Test case generation method and system for unmanned aerial vehicle aerial test requirement formalization
Patent Information
- Application Number
- CN202611426977.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-15
- Publication Date
- 2026-10-30
AI Technical Summary
[0004]然而，扩散模型能够生成高质量视觉图像，但现有方法通常依赖人工设计提示词（Prompt）控制生成过程，缺少将领域测试需求中的场景语义、环境条件以及物理参数自动映射为扩散模型输入参数的方法，使得生成测试数据过程存在较大的随机性和不确定性，难以保证生成结果的一致性
[0045]本申请通过模型解析器对无人机航拍测试需求模型进行信息提取，生成结构化测试需求数据，并通过将结构化测试需求数据中的语义描述和数值型描述转换为正向提示词，并调用预先设定的负向提示词，生成对应的控制参数。以此避免了人工设计提示词造成的不确定性，保证了生成结果的一致性。
Smart Images
Figure CN123011730A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of software testing technology, specifically to a formalized test case generation method and system for drone aerial photography testing requirements. Background Technology
[0002] With the widespread application of drones in disaster monitoring, forestry inspection, and agricultural remote sensing, their visual perception systems need to maintain stable and reliable operation under complex environments and extreme conditions. Drone visual perception algorithms typically rely on a large amount of image data under different environmental conditions for testing and verification. However, in actual testing, boundary scenarios such as extreme weather, complex lighting, and sensor anomalies are difficult to acquire quickly through real-world data collection methods.
[0003] Model-based systems engineering (MBSE) integrates requirements analysis, design verification, and test management by establishing system models. SysML, a key modeling language in the MBSE field, uses Block Definition Diagrams (BDDs) to describe the structural components, module relationships, and interface information of a system, providing effective support for modeling complex systems.
[0004] However, while diffusion models can generate high-quality visual images, existing methods typically rely on manually designed prompts to control the generation process. They lack methods to automatically map scene semantics, environmental conditions, and physical parameters in domain testing requirements to the input parameters of the diffusion model. This results in significant randomness and uncertainty in the process of generating test data, making it difficult to guarantee the consistency of the generated results. Summary of the Invention
[0005] The purpose of this invention is to address the problem that existing technologies typically rely on manually designed prompts to control the generation process, lacking a method to automatically map scene semantics, environmental conditions, and physical parameters in domain testing requirements to diffusion model input parameters. This results in significant randomness and uncertainty in the test data generation process, making it difficult to guarantee the consistency of the generated results. The invention provides a formalized test case generation method and system for UAV aerial photography testing requirements.
[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:
[0007] A formalized test case generation method for drone aerial photography testing requirements includes the following steps:
[0008] Step 1: Construct an extended SysML model to meet the needs of drone aerial photography testing;
[0009] Step 2: Based on the SysML extended model and combined with the test requirements, construct the drone aerial photography test requirement model;
[0010] Step 3: Use the model parser to extract information from the UAV aerial photography test requirement model and generate structured test requirement data;
[0011] Step 4: Based on the structured test requirement data, convert the semantic and numerical descriptions in the structured test requirement data into positive prompt words, and call the pre-set negative prompt words to generate the corresponding control parameters;
[0012] Step 5: Call the text-generative diffusion model and generate a boundary scene test image that meets the requirements of UAV aerial photography vision testing based on the control parameters.
[0013] Furthermore, the specific steps of step 1 are as follows:
[0014] Based on the SysML metamodel extension mechanism, the Block elements within the block definition graph are extended into a metamodel, and a new TestComponent metaclass is added to the metamodel to represent test scenario information, thus obtaining the extended metamodel.
[0015] Then, based on the extended metamodel, physical entity stereotypes are constructed. <physicalblock>> Test Information Structure < <testinfoblock>>and inherited from the test information stereotype< <testinfoblock>>structural type< <generativetestblock>After that, we obtained the SysML extended model for drone aerial photography testing.
[0016] The physical entity construct < <physicalblock>Used to describe physical entities, including sensors, actuators, and controllers of the UAV;
[0017] The test information stereotype < <testinfoblock>Used to carry various types of test information;
[0018] The inheritance from < <testinfoblock>>structural type< <generativetestblock>>Used to mark model elements that need to participate in the generation of generative test data;
[0019] The TestComponent metaclass contains a scenario name, a semantic description, and a numeric description attribute.
[0020] Furthermore, the specific steps of step 2 are as follows:
[0021] Based on the SysML extended model constructed in step one, an initial test requirement model is created in the block definition diagram. For test scenarios that require the generation of test data, the corresponding TestComponent property is defined in the initial test requirement model, and < is applied to the test scenarios that require the generation of test data. <generativetestblock>>Construct the model to obtain the final test requirement model.
[0022] Furthermore, the specific steps of step 3 are as follows:
[0023] Build a model parser and take the test requirement model as input to identify the < <generativetestblock>>Tag model elements and extract scene names, semantic descriptions, and numerical descriptions from the elements to obtain the output structured test requirement data.
[0024] Furthermore, the specific steps of step 4 are as follows:
[0025] A prompt word compiler is built. The prompt word compiler first extracts the semantic description as the head base of the positive prompt word to form the basic semantic description that the diffusion model can recognize. Then, the key-value pairs in the numerical description are transformed by word weight, and each parameter is strictly formatted into a (keyword: weight) structure to generate feature control labels. Finally, the basic semantic description, feature control labels and pre-set negative prompt words are concatenated by commas to obtain the control parameters.
[0026] A test case generation system for formalizing test requirements for drone aerial photography testing is provided. The system includes a requirement model building module, a model parser module, a prompt word compiler module, and a diffusion model module.
[0027] The requirement model building module is used to build a SysML extended model for UAV aerial photography test requirements, and based on the SysML extended model and combined with the test requirements content, to build a UAV aerial photography test requirement model.
[0028] The model parser module is used to extract information from the UAV aerial photography test requirement model and generate structured test requirement data.
[0029] The prompt word compiler module converts the semantic and numerical descriptions in the structured test requirement data into positive prompt words based on the structured test requirement data, and calls the pre-set negative prompt words to generate corresponding control parameters.
[0030] The diffusion model module is used to call the text-generating diffusion model and generate boundary scene test images that meet the requirements of UAV aerial photography vision testing based on control parameters.
[0031] Furthermore, the specific steps in the requirement model construction module for building the SysML extended model for UAV aerial photography testing requirements are as follows:
[0032] Based on the SysML metamodel extension mechanism, the Block elements within the block definition graph are extended into a metamodel, and a new TestComponent metaclass is added to the metamodel to represent test scenario information, thus obtaining the extended metamodel.
[0033] Then, based on the extended metamodel, physical entity stereotypes are constructed. <physicalblock>> Test Information Structure < <testinfoblock>>and inherited from the test information stereotype< <testinfoblock>>structural type< <generativetestblock>After that, we obtained the SysML extended model for drone aerial photography testing.
[0034] The physical entity construct < <physicalblock>Used to describe physical entities, including sensors, actuators, and controllers of the UAV;
[0035] The test information stereotype < <testinfoblock>Used to carry various types of test information;
[0036] The inheritance from < <testinfoblock>>structural type< <generativetestblock>>Used to mark model elements that need to participate in the generation of generative test data;
[0037] The TestComponent metaclass contains a scenario name, a semantic description, and a numeric description attribute.
[0038] Furthermore, the specific steps in the requirement model construction module for constructing the UAV aerial photography test requirement model based on the SysML extended model and combined with the test requirement content are as follows:
[0039] Based on the SysML extended model constructed in step one, an initial test requirement model is created in the block definition diagram. For test scenarios that require the generation of test data, the corresponding TestComponent property is defined in the initial test requirement model, and < is applied to the test scenarios that require the generation of test data. <generativetestblock>>Construct the model to obtain the final test requirement model.
[0040] Furthermore, the model parser module specifically performs the following steps:
[0041] Build a model parser and take the test requirement model as input to identify the < <generativetestblock>>Tag model elements and extract scene names, semantic descriptions, and numerical descriptions from the elements to obtain the output structured test requirement data.
[0042] Furthermore, the prompt word compiler module specifically performs the following steps:
[0043] A prompt word compiler is built. The prompt word compiler first extracts the semantic description as the head base of the positive prompt word to form the basic semantic description that the diffusion model can recognize. Then, the key-value pairs in the numerical description are transformed by word weight, and each parameter is strictly formatted into a (keyword: weight) structure to generate feature control labels. Finally, the basic semantic description, feature control labels and pre-set negative prompt words are concatenated by commas to obtain the control parameters.
[0044] The beneficial effects of this invention are:
[0045] This application uses a model parser to extract information from the UAV aerial photography test requirement model, generating structured test requirement data. It then converts the semantic and numerical descriptions in the structured test requirement data into positive prompts and calls pre-defined negative prompts to generate corresponding control parameters. This avoids the uncertainty caused by manually designed prompts and ensures the consistency of the generated results. Attached Figure Description
[0046] Figure 1 This is the overall flowchart of this application;
[0047] Figure 2 This is a schematic diagram illustrating the expansion process of the BDD metamodel.
[0048] Figure 3 This is a schematic diagram of the expansion process of the BDD construct. Detailed Implementation
[0049] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.
[0050] Specific Implementation Method 1: The formalized test case generation method for UAV aerial photography testing requirements described in this implementation method specifically includes:
[0051] Step S1: Based on the SysML metamodel extension mechanism, the Block elements in the block definition diagram are extended, and a new TestComponent metaclass is added at the metamodel level to describe test scenario information. The TestComponent includes scenario name, semantic description and numerical description. The corresponding structural extension is constructed according to the extended metamodel to realize the unified modeling of system entity information and test information.
[0052] Step S2: Construct a model parser, taking the test information model as input, identify the test information constructs in the model, and extract the scene names, semantic descriptions and numerical descriptions contained therein to generate structured test requirement data.
[0053] Step S3: Build a prompt word compiler. Based on the structured test requirement data, convert the semantic description and numerical description into positive prompt words that conform to the input rules of the diffusion model, and call the pre-set negative prompt words to generate the corresponding control parameters.
[0054] Step S4: Call the text-generative diffusion model to automatically generate a boundary scene test image that meets the requirements of UAV aerial photography vision testing based on the parameters generated by the prompt word compiler.
[0055] In step S1, a physical entity stereotype is constructed based on the extended metamodel. <physicalblock>> is used to describe the sensors, actuators, and controllers in a drone; constructs the test information stereotype < <testinfoblock>> is used to describe test information and further construct < <generativetestblock>>Inherited from< <testinfoblock>> is used to identify test model elements that need to participate in the generation of generative test data so that the model parser can recognize and extract them.
[0056] In step S2, the model parser reads the test requirement model file and identifies < <generativetestblock>>Tag model elements and extract structured test information from them.
[0057] In step S3, the prompt word compiler converts the semantic description in the test requirement model into text prompt information that the diffusion model can recognize, and adjusts the corresponding visual feature weights according to the numerical parameters to realize the mapping between test requirements and generated image features.
[0058] Example:
[0059] Combined with appendix Figures 1 to 3 This embodiment describes a formalized test case generation method for UAV aerial photography testing requirements, used to automatically generate boundary scene test images from a formalized test requirement description.
[0060] Step 1: Construct a SysML extended model to meet the needs of drone aerial photography testing.
[0061] Specifically, a test requirement extension model is established based on the meta-model extension mechanism in the SysML modeling tool. While retaining the original Block structure description capabilities of SysML, the Block meta-elements are extended to add a new TestComponent meta-class for describing test information. The TestComponent is used to store test scenario-related information, including scenario name, semantic description, and numerical description.
[0062] Furthermore, the corresponding model stereotype is constructed based on the extended metamodel. Specifically, the construction of < <physicalblock>>Used to describe the physical entity components of an unmanned aerial vehicle (UAV) system, including devices such as sensors, actuators, and controllers; constructing< <testinfoblock>>Used to describe system test-related information; further construct< <generativetestblock>>Inherited from the above< <testinfoblock>> is used to identify test model elements that need to participate in the generation of generative test data.
[0063] Specifically, in the UAV aerial photography visual perception system model, through < <physicalblock>Describe the system composition of the drone platform, airborne camera equipment, and vision processing module, etc., through < <generativetestblock>By associating corresponding test requirement information, a unified modeling of system structure information and test requirement information can be achieved.
[0064] Step 2: Establish a drone aerial photography test requirement model and generate structured test requirement data.
[0065] Specifically, based on the SysML extended model constructed in step one, a test requirement model is created in the block definition diagram, and applied to the test scenarios that require the generation of test data. <generativetestblock>>Structure type.
[0066] Furthermore, based on the test requirements, corresponding TestComponent attributes are defined in the test information model. The scenario name identifies the current test scenario type; the semantic description describes the environmental factors and target characteristics in the test scenario in textual form, such as heavy rain, light rain, low light, and strong light; the numerical description describes the corresponding environmental factors or target characteristics numerically, for example, using normalized values from 0 to 1 to describe rainfall intensity, where 0.5 represents moderate rain and 0.2 represents light rain.
[0067] For example, in the test scenario of UAV visual target detection, the test scenario name can be defined as "target detection in low light during rain", the semantic description is "target image taken by UAV under heavy rain conditions", and the numerical description includes parameters such as rainfall intensity, ambient brightness and target size. The rainfall intensity can be represented by a value between 0 and 1, thus forming a structured test requirement.
[0068] Step 3: Extract test requirement information using the model parser and convert it into generation parameters.
[0069] Specifically, the completed SysML test requirement model is exported as a model exchange file, and the model parser reads the model file and parses the stereotype information in the model.
[0070] Furthermore, the model parser identifies applications. <generativetestblock>>Construct test model elements and extract the scenario names, semantic descriptions, and numerical descriptions contained therein, and convert them into structured test requirement data.
[0071] Specifically, the structured test requirement data is organized as key-value pairs and used as input for the subsequent prompt word compilation process, thereby establishing a data connection between the test requirement model and the generative model.
[0072] Step four: Generate the diffusion model input parameters using the prompt word compiler.
[0073] Specifically, the prompt word compiler receives the structured test requirement data output by the model parser and transforms the semantic and numerical information in the test requirements according to the input requirements of the diffusion model.
[0074] Furthermore, the prompt word compiler first directly extracts the semantic description as the initial base of the positive prompt word to form a basic semantic description that the diffusion model can recognize. Then, it performs lexical weight transformation on the key-value pairs in the numerical description, strictly formatting each parameter into a (keyword: weight) structure to generate feature control labels. Finally, it concatenates the basic semantic description, feature control labels, and fixed negative prompt words separated by commas, and packages them into a standard parameter set, thereby realizing the mapping between environmental factors, target features, scene constraints, and generated image features in the test requirements.
[0075] Specifically, for example, in response to the test requirement of "target detection in low light during rainy weather", the prompt word compiler directly uses the semantic content such as drone aerial photography, rainy weather environment, and low light conditions in the semantic description as the beginning of the positive prompt word, and converts the key-value pairs such as rainfall intensity and light intensity in the numerical description into feature control labels in the form of (rain: 0.8) and (illumination: 0.3). Then, it assembles them with the fixed negative prompt word separated by commas to form complete generation control parameters, which can be called by the diffusion model to generate the corresponding test image.
[0076] Step 5: Generate test images based on a text-generative diffusion model.
[0077] Specifically, the standard parameter set generated in step four is input into the text-generative diffusion model. The standard parameter set includes at least basic semantic descriptions, feature control labels, and fixed negative prompt words. The diffusion model generates drone aerial visual test images based on the standard parameter set.
[0078] Furthermore, during the generation process, the diffusion model extracts scene semantic features based on the basic semantic description and combines feature control labels to constrain weather conditions, lighting environment, target scale, target sharpness, and background complexity in the image, so that the generated result is consistent with the environmental factors and target features in the test requirements. At the same time, fixed negative prompts are used to suppress interference content that is irrelevant to the test requirements, thereby improving the scene consistency and test relevance of the generated image.
[0079] Specifically, during the generation process, fixed random seeds, fixed sampling steps, and fixed resolution can be set to reduce generation differences caused by the random sampling process of the diffusion model, enabling the same test requirement to obtain highly consistent test data samples when called multiple times. For the "rainy day low light target detection" test requirement, the diffusion model generates aerial test images that meet the conditions of heavy rain, low light, and limited target visibility based on the input basic semantic descriptions such as UAV aerial photography, rainy environment, and low light conditions, as well as feature control labels such as (rain: 0.8) and (illumination: 0.3).
[0080] The final generated boundary scene test images can be used as test data for UAV visual perception algorithms to verify the performance of intelligent algorithms such as target detection and image recognition under complex environmental conditions.
[0081] This embodiment extends the SysML model to unify the description of UAV system structure information and test requirements information, and uses a model parser and prompt word compiler to realize the automatic conversion of test requirements into diffusion model generation parameters, ultimately achieving automated test data generation for UAV aerial vision testing requirements.
[0082] This application extends the SysML metamodel and constructive mechanisms to unify the modeling of UAV system structural information and test requirement information, transforming test requirements from traditional natural language descriptions into formal model descriptions, thereby improving the standardization and parsability of test requirement expression.
[0083] This application utilizes a diffusion model to generate high-quality visual test data, which can quickly construct boundary scenarios such as extreme weather, complex lighting, and sensor anomalies that are difficult to obtain in real environments. By constraining the random generation process of the diffusion model with fixed generation parameters, the consistency and reproducibility of the test data generation process are improved, providing stable and reliable data support for the testing of UAV visual perception algorithms.
[0084] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.< / generativetestblock> < / generativetestblock> < / generativetestblock> < / physicalblock> < / testinfoblock> < / generativetestblock> < / testinfoblock> < / physicalblock> < / generativetestblock> < / testinfoblock> < / generativetestblock> < / testinfoblock> < / physicalblock> < / generativetestblock> < / generativetestblock> < / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock> < / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock> < / generativetestblock> < / generativetestblock> < / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock> < / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock>
Claims
1. A formalized test case generation method for drone aerial photography testing requirements, characterized in that, Includes the following steps: Step 1: Construct an extended SysML model to meet the needs of drone aerial photography testing; Step 2: Based on the SysML extended model and combined with the test requirements, construct the drone aerial photography test requirement model; Step 3: Use the model parser to extract information from the UAV aerial photography test requirement model and generate structured test requirement data; Step 4: Based on the structured test requirement data, convert the semantic and numerical descriptions in the structured test requirement data into positive prompt words, and call the pre-set negative prompt words to generate the corresponding control parameters; Step 5: Call the text-generative diffusion model and generate a boundary scene test image that meets the requirements of UAV aerial photography vision testing based on the control parameters.
2. The test case generation method for formalizing UAV aerial photography testing requirements according to claim 1, characterized in that, The specific steps of step 1 are as follows: Based on the SysML metamodel extension mechanism, the Block elements within the block definition graph are extended into a metamodel, and a new TestComponent metaclass is added to the metamodel to represent test scenario information, thus obtaining the extended metamodel. Then, based on the extended metamodel, physical entity stereotypes are constructed. <physicalblock>> Test Information Structure < <testinfoblock>>and inherited from the test information stereotype< <testinfoblock>>structural type< <generativetestblock> After that, we obtained the SysML extended model for drone aerial photography testing.< / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock> The physical entity construct < <physicalblock> Used to describe physical entities, including sensors, actuators, and controllers of the UAV;< / physicalblock> The test information stereotype < <testinfoblock> Used to carry various types of test information;< / testinfoblock> The inheritance from < <testinfoblock>>structural type< <generativetestblock> >Used to mark model elements that need to participate in the generation of generative test data;< / generativetestblock> < / testinfoblock> The TestComponent metaclass contains a scenario name, a semantic description, and a numeric description attribute.
3. The method for generating test cases for formalizing UAV aerial photography testing requirements according to claim 2, characterized in that, The specific steps of step 2 are as follows: Based on the SysML extended model constructed in step one, an initial test requirement model is created in the block definition diagram. For test scenarios that require the generation of test data, the corresponding TestComponent property is defined in the initial test requirement model, and < is applied to the test scenarios that require the generation of test data. <generativetestblock> >Construct the model to obtain the final test requirement model.< / generativetestblock> 4. The method for generating test cases for formalizing UAV aerial photography testing requirements according to claim 3, characterized in that, The specific steps of step 3 are as follows: Build a model parser and take the test requirement model as input to identify the < <generativetestblock> >Tag model elements and extract scene names, semantic descriptions, and numerical descriptions from the elements to obtain the output structured test requirement data.< / generativetestblock> 5. The test case generation method for formalizing UAV aerial photography testing requirements according to claim 4, characterized in that, The specific steps of step 4 are as follows: A prompt word compiler is built. The prompt word compiler first extracts the semantic description as the head base of the positive prompt word to form the basic semantic description that the diffusion model can recognize. Then, the key-value pairs in the numerical description are transformed by word weight, and each parameter is strictly formatted into a (keyword: weight) structure to generate feature control labels. Finally, the basic semantic description, feature control labels and pre-set negative prompt words are concatenated by commas to obtain the control parameters.
6. A formalized test case generation system for UAV aerial photography testing requirements, characterized in that: The system includes a demand model building module, a model parser module, a prompt word compiler module, and a diffusion model module; The requirement model building module is used to build a SysML extended model for UAV aerial photography test requirements, and based on the SysML extended model and combined with the test requirements content, to build a UAV aerial photography test requirement model. The model parser module is used to extract information from the UAV aerial photography test requirement model and generate structured test requirement data. The prompt word compiler module converts the semantic and numerical descriptions in the structured test requirement data into positive prompt words based on the structured test requirement data, and calls the pre-set negative prompt words to generate corresponding control parameters. The diffusion model module is used to call the text-generating diffusion model and generate boundary scene test images that meet the requirements of UAV aerial photography vision testing based on control parameters.
7. The test case generation system for formalizing UAV aerial photography testing requirements according to claim 6, characterized in that, The specific steps for constructing the SysML extended model for UAV aerial photography testing requirements in the requirement model construction module are as follows: Based on the SysML metamodel extension mechanism, the Block elements within the block definition graph are extended into a metamodel, and a new TestComponent metaclass is added to the metamodel to represent test scenario information, thus obtaining the extended metamodel. Then, based on the extended metamodel, physical entity stereotypes are constructed. <physicalblock>> Test Information Structure < <testinfoblock>>and inherited from the test information stereotype< <testinfoblock>>structural type< <generativetestblock> After that, we obtained the SysML extended model for drone aerial photography testing.< / generativetestblock> < / testinfoblock> < / testinfoblock> < / physicalblock> The physical entity construct < <physicalblock> Used to describe physical entities, including sensors, actuators, and controllers of the UAV;< / physicalblock> The test information stereotype < <testinfoblock> Used to carry various types of test information;< / testinfoblock> The inheritance from < <testinfoblock>>structural type< <generativetestblock> >Used to mark model elements that need to participate in the generation of generative test data;< / generativetestblock> < / testinfoblock> The TestComponent metaclass contains a scenario name, a semantic description, and a numeric description attribute.
8. The test case generation system for formalizing UAV aerial photography testing requirements according to claim 7, characterized in that, The specific steps for constructing the UAV aerial photography test requirement model in the requirement model construction module, based on the SysML extended model and combined with the test requirement content, are as follows: Based on the SysML extended model constructed in step one, an initial test requirement model is created in the block definition diagram. For test scenarios that require the generation of test data, the corresponding TestComponent property is defined in the initial test requirement model, and < is applied to the test scenarios that require the generation of test data. <generativetestblock> >Construct the model to obtain the final test requirement model.< / generativetestblock> 9. The test case generation system for formalizing UAV aerial photography testing requirements according to claim 8, characterized in that, The model parser module specifically performs the following steps: Build a model parser and take the test requirement model as input to identify the < <generativetestblock> >Tag model elements and extract scene names, semantic descriptions, and numerical descriptions from the elements to obtain the output structured test requirement data.< / generativetestblock> 10. The test case generation system for formalizing UAV aerial photography testing requirements according to claim 9, characterized in that, The prompt word compiler module specifically performs the following steps: A prompt word compiler is built. The prompt word compiler first extracts the semantic description as the head base of the positive prompt word to form the basic semantic description that the diffusion model can recognize. Then, the key-value pairs in the numerical description are transformed by word weight, and each parameter is strictly formatted into a (keyword: weight) structure to generate feature control labels. Finally, the basic semantic description, feature control labels and pre-set negative prompt words are concatenated by commas to obtain the control parameters.