Method for evaluating environmental adaptability of imaging system based on infrared simulation

Abstract

The present application relates to a kind of based on infrared simulation imaging system environment adaptability evaluation method, first, using high-dimensional Gaussian distribution model to model environment parameter space, based on the model generates environment sample set;Second, design evaluation index to evaluate the quality of environment sample set;Then, using infrared simulation system based on environment sample generates corresponding simulation image, adopts TTP model to estimate the imaging quality of simulation image, simulation sample is marked to generate the training sample of environment adaptability evaluation model;Finally, using training sample set to train Levenberg-Marquardt backpropagation algorithm neural network obtains environment adaptability evaluation model;Design experiment, collect infrared image under outdoor environment to carry out model verification.The present application is by using infrared visual simulation system, increase the quantity of experimental data, solve the problem that outdoor test cost is high, environment condition is single, increase the credibility of environment adaptability evaluation result.

Description

Technical Field

[0001] This invention relates to the field of environmental adaptability assessment technology for imaging systems, and more specifically to an environmental adaptability assessment method for imaging systems based on infrared simulation. Background Technology

[0002] In the daily operations of unmanned ground-based photoelectric detection systems, meteorological conditions determine whether they can function properly. Therefore, conducting environmental adaptability assessment research on unmanned ground-based photoelectric detection systems has significant application value.

[0003] In recent years, scholars both domestically and internationally have conducted relevant research on environmental adaptability assessment. Existing technologies, based on a large amount of environmental factor and equipment failure data obtained from experiments, and combined with the requirements that environmental adaptability should meet, have established assessment models using the direct evaluation method. Alternatively, by analyzing the fuzziness between assessment index factors and judgment states, comprehensive fuzzy assessment models have been established. However, in the process of constructing the assessment model, the grey relationship between the judgment state and the index factors is not considered, resulting in a lack of reliability in the assessment index values. Another approach is to rationally integrate subjective and objective information to assign weights to the assessment indicators, establishing a self-learning model of weight coefficients based on Bayesian estimation to achieve adaptive adjustment of subjective and objective weight coefficients in the combined weighting method. By analyzing the fuzzy relationship between assessment index factors and judgment states, membership functions are constructed to achieve assessment index values ​​under uncertain conditions. After determining the information grey level by integrating expert opinions, a self-learning fuzzy grey level model for assessing the adaptability of air defense early warning radar to the plateau environment is established. However, this method, when experimental data is limited, mainly relies on expert experience for assessment, increasing the risk of subjective judgment and making the assessment results lack objectivity.

[0004] Currently, environmental adaptability assessments mainly rely on outdoor testing, which is challenging for unmanned ground-based photoelectric imaging systems. To address these needs, this invention proposes an environmental adaptability assessment method based on an infrared simulation system. Summary of the Invention

[0005] This invention addresses the environmental adaptability of unmanned ground imaging systems. Given the significant challenges of outdoor testing for unmanned ground optoelectronic imaging systems and the need for substantial experimental data, this invention proposes an evaluation framework based on an infrared simulation system. This framework generates simulation experimental data by combining an infrared visual simulation system with spatial modeling of environmental parameters. Furthermore, it learns the nonlinear relationship between environmental adaptability and environmental parameters through a neural network, and finally derives the confidence level of the evaluation model through fuzzy band analysis.

[0006] The technical solution adopted by this invention to achieve the above objectives is: an environmental adaptability assessment method for imaging systems based on infrared simulation, comprising the following steps:

[0007] Historical meteorological data of the target area are obtained and fitted with a high-dimensional Gaussian distribution. Then, the central limit theorem is used to generate an environmental sample set with the same distribution.

[0008] Determine environmental sample set quality assessment indicators and comprehensively evaluate the quality of environmental sample sets;

[0009] An infrared simulation system is used to generate corresponding simulation images based on environmental samples. The TTP model is used to estimate the imaging quality of the simulation images. The simulation samples are labeled to generate training samples for the environmental adaptability assessment model.

[0010] The neural network is trained using a training sample set to obtain an environmental adaptability assessment model;

[0011] Infrared images of outdoor scenes were collected for model verification.

[0012] The process involves fitting a high-dimensional Gaussian distribution and then using the central limit theorem to generate an identically distributed environmental sample set. Specifically, this involves performing high-dimensional Gaussian analysis on historical meteorological data to obtain its mean and covariance matrix, and then using the central limit theorem to generate an identically distributed environmental sample set. This includes the following steps:

[0013] x1, x2, ..., x n It follows a mean μ and a variance σ. 2 Let a certain distribution of ,

[0014]

[0015] Where x1, x2, ..., x n It consists of n meteorological data points in a certain dimension, where the dimension represents the type of meteorological data; ξ indicates that it follows a normal distribution.

[0016] According to the central limit theorem, n multidimensional independent and identically distributed uniform distributions are generated, forming an environmental sample set.

[0017] The process of determining environmental sample set quality assessment indicators and comprehensively evaluating the quality of the environmental sample set includes the following steps:

[0018] Determine whether the environmental sample set passes the quality assessment of homogeneity, orthogonality, distribution consistency, and typicality; if the environmental sample set passes all indicators, proceed to the next step.

[0019] Otherwise, regenerate the environment sample set.

[0020] The determination of whether the environmental sample set passes the homogeneity quality assessment is achieved by the following formula:

[0021]

[0022] Where m is the dimension of the environmental sample set, n is the number of environmental samples in the environmental sample set, and the environmental sample set P is an n*m matrix; MD 2 x is a homogeneity quality assessment index, representing the mixing deviation; ij This represents the sample in the i-th row and j-th column of the environmental sample set;

[0023] If MD 2 If (P) exceeds the threshold, it means that the homogeneity quality assessment has been passed; otherwise, it means that the homogeneity quality assessment has not been passed and the environmental sample set needs to be regenerated.

[0024] The determination of whether the environmental sample set passes the orthogonal quality assessment is achieved by the following formula:

[0025]

[0026] Where, ρ ij ρ represents the linear correlation coefficient between any two columns of the simulation sample set matrix; m represents the dimension of the environmental sample set; ρ is the correlation coefficient.

[0027] Orthogonality quality assessment index ρ 2 If the result is less than the threshold, it means the orthogonality quality assessment has been passed; otherwise, it means the orthogonality quality assessment has not been passed and the environmental sample set needs to be regenerated.

[0028] The determination of whether the environmental sample set passes the distribution consistency quality assessment is achieved by the following formula:

[0029]

[0030] Where σ1 and σ2 are the covariance matrices of historical meteorological data and environmental sample set, respectively; μ1 and μ2 are the mean matrices of historical meteorological data and test samples, respectively; m is the dimension of environmental samples; KL is the distribution consistency quality assessment index, which represents the KL divergence used to measure the consistency of two high-dimensional Gaussian distributions; tr represents the trace of the matrix.

[0031] If KL is greater than the threshold, it means that the distribution consistency quality assessment has been passed; otherwise, it means that the distribution consistency quality assessment has not been passed and the environmental sample set needs to be regenerated.

[0032] The determination of whether the environmental sample set passes the typicality quality assessment is achieved by the following formula:

[0033] ty=N c / n

[0034] Where n is the number of environmental samples in the environmental sample set, N c The number of preset typical points included in the simulation test sample; ty is the typicality evaluation index;

[0035] If ty is greater than the threshold, it means that the typicality quality assessment has been passed; otherwise, it means that the typicality quality assessment has not been passed and the environmental sample set needs to be regenerated.

[0036] The process of generating corresponding simulated images based on environmental samples using an infrared simulation system, estimating the imaging quality of the simulated images using a TTP model, and labeling the simulated samples to generate training samples for the environmental adaptability assessment model includes the following steps:

[0037] By comparing the environmental adaptability deviations of N sets of real-world and simulated scenarios, the fuzzy band parameters are obtained:

[0038]

[0039] Among them, T i The image quality measurement value for a real scene, t i These are the corresponding simulated scene imaging quality measurement values;

[0040] If the imaging quality of a certain simulated scene is t, then the corresponding real scene imaging quality is within the blur zone: [t-δ, t+δ].

[0041] When using the TTP model for sample labeling, the measured image quality t and the left and right boundaries of the image quality t-δ and t+δ are used as labels.

[0042] The process of training a neural network using a training sample set to obtain an environmental adaptability assessment model includes the following steps:

[0043] Using the training set of environmental samples as input and the imaging quality t as output, a standard model is obtained by training a neural network.

[0044] The training set of the environmental sample set is used as input, and the left interval t-δ of the fuzzy band is used as output. The left boundary model is obtained by training a neural network.

[0045] The training set of the environmental sample set is used as input, and the right interval t+δ of the fuzzy band is used as output. The right boundary model is obtained by training a neural network.

[0046] The verification and evaluation of the model's reliability under outdoor conditions includes the following steps:

[0047] 1) Input the test set of the environmental sample set into the three models respectively, and the model output results are the imaging quality;

[0048] If all three results are greater than the threshold, the classification result is environmental adaptation, meaning that the imaging system can work normally under the current weather conditions.

[0049] If any one of the three results is less than the threshold, the classification result is environmental maladaptation, meaning that the imaging system cannot work properly under the current weather conditions.

[0050] Then, the classification results are compared with the labels. If they match, the classification is correct. The ratio of the number of correctly classified samples to the number of test set samples is the model confidence level p.

[0051] If the model confidence level p is greater than the set value, it indicates that the model is initially judged to be reliable; otherwise, the initial judgment is unreliable.

[0052] 2) Input the test set of the environmental sample set into the standard model, left boundary model, and right boundary model, and output three imaging qualities respectively. Select the minimum value 'a' and the maximum value 'b' to obtain an imaging quality range [a, b]. Set the imaging quality threshold as α:

[0053] If both a and b are greater than α, then this scenario is considered environmental adaptation.

[0054] If both a and b are less than α, then this scenario is considered an environment maladaptation.

[0055] If a is less than α, b is greater than α, and (b-α) > (α-a), then this scenario is considered environmental adaptation.

[0056] If a is less than α, b is greater than α and (b-α) < (α-a), then this scenario is considered an environment maladaptation.

[0057] Infrared images acquired outdoors were used to measure image quality using a TTP model to obtain environmental adaptability assessment results.

[0058] Compare the environmental adaptability results output by the model with those from the analysis and evaluation under the same conditions. If they are consistent, the model is reliable; otherwise, it is unreliable.

[0059] The present invention has the following beneficial effects and advantages:

[0060] 1. By utilizing an infrared visual simulation system, the amount of experimental data was increased, solving the problem of the difficulty of outdoor testing and increasing the objectivity of the evaluation model.

[0061] 2. Simulation data was generated by modeling environmental parameters and evaluated using simulation data for uniformity, orthogonality, distribution consistency, and typicality, thus verifying the usability of the simulation data.

[0062] 3. By utilizing neural networks to learn the nonlinear relationship between environmental parameters and environmental adaptability, an environmental adaptability assessment model can be modeled more accurately.

[0063] 4. By incorporating fuzzy band analysis into the credibility of the environmental adaptability assessment model, the credibility of the assessment model is described more comprehensively.

[0064] 5. Through verification in a small number of real-world scenarios, the method of this invention has shown good evaluation results in environmental adaptability assessment for various targets. Attached Figure Description

[0065] Figure 1 This is an overall flowchart of the present invention;

[0066] Figure 2 This is a schematic diagram illustrating the comprehensive evaluation of the simulation samples of the present invention;

[0067] Figure 3 This is a schematic diagram illustrating the uniformity of the present invention. Detailed Implementation

[0068] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0069] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0070] The method for evaluating the environmental adaptability of imaging systems based on infrared simulation includes the following steps:

[0071] A high-dimensional Gaussian distribution model is used to model the environmental parameter space, and an environmental sample set is generated based on this model.

[0072] Design evaluation indicators to assess the quality of the environmental sample set;

[0073] An infrared simulation system is used to generate corresponding simulation images based on environmental samples. The TTP model is used to estimate the imaging quality of the simulation images. The simulation samples are labeled to generate training samples for the environmental adaptability assessment model.

[0074] The Levenberg-Marquardt backpropagation algorithm neural network was trained using the training sample set to obtain an environmental adaptability assessment model.

[0075] Design an experiment to collect outdoor infrared images for model verification.

[0076] The environmental parameter space modeling is achieved by generating identically distributed simulation samples using the central limit theorem. That is:

[0077] Let x1, x2, ..., x n It follows a pattern with mean μ and variance σ. 2 Let a certain distribution of ,

[0078]

[0079] Where x1, x2, ..., x nIt refers to n data points in a certain dimension, such as n temperature values ​​or n pressure values, where n refers to n sets of data.

[0080] When n is sufficiently large, ξ asymptotically follows a normal distribution, meaning that the distribution among n independent and identically distributed random variables approximates a normal distribution; the larger n is, the better the approximation. According to the Central Limit Theorem, generating a high-dimensional Gaussian distribution can be achieved by generating n high-dimensional independent and identically distributed uniform distributions.

[0081] The environmental sample set evaluation indicators are: uniformity, orthogonality, distribution consistency, and typicality. Among them:

[0082] Uniformity: The uniformity criterion (i.e. the deviation-based criterion) can not only evaluate the uniformity of sampling points in a multidimensional cube, but also guarantee the uniformity of their projection in a low dimension. Therefore, it is often used to measure the space-filling property of experimental tests.

[0083] The main deviation measurement methods include central L2 deviation, convoluted L2 deviation, and mixed deviation. Central L2 deviation focuses more on points near the vertex and has lower sensitivity to points in the central region; convoluted L2 deviation remains unchanged after various factor translation transformations; while mixed deviation can effectively overcome the shortcomings of both, therefore, mixed deviation is used to measure the space-filling property in experiments. Its formula is:

[0084]

[0085] m represents the dimension, and n is the number. In this paper, m = 5, meaning five dimensions, and n is the number of environmental samples. The dataset is an n*m matrix, and larger indicator values ​​are better.

[0086] Orthogonality: Experimental design often requires the experimental design matrix to have the property of being "neatly comparable," meaning that the orthogonality of the experimental design matrix is ​​required. The higher the degree of orthogonality, the lower the correlation between the test samples, and the closer the distribution of the test samples is to being "neatly comparable." A commonly used metric for orthogonality is the correlation coefficient ρ between the experimental factors.

[0087]

[0088] Where, ρ ij This represents the linear correlation coefficient between two columns of the simulation sample set matrix. m represents the number of columns, i.e., the dimension of the environmental parameters. A smaller value is better.

[0089] Distribution Consistency: Real-world system factors exhibit certain distribution patterns under certain circumstances. Therefore, the quality of an experimental design can be analyzed by measuring the consistency between the distribution of factors in the test sample and the actual system. When the distribution of factors in the actual system is known, hypothesis testing can be performed on the point factors in the experimental test to determine whether the distribution of factors in the simulated test sample and the actual system is consistent. A higher degree of consistency indicates that the selection of test samples is more in line with reality, and the experimental results are more reliable. Historical meteorological data conforms to a high-dimensional Gaussian distribution, and the selected environmental sample set also conforms to a high-dimensional Gaussian distribution. KL divergence is used to measure the consistency between the two high-dimensional Gaussian distributions. A smaller KL divergence indicates higher consistency. The KL divergence formula is:

[0090]

[0091] Where σ1 and σ2 are the covariance matrices of the real meteorological data and the generated environmental sample set, respectively; μ1 and μ2 are the mean matrices of the real meteorological data and the test samples, respectively; and m is the dimension of the test samples. A larger index value is better.

[0092] Typicality: The typicality of the simulation sample is measured by the degree to which the simulation test sample covers the preset typical points.

[0093] ty=N c / n

[0094] Where n is the number of test samples, N c The number of preset typical points included in the simulation test sample.

[0095] The bigger the better

[0096] Due to certain deviations between the simulation system and the real scene in the process of labeling the simulated samples, a fuzzy band analysis method was proposed and incorporated into the experimental analysis to address this issue.

[0097] By comparing the environmental adaptability deviations of N sets of real-world and simulated scenarios, the fuzzy band parameters are obtained:

[0098]

[0099] Among them, T i The image quality measurement value for a real scene, t i This represents the measured image quality value for the corresponding simulated scene. Therefore, if the image quality of a simulated scene is t, then the corresponding real scene image quality falls within the fuzzy band interval: [t-δ, t+δ]. When using the TTP model for sample labeling, it is necessary not only to label the measured image quality t, but also to use the fuzzy band parameters to obtain the left and right boundaries of the image quality, t-δ and t+δ.

[0100] The Levenberg-Marquardt backpropagation algorithm, compared to the fastest gradient descent method of the traditional backpropagation algorithm, has a faster convergence speed and better convergence performance. This invention uses the Levenberg-Marquardt backpropagation algorithm for training.

[0101] The environmental adaptability assessment model is as follows: After fuzzy band analysis, this invention uses a neural network to train three prediction models: a standard model, a left boundary model, and a right boundary model. The standard model is the classifier for the simulation system under error-free conditions, the left boundary model is the classifier for the simulation system under maximum negative error conditions, and the right boundary model is the classifier for the simulation system under maximum positive error conditions. Integrating these three models forms the environmental assessment model of this invention. The reliability of the environmental assessment model is verified using results: The test set is input into each of the three models, and the model outputs results (image quality). If all three results are greater than a threshold, the classification result is environmental adaptability, meaning the imaging system can function normally under the current weather conditions. If any of the three results is less than the threshold, the classification result is environmental inadaptability, meaning the imaging system cannot function normally under the current weather conditions. Then, the classification results are compared with the labels; the ratio of the number of correctly classified results to the total number of samples is the model reliability p.

[0102] The experimental verification involves evaluating and comparing the results using both the environmental adaptability assessment model and real-world experiments in typical real-world scenarios.

[0103] An infrared simulation-based method for evaluating the environmental adaptability of imaging systems includes the following steps:

[0104] Step 1: Obtain historical meteorological data of the target area, fit it with a high-dimensional Gaussian distribution model, and then use the central limit theorem to generate an environmental sample set with the same distribution;

[0105] Step 2: Design environmental sample set quality assessment indicators and comprehensively evaluate the quality of the environmental sample set generated in Step 1;

[0106] Step 3: Use the infrared simulation system to generate corresponding simulation images based on environmental samples, use the TTP model to estimate the imaging quality of the simulation images, and label the simulation samples to generate training samples for the environmental adaptability assessment model.

[0107] Step 4: Train the Levenberg-Marquardt backpropagation algorithm neural network using the training sample set to obtain the environmental adaptability assessment model;

[0108] Step 5: Design an experiment to collect outdoor infrared images for model verification.

[0109] The generation of the environmental sample set includes the following process:

[0110] First, high-dimensional Gaussian analysis is performed on historical meteorological data to obtain its mean and covariance matrix. Then, the central limit theorem is used to generate identically distributed simulation samples.

[0111] The environmental sample set assessment includes the following process:

[0112] The homogeneity criterion (i.e. the deviation-based criterion) can not only evaluate the homogeneity of sampling points in a multidimensional cube, but also guarantee the homogeneity of their projection in a lower dimension. Therefore, it is often used to measure the space-filling property of experimental tests.

[0113] Based on the orthogonality criterion: Experimental test design often requires the experimental design matrix to have the property of "uniform comparability," that is, the orthogonality of the experimental design matrix is ​​required. The higher the degree of orthogonality, the lower the correlation between test samples, and the closer the distribution of test samples is to "uniform comparability." A commonly used measure of orthogonality is the correlation coefficient ρ between experimental factors.

[0114] Based on the distribution consistency criterion: Real-world system factors exhibit certain distribution patterns under certain circumstances. Therefore, the quality of an experimental design can be analyzed by measuring the consistency between the distribution of factors in the test sample and the actual system. When the distribution of factors in the actual system is known, hypothesis testing can be performed on the point factors in the experimental test to determine whether the distribution of factors in the simulated test sample and the actual system is consistent. The better the consistency, the more realistic the selection of the test sample, and the more reliable the experimental results.

[0115] Based on the typicality criterion: the typicality of the simulation sample is measured by the degree to which the simulation test sample covers the preset typical points.

[0116] The outdoor validation and reliability assessment of the model includes the following processes:

[0117] Inputting the test sample into three models will output three image qualities, resulting in an image quality range [a, b]. This range represents the estimated blur band range. The image quality threshold is set to α.

[0118] If both a and b are greater than α, then this scenario is considered environmental adaptation.

[0119] If both a and b are less than α, then this scenario is considered an environment maladaptation.

[0120] If a is less than α, b is greater than α, and (b-α) > (α-a), then this scenario is considered environmental adaptation.

[0121] If a is less than α, b is greater than α, and (b-α) < (α-a), then this scenario is considered an example of environmental maladaptation.

[0122] Finally, it was verified whether the environmental adaptability assessment results were consistent with those actually measured under the same conditions (multiple dimensions: temperature, air pressure, humidity, wind speed, visibility, etc.).

[0123] like Figure 1 As shown, the environmental adaptability assessment and reliability analysis method based on an infrared simulation system of the present invention includes the following steps:

[0124] (1): Obtain historical meteorological data of the target area, fit a high-dimensional Gaussian distribution model, and then use the central limit theorem to generate an environmental sample set with the same distribution.

[0125] (2): Design environmental sample set quality assessment indicators and comprehensively evaluate the quality of the environmental sample set generated in step 1;

[0126] (3) The infrared simulation system generates corresponding simulation images based on environmental samples, the TTP model is used to estimate the imaging quality of the simulation images, and the simulation samples are labeled to generate training samples for the environmental adaptability assessment model.

[0127] (4) The Levenberg-Marquardt backpropagation algorithm neural network was trained using the training sample set to obtain an environmental adaptability assessment model;

[0128] (5) Design an experiment to collect outdoor infrared images for model verification.

[0129] All experiments were conducted under summer weather conditions in Shenyang, generating a total of 2646 simulated samples. Figure 2 The composition of the comprehensive evaluation indicators is described. Figure 3 The principle of uniformity is described. The evaluation of the simulation samples is shown in Table 1:

[0130] Table 1 Evaluation Data

[0131]

[0132] As can be seen from the indicators in the table above, the meteorological conditions corresponding to the selected simulation samples are evenly distributed, the correlation between meteorological conditions is low, and the distribution is basically consistent with the real samples. The proportion of typical scenarios is high, indicating that the quality of the generated simulation samples can be used for the following research.

[0133] By comparing the environmental adaptability deviations of N sets of real and simulated scenarios, the fuzzy band parameter was obtained, and the fuzzy band error δ was calculated and set to 0.05.

[0134] Under the conditions in Table 2:

[0135] Table 2. A set of meteorological data

[0136]

[0137] The image quality under these conditions is: 0.668418

[0138] After adding the error range, the left boundary of the image quality is 0.618418, and the right boundary is 0.718418.

[0139] The 2646 samples were divided into a training set and a test set in a 7:3 ratio. The model was trained on the training set and tested on the test set to determine its reliability.

[0140] The model confidence score was 0.9733 after testing on 793 test sets.

[0141] Verification was performed under the outdoor conditions described in Table 3:

[0142] Table 3 Outdoor test meteorological conditions

[0143]

[0144] The environmental assessment model output is adaptive, and the actual measured results are also adaptive, proving the effectiveness of the method of this invention.

Claims

1. A method for evaluating the environmental adaptability of an imaging system based on infrared simulation, characterized in that, The method comprises the following steps: Obtaining historical meteorological data of a target area to perform high-dimensional Gaussian distribution fitting, and then generating an environment sample set with the same distribution by using the central limit theorem; Determining an environment sample set quality evaluation index and comprehensively evaluating the quality of the environment sample set; Generating corresponding simulation images based on the environment sample set by using an infrared simulation system, estimating the imaging quality of the simulation images by using a TTP model, marking the simulation samples to generate training samples of the environment adaptability evaluation model; Training a neural network by using the training sample set to obtain the environment adaptability evaluation model; Collecting infrared images in an outdoor scene to verify the model; The method of determining the environment sample set quality evaluation index and comprehensively evaluating the quality of the environment sample set comprises the following steps: Judging whether the environment sample set passes the quality evaluation of uniformity, orthogonality, distribution consistency and typicality; if the environment sample set passes all the indexes, the next step is performed; Otherwise, the environment sample set is regenerated; The judgment of whether the environment sample set passes the quality evaluation of uniformity is implemented by the following formula: ； Wherein, m is the dimension of the environment sample set, n is the number of environment samples in the environment sample set, the environment sample set P is an n m matrix; is a uniformity quality evaluation index, indicating the mixed deviation; Indicates the i-th row and j-th column sample of the environment sample set. If Exceeding the threshold value indicates passing the uniformity quality assessment; otherwise, failing the uniformity quality assessment, requiring re-generating the ambient sample set. The judgment of whether the environment sample set passes the quality evaluation of orthogonality is implemented by the following formula: ； wherein, represents a linear correlation coefficient between any two columns of the simulated sample set matrix; represents the environmental sample set dimension; is a correlation coefficient; Orthogonality quality evaluation index Less than the threshold value, indicating that the environmental sample set passes the orthogonality quality evaluation; otherwise, indicating that the environmental sample set fails the orthogonality quality evaluation and needs to be regenerated. The judgment of whether the environment sample set passes the quality evaluation of distribution consistency is implemented by the following formula: ； wherein, are the historical meteorological data and environmental sample set covariance matrices, respectively, are the mean matrices of the historical meteorological data and test samples, respectively, is the environmental sample set dimension; is the distribution consistency quality evaluation index, representing the KL divergence used to measure the consistency of two high-dimensional Gaussian distributions; denotes the trace of a matrix; greater than the threshold value, it indicates that the distribution consistency quality assessment passes; otherwise, it indicates that the distribution consistency quality assessment fails, and the environment sample set needs to be regenerated; The judgment of whether the environment sample set passes the quality evaluation of typicality is implemented by the following formula: ； wherein, is the number of environment samples in the environment sample set, is the number of preset typical points contained in the simulation test sample; is the typicality evaluation index; greater than the threshold value, it indicates passing the typicality quality assessment; otherwise, it indicates failing the typicality quality assessment and the environmental sample set needs to be regenerated.

2. The method of claim 1, wherein, Performing high-dimensional Gaussian distribution fitting, and then generating an environment sample set with the same distribution by using the central limit theorem, specifically comprising the following steps: subordinate to the mean , variance of a certain distribution, let ； wherein, is n weather data of a certain dimension, the dimension representing the weather data type; denotes a normal distribution; According to the Central Limit Theorem, generate a multidimensional set of independent and identically distributed uniform distributions, constituting the environmental sample set.

3. The method of claim 1, wherein the infrared-simulation-based imaging system environment suitability assessment is performed in real-time. Generating corresponding simulation images based on the environment sample set by using an infrared simulation system, estimating the imaging quality of the simulation images by using a TTP model, marking the simulation samples to generate training samples of the environment adaptability evaluation model, comprising the following steps: Obtaining the fuzzy band parameters by comparing the environment adaptability deviations of N groups of real scenes and simulation scenes: ； wherein is an imaging quality measure value for the real scene, is a corresponding imaging quality measure value for the simulated scene; If the imaging quality of a simulation scene is , then the imaging quality of the corresponding real scene is within the blur band interval: ; When using the TTP model for sample labeling, the imaging quality measured by the label and the left and right boundaries of the imaging quality as a tag.

4. The method of claim 1, wherein the infrared-simulation-based imaging system environment suitability assessment is performed in real-time. Training a neural network by using the training sample set to obtain the environment adaptability evaluation model, comprising the following steps: Taking the training set of the environment sample set as input and the imaging quality t as output, and training a standard model by using a neural network; Taking the training set of the environmental sample set as input, the fuzzy band left interval Taking the training set of the environmental sample set as input, the fuzzy band left interval Taking the training set of the environmental sample set as input, the fuzzy band left interval Taking the training set of the environmental sample set as input, the The training set of the environmental sample set is taken as input, and the fuzzy band right interval The right boundary model is obtained by using neural network training for output.

5. The method of claim 1, wherein the infrared-simulation-based imaging system environment suitability assessment is performed in real-time. Verifying the reliability of the evaluation model in an outdoor scene, comprising the following steps: 1) inputting the test set of the environment sample set into three models respectively, and outputting the imaging quality from the models; If all the three results are greater than a threshold value, the classification result is environment adaptation, i.e., the imaging system can work normally under the current meteorological condition; If any one of the three results is less than the threshold value, the classification result is environment inadaptation, i.e., the imaging system cannot work normally under the current meteorological condition; Then the classification results are compared with the labels, and if they are consistent, the classification is correct; the ratio of the number of correct classifications to the number of test set samples is the model reliability ; If the model's credibility If the value is greater than the set value, it indicates that the preliminary judgment of the evaluation model is reliable; otherwise, the preliminary judgment is unreliable. 2) input the test set of the environmental sample set into the standard model, the left boundary model and the right boundary model, respectively output three imaging qualities, select the minimum value a and the maximum value b, and obtain an imaging quality range , set the imaging quality threshold value as : If are all greater than at this time, the scene is determined to be environment adaptation; If are all less than at this time, the scene is determined to be not suitable for the environment. if less than , greater than and ; at this time, the scene is judged as environment adaptation; if less than , greater than and ; at this time, the scene is judged as environment inadaptation; Obtaining the environment adaptability evaluation result by measuring the imaging quality of the infrared images collected outdoors by using the TTP model, Comparing the environment adaptability result obtained by analyzing and evaluating the model with the environment adaptability result under the same condition, if they are consistent, the model is reliable; otherwise, the model is unreliable.

Images