiment, the data test coverage is TU1=0.9, the algorithm test coverage is TU2=0.95, the model test coverage is TU3=0.9, and the system test coverage is TU4=0.85.

[0138] Step 6.3: Calculate the test coverage results. According to the top-level indicator values ​​calculated in the previous step, calculate the test coverage rate:

[0139]

[0140] In this embodiment, the test coverage is obtained:

[0141]

[0142] Step 6.4: Determine test coverage.

[0143] When Result<100%, add test cases and return to step 5.

[0144] When the test coverage result Result ≥ 100%, it means that the test coverage meets the requirements, and the intelligent software test can be ended at this time.

[0145] When the test coverage result is <100%, it means that the test coverage does not meet the requirements. At this time, it is necessary to add test cases and re-execute the test. The specific method of adjusting the test cases is determined according to the top-level indicators, which are:

[0146] (1) When the data distribution coverage (TD2) deviates too much from the actual distribution, for example, when the data distribution coverage deviates from the actual distribution by more than 5%, the distribution of the test data should be corrected.

[0147] (2) Abnormal situation coverage (TD 12) The relevant test content is relatively small. For example, when the abnormal situation coverage is less than 95%, the test content of the abnormal situation in the algorithm should be increased.

[0148] (3) Feature coverage (TD 13 ) There are many test contents. For example, when the feature coverage rate is greater than 100%, it means that some features are covered repeatedly, and their test coverage metric construction process and test process should be sorted out.

[0149] (4) Fault recovery coverage (TD 23 ) is too low. For example, when the fault recovery coverage is less than 95%, it is necessary to construct more fault conditions to increase the fault recovery coverage.

[0150] In this embodiment, the test case is adjusted as shown in Table 3, and the test is executed again by returning to step 5.

[0151] Table 3

[0152]

[0153] After adjusting the test cases, the data test coverage is TU1=1, the algorithm test coverage is TU2=1, the model test coverage is TU3=1, the system test coverage is TU4=1, and the test coverage is:

[0154]

[0155] When the test coverage requirements are met, the intelligent software testing is completed.

[0156] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A test coverage measurement method for deep learning intelligent software, characterized by: It includes: Step 1: Establish an intelligent software test coverage measurement model; Step 1.1: Define the intelligent software test coverage measurement indicator set TC, TC = {TU, TM, TD}, where the top-level indicator set TU = {TU1, TU2, TU3, TU4}, the second-level indicator set TM = {TM1, TM2, …, TM 10 } and the underlying indicator set TD={TD1,TD2,…,TD 25 }; Step 1.2: Establish a mapping relationship; Establish a primary mapping relationship f1(u->m) between the top-level indicator TU and the second-level indicator TM; Step 1.3: j establishes a secondary mapping relationship; Establish a secondary mapping relationship f2 (m->d) between the second-layer indicator TM and the bottom-layer indicator TD; Step 1.4: Construct the intelligent software test coverage measurement model AITC = {TC, f1, f2}; Step 2: Determine the metric selection strategy S; Determine the user requirement metric s1 according to the user requirement, determine the test target metric s2 according to the test target, determine the selection standard metric s3 according to the selection standard, determine the software level metric s4 according to the software level, determine the algorithm type metric s5 according to the algorithm type, and determine the metric selection strategy S; S = {s1, s2, s3, s4, s5}; Step 3: Adjust the first weight Vu j and the second weight Vt k ; Step 4: Select the metric set; According to the metric selection strategy S, the specific metric set STC is determined from the intelligent software test coverage metric set: STC={if(V2 k ≠0)TD k ,k=1,...′25} (10); Step 5: Get the indicator value of the underlying indicator; According to the metric selected in step 4, perform intelligent software testing to obtain the indicator value of the underlying indicator; Step 6: Get test coverage; According to the metric set STC and the primary mapping relationship and secondary mapping relationship in the intelligent software test coverage metric model, after obtaining the indicator value of the top-level indicator, the test coverage is obtained according to formula (11): Among them, Result is the test coverage, TU u is the top-level indicator; If the test coverage Result is <100%, add test cases and return to step 5. Otherwise, the intelligent software test ends. Specifically: When the test coverage result Result < 100%, it means that the test coverage does not meet the requirements. At this time, you need to add test cases and return to step 5 to re-execute the test; When the test coverage result Result ≥ 100%, it means that the test coverage meets the requirements, and the intelligent software test can be ended at this time.

2. The test coverage measurement method for deep learning intelligent software according to claim 1, characterized in that: In the step 1.1, the intelligent software test coverage measurement index set TC, TC = {TU, TM, TD}, wherein the top-level index set TU = {TU1, TU2, TU3, TU4}, the second-level index set TM = {TM1, TM2, ..., TM 10 } and the underlying indicator set TD={TD1,TD2,…,TD 25 }, specifically: The top-level indicators TU1, TU2, TU3, and TU4 are data coverage, algorithm coverage, model coverage, and system coverage, respectively. The second-level indicators TM1 to TM 10 They are data integrity, data correctness, algorithm adaptability, algorithm reliability, model stability, model functionality, model adaptability, system function, system performance, and other system attributes. The underlying indicator set TD1 to TD 25 They are dataset coverage, data distribution coverage, label space coverage, feature space coverage, time coverage, data type coverage, time complexity and space complexity, code coverage, regression task evaluation indicators, classification task evaluation indicators, response surface analysis, abnormal situation coverage, feature coverage, edge case coverage, generalization ability, scenario coverage, user group coverage, system function coverage, function implementation quality, system compliance, performance indicator coverage, resource usage coverage, fault recovery coverage, system reliability, and system security.

3. The test coverage measurement method for deep learning intelligent software according to claim 2, characterized in that: The primary mapping relationship f1(u->m) in step 1.2 is specifically: Among them, Vu j Indicates the second-level indicator TM j At the top level indicator TU i Calculate weight, V1 j The default value is The secondary mapping relationship f2 (m->d) in step 1.3 is specifically: Among them, V2 k Represents the underlying indicator TD k In the second-level indicator TM j Calculate weights, V2 k The default value is 4. The test coverage measurement method for deep learning intelligent software according to claim 2, characterized in that: The step 2: determining the measurement indicator selection strategy S, specifically includes the following steps: Step 2.1: Determine s1 according to user requirements; Among them, M1 k For the indicator TD k The measurement value in the user requirement; s1 is the user requirement measurement index; Step 2.2: Determine s2 according to the test objectives; Among them, M2 k For the indicator TD k The measurement value in the test target; s2 is the test target measurement indicator; Step 2.3: Determine s3 according to the selection criteria; Among them, M3 k For the indicator TD k The measurement value in the selection standard; s3 is the measurement index of the selection standard; Step 2.4: Determine s4 according to the software level; Among them, M4 k For the indicator TD k The measurement value at the software level; s4 is the software level measurement indicator; Step 2.5: Determine s5 according to the algorithm type; Among them, M5 k is the measurement value in the algorithm type; s5 is the algorithm type measurement indicator; Step 2.6: Determine the metric selection strategy. Select the metric selection strategy according to steps 2.1-2.5 and determine the metric selection strategy S, where S = {s1, s2, s3, s4, s5}.

5. The test coverage measurement method for deep learning intelligent software according to claim 4, characterized in that: Step 3 adjusts the first weight Vu j and the second weight Vt k , specifically:

6. The test coverage measurement method for deep learning intelligent software according to claim 5, characterized in that: The step 6 is specifically as follows: Step 6.1: Calculate the value of the selected second-layer indicator. According to the value of the bottom-layer indicator and the secondary mapping relationship f2(m->d), calculate the value of the selected second-layer indicator. Step 6.2: Calculate the values of the selected top-level indicators. According to the values of the second-level indicators and the one-time mapping relationship f1(u->m), calculate the values of the selected top-level indicators: Step 6.3: Calculate the test coverage result. According to the values of the top-level indicators calculated in the previous step, calculate the test coverage rate: Step 6.4: Judge the test coverage rate; When Result < u00%, add test cases and return to Step 5; When the test coverage result Result ≥ u00%, it means that the test coverage meets the requirements, and the intelligent software test is ended.

Images