A fast prediction method for space infrared target characteristics based on language large model

Abstract

This invention discloses a method for rapidly predicting spatial infrared target characteristics based on a large language model. By constructing a causal inference feature extraction framework and integrating multi-source information such as expert knowledge and intelligence reports, the method performs feature inversion on unknown target parameters. Furthermore, a cross-domain migration prediction model for target characteristics is established to enable cross-domain generation of characteristic images, ultimately forming a rapid target characteristic prediction system for complex spatial environments. This method effectively addresses issues such as missing target model parameters, dynamic evolution of characteristics across time domains, and significant differences in characteristics across multiple operating scenarios, significantly improving the efficiency and accuracy of infrared target characteristic prediction.

Description

Technical Field

[0001] The present invention belongs to the field of aerospace technology, and in particular relates to a method for quickly predicting characteristics of space infrared targets based on a large language model. Background Art

[0002] With the rapid development of aerospace technology, missions such as on-orbit maintenance and space debris removal are increasing. These missions often require autonomous identification and characteristic analysis of non-cooperative targets. Data-driven autonomous identification methods require pre-acquisition of real active and passive characteristic data samples of the target for template construction or training set generation. However, practical applications face the following technical challenges:

[0003] (1) Difficulty in obtaining real data: It is extremely difficult to obtain real measured infrared data of space targets, especially infrared characteristic data under different environmental conditions, which is costly and time-consuming.

[0004] (2) Limitations of traditional physical modeling methods: The current mainstream target characteristic data generation methods mainly rely on physical mechanism modeling, which requires precise target parameters and environmental parameters, and the modeling process is complex and consumes a lot of computing resources. These methods have obvious defects:

[0005] (1) Time lag: The modeling process takes a long time and cannot meet the needs of rapid response;

[0006] (2) Weak cross-scenario generalization capability: remodeling is required for new environmental conditions;

[0007] (3) Strong parameter dependence: It is difficult to accurately model unknown targets;

[0008] (4) Dynamic changes in target characteristics: Space target characteristics will change dynamically with time, orbital position, attitude, solar radiation angle and other factors. Traditional static modeling is difficult to accurately reflect this dynamic characteristic.

[0009] (5) Significant differences in multiple working scenarios: The infrared characteristics under different environmental conditions vary significantly, and existing methods are difficult to efficiently achieve cross-scenario characteristic prediction.

[0010] Several solutions have been proposed to address these technical challenges, such as deep learning-based image conversion methods and domain adaptation methods. However, these methods often require large amounts of labeled data, struggle to effectively leverage domain expert knowledge and physical laws, and are unable to meet the demands of rapid algorithm updates and iterations. Therefore, there is an urgent need to develop a new method that can fuse multi-source information and rapidly predict target infrared characteristics. Summary of the Invention

[0011] To overcome the shortcomings of existing technologies, the present invention provides a method for rapid prediction of spatial infrared target characteristics based on a large language model. By constructing a causal inference feature extraction framework, the method integrates expert knowledge, intelligence reports, and other multi-source information to perform feature inversion on unknown target parameters. Furthermore, a cross-domain migration prediction model for target characteristics is established to enable cross-domain generation of characteristic images, ultimately forming a rapid target characteristic prediction system for complex spatial environments. This method effectively addresses issues such as missing target model parameters, dynamic evolution of characteristics across time domains, and significant differences in characteristics across multiple operating scenarios, significantly improving the efficiency and accuracy of infrared target characteristic prediction.

[0012] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0013] Step 1: Multi-source data acquisition and preprocessing;

[0014] Step 2: Causal feature extraction and knowledge representation based on the language model;

[0015] Step 3: Unsupervised domain adaptive image semantic segmentation;

[0016] Step 4: Assign infrared features and generate gradients based on the language model;

[0017] Step 5: Model calibration and accuracy optimization based on measured data.

[0018] Preferably, the step 1 is specifically:

[0019] Step 1-1: Source domain image data: Obtain known source domain images ;

[0020] Step 1-2: Target scenario knowledge: Extract expert knowledge for the target scenario that needs to be predicted, i.e. the target domain ;

[0021] Steps 1-3: Building a domain knowledge base: integrating common sense of physical laws , expert knowledge and domain expertise , forming a comprehensive knowledge base ;

[0022] Steps 1-4: Source domain image and expert knowledge After data preprocessing, the preprocessed source domain image is obtained and structured knowledge representation .

[0023] Preferably, the source domain image It is infrared target data or visible light image data.

[0024] Preferably, the target scene includes space environment, relevant documents, and report information.

[0025] Preferably, the data preprocessing includes image standardization, noise removal, and geometric correction.

[0026] Preferably, the step 2 is specifically as follows:

[0027] Step 2-1: Design of prompt words for the open source language model;

[0028] A structured causal reasoning prompt word engineering design is adopted, including the following components:

[0029] a) Context setting module ;

[0030] b) Causal Link Guidance Module ;

[0031] c) Structured output specification module ;

[0032] d) Uncertainty quantification module ;

[0033] e) Counterfactual Thinking Module ;

[0034] The complete prompt word template is constructed as follows:

[0035]

[0036] in, Indicates a complete prompt word template;

[0037] Step 2-2: Causal feature extraction of unstructured knowledge;

[0038] Define the causal feature extraction function:

[0039]

[0040] in, represents the set of extracted causal features, A language model function representing feature extraction;

[0041] Step 2-3: Causal chain analysis and knowledge representation framework construction;

[0042] Based on the extracted causal features, a causal chain analysis model is constructed:

[0043]

[0044] in, represents a causal diagram, Indicates that the node set corresponds to each part or characteristic attribute of the target. represents the causal relationship between edge sets, i.e. parts, Represents the weight matrix, i.e., the strength of causal relationship;

[0045] By analyzing the structured JSON data output by the language model, a causal graph is automatically constructed; the edge weights are calculated based on the causal relationship strength and confidence given by the model;

[0046] Steps 2-4: Reasoning about target characteristics in a specific state;

[0047] Based on the constructed causal graph, the probability reasoning method is used to calculate the target in a specific state. Temperature values ​​of each part and radiation values :

[0048]

[0049] in, Indicates a specific state and causal diagrams Probability distribution of temperature and radiation values ​​under the conditions, Represents the inference function based on the language model;

[0050] The Monte Carlo sampling method is used to estimate the probability distribution of temperature and radiation values, and the confidence interval is calculated using the uncertainty range given by the language model;

[0051] Output the temperature value of each target part under specific conditions and radiation values ,in is the number of target sites.

[0052] Preferably, the step 3 is specifically:

[0053] Step 3-1: Design the overall architecture of the segmentation model;

[0054] The segmentation model consists of the following four modules:

[0055] a) Feature Encoder : Located at the forefront of the segmentation model, it is responsible for extracting multi-scale features of the image; ResNet-50 is used as the backbone network and pre-trained and initialized; the feature encoder consists of 5 stages, each stage downsampling once to generate feature maps of different scales ;

[0056] b) Domain Adaptation Module : Located between the feature encoder and feature decoder, it consists of multiple domain adversarial discriminators corresponding to features at different levels; through adversarial training, it reduces the difference in feature distribution between the source domain and the target domain; the domain adaptation module performs the following mapping:

[0057]

[0058] in, Output features for the last layer of the encoder, is the domain discriminator, is the feature after domain adaptation; represents the domain adaptation transformation function;

[0059] c) Feature Decoder : Located in the second half of the segmentation model, it is responsible for mapping the features of the feature encoder and domain adaptation module to the segmentation results; the feature decoder adopts a cascade upsampling structure to fuse skip connection features of different scales:

[0060]

[0061] in, Indicates the decoder Layer output features, Indicates the upsampling result of the previous layer feature, represents the skip connection feature of the corresponding encoder layer, represents the feature decoding transformation function;

[0062] d) Split Head : Located at the end of the segmentation model, it maps the features output by the feature decoder to the final segmentation result:

[0063]

[0064] in, represents the output segmentation mask, Represents the output features of the last layer of the feature decoder, Represents the feature mapping function of the segmentation head;

[0065] Step 3-2: Training phase;

[0066] The unsupervised domain adaptation strategy is adopted in the training phase, as follows:

[0067] a) Source domain supervised learning: Use preprocessed source domain images for supervised training to optimize the segmentation loss:

[0068]

[0069] in, represents the cross entropy loss function, represents the segmentation label of the source domain image, represents the feature encoding transformation function;

[0070] b) Adversarial Domain Adaptation: Introducing Domain Discriminators , reducing inter-domain differences through adversarial training:

[0071]

[0072] in, represents the target domain image, represents the natural logarithm function, represents the expected operation on the target domain image, represents the expected operation on the source domain image, represents the adversarial loss function;

[0073] c) Pseudo-label generation: Reasoning on the target domain image to generate high-confidence pseudo-labels:

[0074]

[0075] in, represents the generated pseudo-label, represents the confidence threshold, represents a threshold-based binarization function;

[0076] d) Self-training iteration: Use the generated pseudo labels to iteratively optimize the model:

[0077]

[0078] in, represents the self-training loss function;

[0079] e) Overall optimization goal:

[0080]

[0081] in, represents the regularization term, , , All are weight coefficients;

[0082] Step 3-3: Testing phase;

[0083] a) Input the image to be segmented ;

[0084] b) Extract features through feature encoder: ;

[0085] c) Processing features through domain adaptation module: ;

[0086] d) Generate multi-scale features through feature decoder: ;

[0087] e) Generate the final segmentation result through the segmentation head: ;

[0088] f) Apply post-processing to refine the segmentation boundaries and obtain the final segmentation mask:

[0089]

[0090] Among them, the final segmentation mask Include semantic regions , each region corresponds to a component of the target; represents the post-processing optimization function;

[0091] Preferably, the step 4 is specifically as follows:

[0092] Step 4-1: Assign regional infrared characteristics;

[0093] The temperature value obtained in step 2 and radiation values , assign infrared features to each semantic area segmented in step 3:

[0094]

[0095] in, Represents pixel points The initial infrared characteristic value at Represents pixel points Belong to the region probability; Represent the temperature and radiation values ​​of the i-th region respectively;

[0096] Step 4-2: Generate temperature gradient based on heat transfer model;

[0097] The conduction coefficient matrix between components given by the language model , construct the temperature gradient formula:

[0098]

[0099] in, Indicates the infrared characteristic value after considering the gradient, Indicates the The influence coefficient of the region, represents the attenuation parameter, Indicates a point To area The shortest distance to the border;

[0100] Step 4-3: Target domain adaptation;

[0101] According to the environmental parameters of the target domain , adaptively adjust the infrared characteristics:

[0102]

[0103] in, represents the infrared characteristic value after adapting to the target domain, represents the global adjustment parameters, Represents the influence function of environmental parameters.

[0104] Preferably, the step 5 is specifically as follows:

[0105] Step 5-1: Construction of difference metric and loss function;

[0106] Defining Measured Characteristics and predictive features The difference measure between:

[0107]

[0108] in, represents the mean square error, represents the structural similarity index, and represents the weight coefficient, represents the difference metric loss function;

[0109] Step 5-2: Parameter optimization and calibration;

[0110] Optimize the model parameters by minimizing the discrepancy metric:

[0111]

[0112] in, represents the optimized parameters;

[0113] Step 5-3: Final infrared characteristics generation;

[0114] Generate the final infrared characteristics prediction results based on the optimized parameters :

[0115] .

[0116] The beneficial effects of the present invention are as follows:

[0117] 1. By employing a causal inference feature extraction framework based on a large language model, this paper overcomes the strong dependence on target parameters in traditional methods and achieves the effect of automatically extracting key parameters from unstructured knowledge. The large language model can understand and integrate physical laws, expert knowledge, and domain expertise, extracting causally meaningful prior information from them, significantly reducing the dependence on precise parameters.

[0118] 2. By employing unsupervised domain-adaptive semantic segmentation technology, this method addresses the weak cross-scenario generalization capabilities of traditional methods and achieves effective migration from the source domain to the target domain. This method eliminates the need for labeled data in the target domain, significantly reducing data acquisition costs.

[0119] 3. By employing a large-scale model-guided infrared characteristic assignment and gradient generation method, this invention overcomes the difficulty of traditional methods in expressing inter-component thermal conduction effects, achieving a more realistic infrared characteristic simulation. The temperature gradient generation based on the heat transfer model makes the temperature transition between different target parts more natural and in line with physical laws.

[0120] 4. By employing an adaptive calibration mechanism based on measured data, this method overcomes the time lag associated with traditional methods and enables rapid model updates and iterations. Only a small amount of measured data is required to effectively calibrate the model, significantly improving prediction accuracy.

[0121] In summary, the present invention breaks through the bottleneck of traditional infrared target characteristic prediction technology. Through intelligent analysis and prediction driven by a large language model, it greatly improves the efficiency, accuracy and adaptability of infrared target characteristic prediction, and provides reliable technical support for the autonomous identification of space targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0122] Figure 1 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0123] The present invention will be further described below with reference to the accompanying drawings and examples.

[0124] like Figure 1 As shown, the present invention provides a method for rapid prediction of infrared target characteristics based on a large language model. By constructing a causal inference feature extraction framework, it integrates expert knowledge, intelligence reports, and other multi-source information to perform feature inversion on unknown target parameters. A cross-domain migration prediction model for target characteristics is established, enabling cross-domain generation of characteristic images. Ultimately, this system forms a rapid prediction system for target characteristics in complex spatial environments. This method effectively addresses issues such as missing target model parameters, dynamic evolution of characteristics across time, and significant differences in characteristics across multiple operating scenarios, significantly improving the efficiency and accuracy of infrared target characteristic prediction.

[0125] Step 1: Multi-source data acquisition and preprocessing;

[0126] Step 1-1: Source domain image data: Obtain known source domain images , such as infrared target data in plain areas, or visible light image data;

[0127] Step 1-2: Target scene knowledge: For the target scene that needs to be predicted, that is, the target domain, such as the spatial environment, collect relevant literature, reports and other information to extract expert knowledge ;

[0128] Steps 1-3: Building a domain knowledge base: integrating common sense of physical laws , expert knowledge and domain expertise , forming a comprehensive knowledge base ;

[0129] Steps 1-4: Source domain image and expert knowledge After data preprocessing, including image standardization, noise removal, geometric correction, etc., the preprocessed source domain image is obtained and structured knowledge representation .

[0130] Step 2: Causal feature extraction and knowledge representation based on the language model;

[0131] Step 2-1: Design of prompt words for the open source language model;

[0132] A structured causal reasoning prompt word engineering design is adopted, including the following components:

[0133] a) Context setting module ;

[0134] b) Causal Link Guidance Module ;

[0135] c) Structured output specification module ;

[0136] d) Uncertainty quantification module ;

[0137] e) Counterfactual Thinking Module ;

[0138] The complete prompt word template is constructed as follows:

[0139]

[0140] Step 2-2: Causal feature extraction of unstructured knowledge;

[0141] Define the causal feature extraction function:

[0142]

[0143] Step 2-3: Causal chain analysis and knowledge representation framework construction;

[0144] Based on the extracted causal features, a causal chain analysis model is constructed:

[0145]

[0146] By analyzing the structured JSON data output by the language model, a causal graph is automatically constructed; the edge weights are calculated based on the causal relationship strength and confidence given by the model;

[0147] Steps 2-4: Reasoning about target characteristics in a specific state;

[0148] Based on the constructed causal graph, the probability reasoning method is used to calculate the target in a specific state. Temperature values ​​of each part and radiation values :

[0149]

[0150] in, Indicates a specific state and causal diagrams Probability distribution of temperature and radiation values ​​under the conditions, Represents the inference function based on the language model;

[0151] The Monte Carlo sampling method is used to estimate the probability distribution of temperature and radiation values, and the confidence interval is calculated using the uncertainty range given by the language model;

[0152] Output the temperature value of each target part under specific conditions and radiation values ,in is the number of target sites.

[0153] Step 3: Unsupervised domain adaptive image semantic segmentation;

[0154] Step 3-1: Design the overall architecture of the segmentation model;

[0155] The segmentation model consists of the following four modules:

[0156] a) Feature Encoder : Located at the forefront of the segmentation model, it is responsible for extracting multi-scale features of the image; ResNet-50 is used as the backbone network and pre-trained and initialized; the feature encoder consists of 5 stages, each stage downsampling once to generate feature maps of different scales ;

[0157] b) Domain Adaptation Module : Located between the feature encoder and feature decoder, it consists of multiple domain adversarial discriminators corresponding to features at different levels; through adversarial training, it reduces the difference in feature distribution between the source domain and the target domain; the domain adaptation module performs the following mapping:

[0158]

[0159] c) Feature Decoder : Located in the second half of the segmentation model, it is responsible for mapping the features of the feature encoder and domain adaptation module to the segmentation results; the feature decoder adopts a cascade upsampling structure to fuse skip connection features of different scales:

[0160]

[0161] d) Split Head : Located at the end of the segmentation model, it maps the features output by the feature decoder to the final segmentation result:

[0162]

[0163] Step 3-2: Training phase;

[0164] The unsupervised domain adaptation strategy is adopted in the training phase, as follows:

[0165] a) Source domain supervised learning: Use preprocessed source domain images for supervised training and optimize the segmentation loss:

[0166]

[0167] b) Adversarial Domain Adaptation: Introducing Domain Discriminators , reducing inter-domain differences through adversarial training:

[0168]

[0169] c) Pseudo-label generation: Reasoning on the target domain image to generate high-confidence pseudo-labels:

[0170]

[0171] d) Self-training iteration: Use the generated pseudo labels to iteratively optimize the model:

[0172]

[0173] e) Overall optimization goal:

[0174]

[0175] Step 3-3: Testing phase;

[0176] a) Input the image to be segmented ;

[0177] b) Extract features through feature encoder: ;

[0178] c) Processing features through domain adaptation module: ;

[0179] d) Generate multi-scale features through feature decoder: ;

[0180] e) Generate the final segmentation result through the segmentation head: ;

[0181] f) Apply post-processing to refine the segmentation boundaries and obtain the final segmentation mask:

[0182]

[0183] Among them, the final segmentation mask Include semantic regions , each region corresponds to a component of the target; represents the post-processing optimization function;

[0184] Step 4: Assign infrared features and generate gradients based on the language model;

[0185] Step 4-1: Assign regional infrared characteristics;

[0186] The temperature value obtained in step 2 and radiation values , assign infrared features to each semantic area segmented in step 3:

[0187]

[0188] Step 4-2: Generate temperature gradient based on heat transfer model;

[0189] The conduction coefficient matrix between each component given by the language model , construct the temperature gradient formula:

[0190]

[0191] Step 4-3: Target domain adaptation;

[0192] According to the environmental parameters of the target domain , adaptively adjust the infrared characteristics:

[0193]

[0194] Step 5: Model calibration and accuracy optimization based on measured data;

[0195] Step 5-1: Construction of difference metric and loss function;

[0196] Defining Measured Characteristics and predictive features The difference measure between:

[0197]

[0198] Step 5-2: Parameter optimization and calibration;

[0199] Optimize the model parameters by minimizing the discrepancy metric:

[0200]

[0201] Step 5-3: Final infrared characteristics generation;

[0202] Generate the final infrared characteristics prediction results based on the optimized parameters :

[0203] .

[0204] Step 6: System integration and application deployment;

[0205] This step integrates the above steps into a complete infrared target characteristics rapid prediction system and deploys it for practical application.

[0206] 1) Process automation and interface design;

[0207] Design a unified data input interface and result output interface to realize the automated execution of steps 1 to 5.

[0208] 2) Construction of multi-scenario parameter configuration library;

[0209] Establish a parameter configuration library for different application scenarios to support fast switching and deployment.

[0210] 3) Real-time prediction and result visualization;

[0211] Realize real-time prediction and three-dimensional visualization of infrared characteristics, and support interactive analysis and applications.

[0212] Test results:

[0213] Table 1 Performance comparison of different methods in predicting infrared characteristics of space scene targets

[0214] method Structural Similarity (SSIM) Peak signal-to-noise ratio (PSNR / dB) Mean absolute error (MAE / °C) Computation time (s) Method of the present invention 0.89 35.7 2.3 18 Traditional physical modeling 0.74 29.3 5.8 432 Deep Learning Image Conversion Method 0.81 32.5 3.7 46 GAN based on domain adaptation 0.83 33.1 3.2 87

[0215] It can be seen from the experimental results that the method of the present invention is superior to the existing methods in key indicators such as structural similarity, peak signal-to-noise ratio and mean absolute error, while the calculation time is greatly reduced, which fully proves the effectiveness and advancement of the method.

Claims

1. A method for rapid prediction of spatial infrared target characteristics based on a large language model, characterized in that: The steps include: Step 1: Multi-source data acquisition and preprocessing; Step 2: Causal feature extraction and knowledge representation based on the language model; Step 2-1: Design of prompt words for the open source language model; A structured causal reasoning prompt word engineering design is adopted, including the following components: a) Context setting module ; b) Causal Link Guidance Module ; c) Structured output specification module ; d) Uncertainty quantification module ; e) Counterfactual Thinking Module ; The complete prompt word template is constructed as follows: ； in, Indicates a complete prompt word template; Step 2-2: Causal feature extraction of unstructured knowledge; Define the causal feature extraction function: ； in, represents the set of extracted causal features, A language model function representing feature extraction; Step 2-3: Causal chain analysis and knowledge representation framework construction; Based on the extracted causal features, a causal chain analysis model is constructed: ； in, represents a causal diagram, Indicates that the node set corresponds to each part or characteristic attribute of the target. represents the causal relationship between edge sets, i.e. parts, Represents the weight matrix, i.e., the strength of causal relationship; By analyzing the structured JSON data output by the language model, a causal graph is automatically constructed; the edge weights are calculated based on the causal relationship strength and confidence given by the model; Steps 2-4: Reasoning about target characteristics in a specific state; Based on the constructed causal graph, the probability reasoning method is used to calculate the target in a specific state. Temperature values ​​of each part and radiation values : ； in, Indicates a specific state and causal diagrams Probability distribution of temperature and radiation values ​​under the conditions, Represents the inference function based on the language model; The Monte Carlo sampling method is used to estimate the probability distribution of temperature and radiation values, and the confidence interval is calculated using the uncertainty range given by the language model; Output the temperature value of each target part under specific conditions and radiation values ,in is the number of target sites; Step 3: Unsupervised domain adaptive image semantic segmentation; Step 4: Assign infrared features and generate gradients based on the language model; Step 4-1: Assign regional infrared characteristics; The temperature value obtained in step 2 and radiation values , assign infrared features to each semantic area segmented in step 3; Step 4-2: Generate temperature gradient based on heat transfer model; The temperature gradient formula is constructed using the conduction coefficient matrix between components given by the language model; Step 4-3: Target domain adaptation; Adaptively adjust infrared characteristics according to the environmental parameters of the target domain; Step 5: Model calibration and accuracy optimization based on measured data.

2. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 1 is characterized in that: The step 1 is specifically as follows: Step 1-1: Source domain image data: Obtain known source domain images ; Step 1-2: Target scenario knowledge: Extract expert knowledge for the target scenario that needs to be predicted, i.e. the target domain ; Steps 1-3: Building a domain knowledge base: integrating common sense of physical laws , expert knowledge and domain expertise , forming a comprehensive knowledge base ; Steps 1-4: Source domain image and expert knowledge After data preprocessing, the preprocessed source domain image is obtained and structured knowledge representation .

3. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 2 is characterized in that: The source domain image It is infrared target data or visible light image data.

4. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 2, characterized in that: The target scene includes space environment, relevant documents, and report information.

5. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 2 is characterized in that: The data preprocessing includes image standardization, noise removal, and geometric correction.

6. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 5, characterized in that: The step 3 is specifically as follows: Step 3-1: Design the overall architecture of the segmentation model; The segmentation model consists of the following four modules composition: a) Feature Encoder : Located at the front end of the segmentation model, responsible for extracting multi-scale features of the image; ResNet-50 is used as the backbone network and pre-trained and initialized; the feature encoder consists of 5 stages, each stage downsampling once to generate feature maps of different scales ; b) Domain Adaptation Module : Located between the feature encoder and feature decoder, it consists of multiple domain adversarial discriminators corresponding to features at different levels; through adversarial training, it reduces the difference in feature distribution between the source domain and the target domain; the domain adaptation module performs the following mapping: ； in, Output features for the last layer of the encoder, is the domain discriminator, is the feature after domain adaptation; represents the domain adaptation transformation function; c) Feature Decoder : Located in the second half of the segmentation model, it is responsible for mapping the features of the feature encoder and domain adaptation module to the segmentation results; the feature decoder adopts a cascade upsampling structure to fuse skip connection features of different scales: ； in, Indicates the decoder Layer output features, Indicates the upsampling result of the previous layer feature, represents the skip connection feature of the corresponding encoder layer, represents the feature decoding transformation function; d) Split Head : Located at the end of the segmentation model, it maps the features output by the feature decoder to the final segmentation result: ； in, represents the output segmentation mask, Represents the output features of the last layer of the feature decoder, Represents the feature mapping function of the segmentation head; Step 3-2: Training phase; The unsupervised domain adaptation strategy is adopted in the training phase, as follows: a) Source domain supervised learning: Use preprocessed source domain images for supervised training and optimize the segmentation loss: ； in, represents the cross entropy loss function, represents the segmentation label of the source domain image, represents the feature encoding transformation function; b) Adversarial Domain Adaptation: Introducing Domain Discriminators , reducing inter-domain differences through adversarial training: ； in, represents the target domain image, represents the natural logarithm function, represents the expected operation on the target domain image, represents the expected operation on the source domain image, represents the adversarial loss function; c) Pseudo-label generation: Reasoning on the target domain image to generate high-confidence pseudo-labels: ； in, represents the generated pseudo-label, represents the confidence threshold, represents a threshold-based binarization function; d) Self-training iteration: Use the generated pseudo labels to iteratively optimize the model: ； in, represents the self-training loss function; e) Overall optimization goal: ； in, represents the regularization term, , , All are weight coefficients; Step 3-3: Testing phase; a) Input the image to be segmented ; b) Extract features through feature encoder: ; c) Processing features through domain adaptation module: ; d) Generate multi-scale features through feature decoder: ; e) Generate the final segmentation result through the segmentation head: ; f) Apply post-processing to refine the segmentation boundaries and obtain the final segmentation mask: ； Among them, the final segmentation mask Include semantic regions , each region corresponds to a component of the target; Represents a post-processing optimization function.

7. The method for rapid prediction of spatial infrared target characteristics based on a language large model according to claim 6 is characterized in that: The step 4 is specifically as follows: Step 4-1: Assign regional infrared characteristics; The temperature value obtained in step 2 and radiation values , assign infrared features to each semantic area segmented in step 3: ； in, Represents pixel points The initial infrared characteristic value at Represents pixel points Belong to the region probability; Represent the temperature and radiation values ​​of the i-th region respectively; Step 4-2: Generate temperature gradient based on heat transfer model; The conduction coefficient matrix between each component given by the language model , construct the temperature gradient formula: ； in, Indicates the infrared characteristic value after considering the gradient, Indicates the The influence coefficient of the region, represents the attenuation parameter, Indicates a point To area The shortest distance to the border; Step 4-3: Target domain adaptation; According to the environmental parameters of the target domain , adaptively adjust the infrared characteristics: ； in, represents the infrared characteristic value after adapting to the target domain, represents the global adjustment parameters, Represents the influence function of environmental parameters.

8. The method for rapid prediction of space infrared target characteristics based on a language large model according to claim 7 is characterized in that: The step 5 is specifically as follows: Step 5-1: Construction of difference metric and loss function; Defining Measured Characteristics and predictive features The difference measure between: ； in, represents the mean square error, represents the structural similarity index, and represents the weight coefficient, represents the difference metric loss function; Step 5-2: Parameter optimization and calibration; Optimize the model parameters by minimizing the discrepancy metric: ； in, represents the optimized parameters; Step 5-3: Final infrared characteristics generation; Generate the final infrared characteristics prediction results based on the optimized parameters : 。

Images