CN116305705A - Imaging system environment adaptability assessment method based on infrared simulation - Google Patents
Imaging system environment adaptability assessment method based on infrared simulation Download PDFInfo
- Publication number
- CN116305705A CN116305705A CN202111558197.7A CN202111558197A CN116305705A CN 116305705 A CN116305705 A CN 116305705A CN 202111558197 A CN202111558197 A CN 202111558197A CN 116305705 A CN116305705 A CN 116305705A
- Authority
- CN
- China
- Prior art keywords
- environmental
- sample set
- environment
- quality
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an imaging system environment adaptability assessment method based on infrared simulation, which comprises the steps of firstly, modeling an environment parameter space by using a high-dimensional Gaussian distribution model, and generating an environment sample set based on the model; secondly, designing an evaluation index to evaluate the quality of the environment sample set; then, generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by adopting a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model; finally training the Levenberg-Marquardt backpropagation algorithm neural network by using a training sample set to obtain an environmental adaptability assessment model; and (3) designing an experiment, and collecting an infrared image in an outdoor environment for model verification. According to the invention, the infrared visual simulation system is utilized, so that the number of experimental data is increased, the problems of high outdoor test cost and single environmental condition are solved, and the credibility of the environmental adaptability evaluation result is increased.
Description
Technical Field
The invention relates to the technical field of environmental suitability evaluation of imaging systems, in particular to an infrared simulation-based imaging system environmental suitability evaluation method.
Background
In the daily task of an unmanned ground photoelectric detection system, weather conditions determine whether the system can work normally. Therefore, the method for evaluating and researching the environmental suitability of the unmanned ground photoelectric detection system has obvious application value.
In recent years, students at home and abroad have conducted related studies on evaluation of environmental suitability. In the prior art, based on a large amount of environmental factors and equipment failure data obtained in experiments, an evaluation model is established based on a direct evaluation method by combining the requirements that the environmental adaptability should meet; or the comprehensive fuzzy evaluation model is established by analyzing the ambiguity of the evaluation index factors and the evaluation states, but the gray relation between the evaluation states and the index factors is not considered in the process of constructing the evaluation model, so that the evaluation index value lacks credibility. Or weighting the evaluation index by rationally fusing subjective and objective information, and establishing a weight coefficient self-learning model based on Bayesian estimation to realize the self-adaptive adjustment of the subjective and objective weight coefficients in the combined weighting method. And constructing a membership function by analyzing the fuzzy relation between the evaluation index factors and the evaluation states, and realizing evaluation index assignment under uncertain conditions. After the comprehensive expert opinion determines the information gray level, a self-learning fuzzy gray level model for the air defense early warning radar plateau environment adaptability evaluation is established. However, under the condition of less experimental data, the method mainly relies on expert experience to evaluate, increases the risk of subjective evaluation, and makes the evaluation result not objective.
At present, the evaluation of environmental suitability is mainly finished by an outdoor test, and the difficulty of the outdoor test of an unmanned ground photoelectric imaging system is high. In order to meet the above requirements, the invention provides an environment adaptability assessment based on an infrared simulation system.
Disclosure of Invention
Aiming at the environmental adaptability of the unmanned ground imaging system, the invention provides an evaluation framework based on an infrared simulation system, wherein the evaluation framework is used for generating simulation experiment data by combining an infrared visual simulation system with environmental parameter space modeling, learning the nonlinear relation between the environmental adaptability and the environmental parameter through a neural network, and finally obtaining the confidence coefficient of an evaluation model through a fuzzy band analysis method.
The technical scheme adopted by the invention for achieving the purpose is as follows: the imaging system environment adaptability evaluation method based on infrared simulation comprises the following steps:
acquiring historical meteorological data of a target area, performing high-dimensional Gaussian distribution fitting, and generating an environment sample set with the same distribution by using a central limit theorem;
determining an environmental sample set quality evaluation index, and comprehensively evaluating the environmental sample set quality;
generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model;
training the neural network by using a training sample set to obtain an environmental adaptability evaluation model;
and acquiring infrared images under outdoor scenes to perform model verification.
Fitting high-dimensional Gaussian distribution, and generating an environment sample set with the same distribution by using a central limit theorem, wherein the environment sample set is specifically as follows: carrying out high-dimensional Gaussian analysis on historical meteorological data to obtain a mean value and a covariance matrix of the historical meteorological data, and generating an environment sample set with the same distribution by utilizing a central limit theorem, wherein the method comprises the following steps of:
x 1 ，x 2 ，...，x n obeying the mean μ, variance σ 2 Is to make a certain distribution of
Wherein x is 1 ，x 2 ，...，x n N pieces of weather data in a certain dimension, wherein the dimension represents the type of the weather data; ζ represents obeying normal distribution;
according to the central limit theorem, n independent and uniformly distributed multi-dimension are generated to form an environment sample set.
The method for determining the environmental sample set quality evaluation index and comprehensively evaluating the environmental sample set quality comprises the following steps:
judging whether the environmental sample set passes quality assessment of uniformity, orthogonality, distribution consistency and representativeness; if the environment sample set passes all indexes, the next step is carried out;
otherwise, the environmental sample set is regenerated.
The judging whether the environment sample set passes through uniformity quality evaluation is realized by the following formula:
wherein 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 a matrix of n x m; MD (machine direction) device 2 A uniform quality evaluation index representing a mixing dispersion; x is x ij Representing the ith row and jth column samples of the environmental sample set;
if MD 2 (P) exceeding a threshold, indicating passing a homogeneous quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the uniformity quality assessment.
The judging whether the environment sample set passes the orthogonality quality evaluation is realized by the following formula:
wherein ρ is ij Representing a linear phase relationship between any two columns of a simulation sample set matrixA number; m represents the dimension of the environmental sample set; ρ is a correlation coefficient;
orthonormal quality assessment index ρ 2 Less than the threshold, then means through orthogonal quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the orthogonality quality assessment.
The judgment of whether the environment sample set passes the distribution consistency quality assessment is realized by the following formula:
wherein sigma 1 ，σ 2 Covariance matrix, μ of historical meteorological data and environmental sample set, respectively 1 ，μ 2 Respectively obtaining an average matrix of historical meteorological data and a test sample, wherein m is the dimension of an environmental sample; KL is a distribution consistency quality evaluation index, and represents KL divergence used for measuring consistency of two high-dimensional gaussian distributions; tr represents the trace of the matrix;
KL is greater than the threshold, then means through a distribution consistency quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the distribution consistency quality assessment.
The judging whether the environment sample set passes the quality evaluation of the representativeness is realized by the following formula:
ty＝N c /n
wherein N is the number of environmental samples in the environmental sample set, N c The number of the preset typical points is contained for the simulation test sample; ty is a typical evaluation index;
ty is greater than the threshold, then this represents an assessment by typical quality; otherwise, it means that the environment sample set needs to be regenerated without passing the typical quality assessment.
The method for generating the corresponding simulation image based on the environmental sample by using the infrared simulation system, estimating the imaging quality of the simulation image by adopting a TTP model, and marking the simulation sample so as to generate a training sample of the environmental adaptability evaluation model comprises the following steps:
and obtaining fuzzy band parameters by comparing the environmental adaptability deviation of the N groups of real scenes and the simulation scenes:
wherein T is i Measurement of imaging quality for a real scene, t i Imaging quality measurement values for the corresponding simulation scene;
if the imaging quality of a certain simulation scene is t, the imaging quality of the corresponding real scene is in the fuzzy band interval: [ t-delta, t+delta ];
when the TTP model is used for marking the sample, the measured imaging quality t and the left and right boundaries t-delta and t+delta of the imaging quality are marked as labels.
The training sample set is used for training the neural network to obtain an environmental adaptability assessment model, and the method comprises the following steps:
training a training set of an environment sample set as input, imaging quality t as output, and training by using a neural network to obtain a standard model;
training a training set of an environment sample set as input, a fuzzy left interval t-delta as output, and training by using a neural network to obtain a left boundary model;
and training the training set of the environment sample set by using the training set of the environment sample set as input and the fuzzy zone right interval t+delta as output, and obtaining a right boundary model by using the neural network training.
The verification is carried out under the outdoor scene, the reliability of the analysis and evaluation model is analyzed, and the method comprises the following steps:
1) Respectively inputting the test sets of the environment sample set into three models, and outputting results of the models as imaging quality;
if all three results are greater than the threshold, the classification result is environment adaptation, namely the imaging system can work normally under the current meteorological condition;
if any one of the three results is smaller than the threshold value, the classification result is that the environment is not suitable, namely the imaging system can not work normally under the current meteorological condition;
then comparing the classification result with the label, if the classification result is consistent with the label, the classification is correct; the ratio of the correct classification number to the number of samples in the test set is the model credibility p;
if the model reliability p is larger than the set value, the preliminary judgment of the evaluation model is reliable; otherwise, preliminary judgment is unreliable;
2) Inputting a test set of an environment sample set into a standard model, a left boundary model and a right boundary model, respectively outputting three imaging qualities, selecting a minimum value a and a maximum value b to obtain a range [ a, b ] of the imaging quality, and setting an imaging quality threshold value as alpha:
if a and b are both larger than alpha, judging that the scene is environment-adaptive at the moment;
if a and b are both smaller than alpha, judging that the scene is not suitable for the environment;
if a is less than α, b is greater than α and (b- α) > (α -a); judging the scene as environment adaptation at the moment;
if a is less than α, b is greater than α and (b- α) < (α -a); at this time, the scene is judged as being environmentally unadapted.
The imaging quality of the outdoor collected infrared image is measured by using a TTP model to obtain an environmental adaptability evaluation result,
comparing with an environmental adaptability result output by the analysis and evaluation model under the same condition, and if the environmental adaptability results are consistent, ensuring reliability; otherwise, it is unreliable.
The invention has the following beneficial effects and advantages:
1. by utilizing the infrared visual simulation system, the number of experimental data is increased, the problem of high difficulty of outdoor experiments is solved, and the objectivity of an evaluation model is increased.
2. Simulation data are generated through environmental parameter modeling and are evaluated through uniformity, orthogonality, distribution consistency and typical simulation data, so that the availability of the simulation data is verified.
3. By utilizing the neural network to learn the nonlinear relation between the environmental parameters and the environmental adaptability, the environmental adaptability assessment model is modeled more accurately.
4. By the analysis method of the fuzzy band, the reliability of the evaluation model is more fully described by considering the fuzzy band into the reliability of the environment adaptability evaluation model.
5. Through verification on a small number of real scenes, the method has a good evaluation result when performing environmental adaptability evaluation on various targets.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a schematic diagram of a simulation sample comprehensive evaluation of the present invention;
FIG. 3 is a schematic of uniformity of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
The technical scheme adopted by the invention for achieving the purpose is as follows:
the imaging system environment adaptability evaluation method based on infrared simulation comprises the following steps:
modeling an environment parameter space by using a high-dimensional Gaussian distribution model, and generating an environment sample set based on the model;
the quality of the environmental sample set is evaluated by designing an evaluation index;
generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model;
training the Levenberg-Marquardt backpropagation algorithm neural network by using a training sample set to obtain an environmental adaptability assessment model;
and (3) designing an experiment, and collecting outdoor infrared images for model verification.
The environmental parameter space modeling is to generate simulation samples with the same distribution by using a central limit theorem. Namely:
let x be 1 ，x 2 ，...，x n Obeying the mean value as mu and the variance as sigma 2 Is to make a certain distribution of
Wherein x is 1 ，x 2 ，...，x n Refers to n data of a certain dimension, e.g. n temperature values or n pressure values, n refers to n sets of data.
When n is sufficiently large, the xi asymptotes follow a normal distribution, i.e., the distribution among n random variables which are mutually independent and distributed in the same way approximates the normal distribution, and the larger n is, the better the approximation is. According to the central limit theorem, the generation of the high-dimensional Gaussian distribution can be realized by generating the uniform distribution of n independent same distributions of high dimensions.
The environmental sample set evaluation index is: uniformity, orthogonality, distribution uniformity, and representativeness. Wherein:
uniformity: based on the uniformity criterion (i.e. the dispersion-based criterion), not only can the uniformity of the sampling points in the multidimensional cube be evaluated, but also the uniformity of the projection thereof in the low dimension can be ensured, so that the method is commonly used for measuring the space filling property of experimental tests.
The dispersion measurement method mainly comprises the steps of center L2-dispersion, rollable L2-dispersion and mixed dispersion. Wherein the center L2-dispersion is more concerned about points near the vertex, and the sensitivity to points in the center region is lower; the rolling L2-dispersion has unchanged result when each factor translates and changes; the mixing dispersion can overcome the defects of the two, so that the space filling property tested by adopting the mixing dispersion measurement experiment can be well overcome. The formula is as follows:
m is the dimension, n is the number, where m=5 is the number of five dimensions, n is the number of environmental sample sets. The data set is a matrix of n x m, and the larger the index value, the better.
Orthogonality: the experimental design matrix is often required to have a "clean and comparable" nature, i.e., 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 test sample distribution is to "clean up comparable". A common orthogonality metric is the correlation coefficient ρ between experimental factors.
Wherein ρ is ij Representing the linear correlation coefficient between two columns of the matrix of the simulated sample set. m represents the number of columns, i.e. the dimension of the environmental parameter. The smaller the index value is, the better.
Distribution consistency: the actual system factors have a certain distribution rule in some cases, so that the advantages and disadvantages of the experimental design scheme can be analyzed by measuring the consistency of the distribution of the test samples and the actual system factors. When the factor distribution of the actual system is known, hypothesis test can be performed on point factors of experimental tests, so as to judge whether the factor distribution of the simulation test sample and the actual system is consistent. The better the consistency degree is, the more the test sample is selected to be in line with the actual situation, and the more the experimental result is trustworthy. The historical meteorological data accords with high-dimensional Gaussian distribution, the selected environment sample set also accords with the high-dimensional Gaussian distribution, the consistency of the two high-dimensional Gaussian distributions is measured by the KL divergence, the smaller the KL divergence is, the higher the consistency is, and the KL divergence formula is as follows:
wherein sigma 1 ，σ 2 Covariance matrix of real meteorological data and generated environment sample set respectively, mu 1 ，μ 2 The mean matrix is the real meteorological data and the mean matrix of the test sample respectively, and m is the dimension of the test sample. The larger the index value is, the better.
Typically: the typical measure of the simulated sample is calculated by simulating the extent to which the test sample covers a predetermined typical point.
ty＝N c /n
Wherein N is the number of test samples, N c The number of typical points is preset for the simulation test sample.
The larger the better
The simulation sample marking process provides a fuzzy band analysis method for the problem due to certain deviation between a simulation system and a real scene, and the fuzzy band analysis method is combined into experimental analysis.
And obtaining fuzzy band parameters by comparing the environmental adaptability deviation of the N groups of real scenes and the simulation scenes:
wherein T is i Measurement of imaging quality for a real scene, t i Imaging quality measurement values for the corresponding simulation scene. Therefore, if the imaging quality of a certain simulation scene is t, the corresponding real scene imaging quality is within the fuzzy band interval: [ t-delta, t+delta ]]. When the TTP model is used for marking a sample, not only the measured imaging quality t is marked, but also the left and right boundaries t-delta and t+delta of the imaging quality are needed to be obtained by using the fuzzy band parameters.
Compared with the fastest gradient descent method of the traditional BP algorithm, the Levenberg-Marquardt backpropagation algorithm has the advantages of high convergence speed and good convergence. The invention adopts the Levenberg-Marquardt backpropagation algorithm algorithm for training.
The environmental adaptability assessment model is as follows: through fuzzy band analysis, the invention trains three prediction models, namely a standard model, a left boundary model and a right boundary model, by utilizing a neural network. The standard model is a classifier under the condition that the simulation system has no error, the left boundary model is a classifier under the condition that the simulation system has negative maximum error, and the right boundary model is a classifier under the condition that the simulation system has positive maximum error. Integrating three models is the environmental assessment model of the present invention. And (5) giving the credibility of the environment assessment model by adopting result verification: the test set is respectively input into three models, the models output results (the results are imaging quality), if the three results are all larger than a threshold value, the classification results are environment-adaptive, namely the imaging system can work normally under the current meteorological condition, if the three results are smaller than the threshold value, the classification results are environment-unadapted, namely the imaging system can not work normally under the current meteorological condition. And then comparing the classification result with the label, wherein the total sample number on the correct classification number ratio is the model credibility p.
The experimental verification is that the evaluation is carried out through an environmental adaptability evaluation model and a real experiment under a real typical scene, and then the comparison is carried out.
The imaging system environment adaptability evaluation method based on infrared simulation comprises the following steps:
step 1: obtaining historical meteorological data of a target area, performing high-dimensional Gaussian distribution model fitting, and generating an environment sample set with the same distribution by using a central limit theorem;
step 2: designing an environmental sample set quality evaluation index, and comprehensively evaluating the quality of the environmental sample set generated in the step 1;
step 3: generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model;
step 4: training the Levenberg-Marquardt backpropagation algorithm neural network by using a training sample set to obtain an environmental adaptability assessment model;
step 5: and (3) designing an experiment, and collecting outdoor infrared images for model verification.
The environmental sample set generation includes the following processes:
firstly, carrying out high-dimensional Gaussian analysis on historical meteorological data to obtain a mean value and a covariance matrix of the historical meteorological data, and then generating simulation samples with the same distribution by using a central limit theorem.
The environmental sample set evaluation includes the following processes:
based on uniformity criteria (i.e., dispersion-based criteria): not only can the uniformity of sampling points in the multidimensional cube be evaluated, but also the uniformity of projection of the sampling points in the low dimension can be ensured, so that the sampling points are commonly used for measuring the space filling property of experimental tests.
Based on the orthogonality criterion: the experimental design matrix is often required to have a "clean and comparable" nature, i.e., 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 test sample distribution is to "clean up comparable". A common orthogonality metric is the correlation coefficient ρ between experimental factors.
Based on a distribution consistency criterion: the actual system factors have a certain distribution rule in some cases, so that the advantages and disadvantages of the experimental design scheme can be analyzed by measuring the consistency of the distribution of the test samples and the actual system factors. When the factor distribution of the actual system is known, hypothesis test can be performed on point factors of experimental tests, so as to judge whether the factor distribution of the simulation test sample and the actual system is consistent. The better the consistency degree is, the more the test sample is selected to be in line with the actual situation, and the more the experimental result is trustworthy.
Based on the typical criteria: the typical measure of the simulated sample is calculated by simulating the extent to which the test sample covers a predetermined typical point.
The verification is carried out outdoors, and the reliability of the analysis and evaluation model comprises the following processes:
inputting the sample to be tested into three models can output three imaging quality, and a range [ a, b ] of the imaging quality can be obtained, wherein the range is an estimated fuzzy band range, and an imaging quality threshold value is set as alpha:
if a and b are both larger than alpha, judging that the scene is environment-adaptive at the moment;
if a and b are both smaller than alpha, judging that the scene is not suitable for the environment;
if a is less than α, b is greater than α and (b- α) > (α -a). Judging the scene as environment adaptation at the moment;
if a is less than α, b is greater than α and (b- α) < (α -a). At this time, the scene is judged as being environmentally unadapted.
And finally, verifying whether the environmental adaptability evaluation results are consistent with the environmental adaptability evaluation results actually measured under the same conditions (multiple dimensions: temperature, air pressure, humidity, wind speed, visibility and the like).
As shown in fig. 1, the environmental suitability evaluation and reliability analysis method based on the infrared simulation system of the invention comprises the following steps:
(1): obtaining historical meteorological data of a target area, performing high-dimensional Gaussian distribution model fitting, and generating an environment sample set with the same distribution by using a central limit theorem;
(2): designing an environmental sample set quality evaluation index, and comprehensively evaluating the quality of the environmental sample set generated in the step 1;
(3) Generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model;
(4) Training the Levenberg-Marquardt backpropagation algorithm neural network by using a training sample set to obtain an environmental adaptability assessment model;
(5) And (3) designing an experiment, and collecting outdoor infrared images for model verification.
The experiments were all completed under the weather conditions of the Shenyang summer, 2646 groups of simulation samples were generated, FIG. 2 depicts the composition of the comprehensive evaluation index, and FIG. 3 depicts the principle of uniformity. Evaluation of the simulation samples is shown in table 1:
table 1 evaluation data
According to the indexes of the table, the selected simulation samples are uniform in corresponding weather condition distribution, low in correlation among weather conditions, basically consistent with the actual sample distribution, high in proportion of typical scenes and capable of being used for carrying out the following research.
And comparing the environmental adaptability deviation of the N groups of real scenes and the simulation scene to obtain a fuzzy band parameter, and setting the calculated fuzzy band error delta to be 0.05.
Under the conditions of table 2:
TABLE 2 Meteorological data set
The imaging quality under this condition is: 0.668418
Then after adding the error range, the left boundary of the imaging quality is 0.618418 and the right boundary is 0.718418
2646 sets of samples were taken according to 7: and 3, dividing the model into a training set and a testing set, training on the training set, and testing on the testing set to obtain the credibility of the model.
Testing on 793 test sets resulted in model confidence: 0.9733
By verification under the conditions of outdoor table 3:
TABLE 3 outdoor test Meteorological Condition
The output of the environment evaluation model is adaptive, and the actual measured result is also adaptive, thereby proving the effectiveness of the method.
Claims (10)
1. The imaging system environment adaptability evaluation method based on infrared simulation is characterized by comprising the following steps of:
acquiring historical meteorological data of a target area, performing high-dimensional Gaussian distribution fitting, and generating an environment sample set with the same distribution by using a central limit theorem;
determining an environmental sample set quality evaluation index, and comprehensively evaluating the environmental sample set quality;
generating a corresponding simulation image based on the environment sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample so as to generate a training sample of the environment adaptability evaluation model;
training the neural network by using a training sample set to obtain an environmental adaptability evaluation model;
and acquiring infrared images under outdoor scenes to perform model verification.
2. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 1, wherein the method is characterized in that the method comprises the steps of performing high-dimensional Gaussian distribution fitting and generating an environmental sample set with the same distribution by using a central limit theorem, and specifically comprises the following steps: carrying out high-dimensional Gaussian analysis on historical meteorological data to obtain a mean value and a covariance matrix of the historical meteorological data, and generating an environment sample set with the same distribution by utilizing a central limit theorem, wherein the method comprises the following steps of:
x 1 ，x 2 ，...，x n obeying the mean μ, variance σ 2 Is to make a certain distribution of
Wherein x is 1 ，x 2 ，...，x n N pieces of weather data in a certain dimension, wherein the dimension represents the type of the weather data; ζ represents obeying normal distribution;
according to the central limit theorem, n independent and uniformly distributed multi-dimension are generated to form an environment sample set.
3. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 1, wherein the steps of determining an environmental sample set quality evaluation index and comprehensively evaluating the environmental sample set quality include the steps of:
judging whether the environmental sample set passes quality assessment of uniformity, orthogonality, distribution consistency and representativeness; if the environment sample set passes all indexes, the next step is carried out;
otherwise, the environmental sample set is regenerated.
4. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 3, wherein the determining whether the environmental sample set passes the uniformity quality evaluation is implemented by:
wherein 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 a matrix of n x m; MD (machine direction) device 2 A uniform quality evaluation index representing a mixing dispersion; x is x ij Representing the ith row and jth column samples of the environmental sample set;
if MD 2 (P) exceeding a threshold, indicating passing a homogeneous quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the uniformity quality assessment.
5. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 3, wherein the determining whether the environmental sample set passes the orthogonality quality evaluation is implemented by:
wherein ρ is ij Representing linear correlation coefficients between any two columns of the simulation sample set matrix; m represents the dimension of the environmental sample set; ρ is a correlation coefficient;
orthonormal quality assessment index ρ 2 Less than the threshold, then means through orthogonal quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the orthogonality quality assessment.
6. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 3, wherein the determining whether the environmental sample set passes the distribution consistency quality evaluation is implemented by:
wherein sigma 1 ，σ 2 Covariance matrix, μ of historical meteorological data and environmental sample set, respectively 1 ，μ 2 Mean moment of historical meteorological data and test sample respectivelyAn array, m is the dimension of the environmental sample; KL is a distribution consistency quality evaluation index, and represents KL divergence used for measuring consistency of two high-dimensional gaussian distributions; tr represents the trace of the matrix;
KL is greater than the threshold, then means through a distribution consistency quality assessment; otherwise, it means that the environmental sample set needs to be regenerated without passing the distribution consistency quality assessment.
7. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 3, wherein the determining whether the environmental sample set passes the quality-of-representativeness evaluation is implemented by:
ty＝N c /n
wherein N is the number of environmental samples in the environmental sample set, N c The number of the preset typical points is contained for the simulation test sample; ty is a typical evaluation index;
ty is greater than the threshold, then this represents an assessment by typical quality; otherwise, it means that the environment sample set needs to be regenerated without passing the typical quality assessment.
8. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 1, wherein the generating a corresponding simulation image based on an environmental sample by using an infrared simulation system, estimating the imaging quality of the simulation image by using a TTP model, and marking the simulation sample to generate a training sample of the environmental suitability evaluation model comprises the steps of:
and obtaining fuzzy band parameters by comparing the environmental adaptability deviation of the N groups of real scenes and the simulation scenes:
wherein T is i Measurement of imaging quality for a real scene, t i Imaging quality measurement values for the corresponding simulation scene;
if the imaging quality of a certain simulation scene is t, the imaging quality of the corresponding real scene is in the fuzzy band interval: [ t-delta, t+delta ];
when the TTP model is used for marking the sample, the measured imaging quality t and the left and right boundaries t-delta and t+delta of the imaging quality are marked as labels.
9. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 1, wherein the training of the neural network using the training sample set to obtain the environmental suitability evaluation model comprises the following steps:
training a training set of an environment sample set as input, imaging quality t as output, and training by using a neural network to obtain a standard model;
training a training set of an environment sample set as input, a fuzzy left interval t-delta as output, and training by using a neural network to obtain a left boundary model;
and training the training set of the environment sample set by using the training set of the environment sample set as input and the fuzzy zone right interval t+delta as output, and obtaining a right boundary model by using the neural network training.
10. The method for evaluating the environmental suitability of an imaging system based on infrared simulation according to claim 1, wherein the verification is performed under outdoor scenes, the reliability of the evaluation model is analyzed, and the method comprises the following steps:
1) Respectively inputting the test sets of the environment sample set into three models, and outputting results of the models as imaging quality;
if all three results are greater than the threshold, the classification result is environment adaptation, namely the imaging system can work normally under the current meteorological condition;
if any one of the three results is smaller than the threshold value, the classification result is that the environment is not suitable, namely the imaging system can not work normally under the current meteorological condition;
then comparing the classification result with the label, if the classification result is consistent with the label, the classification is correct; the ratio of the correct classification number to the number of samples in the test set is the model credibility p;
if the model reliability p is larger than the set value, the preliminary judgment of the evaluation model is reliable; otherwise, preliminary judgment is unreliable;
2) Inputting a test set of an environment sample set into a standard model, a left boundary model and a right boundary model, respectively outputting three imaging qualities, selecting a minimum value a and a maximum value b to obtain a range [ a, b ] of the imaging quality, and setting an imaging quality threshold value as alpha:
if a and b are both larger than alpha, judging that the scene is environment-adaptive at the moment;
if a and b are both smaller than alpha, judging that the scene is not suitable for the environment;
if a is less than α, b is greater than α and (b- α) > (α -a); judging the scene as environment adaptation at the moment;
if a is less than α, b is greater than α and (b- α) < (α -a); judging the scene as environment-unadapted at the moment;
the imaging quality of the outdoor acquired infrared image is measured by using a TTP model to obtain an environmental adaptability evaluation result, the environmental adaptability evaluation result is compared with the environmental adaptability result output by the analysis evaluation model under the same condition, and if the environmental adaptability evaluation result is consistent, the environmental adaptability evaluation result is reliable; otherwise, it is unreliable.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111558197.7A CN116305705A (en) | 2021-12-20 | 2021-12-20 | Imaging system environment adaptability assessment method based on infrared simulation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111558197.7A CN116305705A (en) | 2021-12-20 | 2021-12-20 | Imaging system environment adaptability assessment method based on infrared simulation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116305705A true CN116305705A (en) | 2023-06-23 |
Family
ID=86783807
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111558197.7A Pending CN116305705A (en) | 2021-12-20 | 2021-12-20 | Imaging system environment adaptability assessment method based on infrared simulation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116305705A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118225254A (en) * | 2024-05-07 | 2024-06-21 | 南京海汇装备科技有限公司 | Multi-parameter anomaly calibration system and method applied to thermal infrared imager |
-
2021
- 2021-12-20 CN CN202111558197.7A patent/CN116305705A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118225254A (en) * | 2024-05-07 | 2024-06-21 | 南京海汇装备科技有限公司 | Multi-parameter anomaly calibration system and method applied to thermal infrared imager |
| CN118225254B (en) * | 2024-05-07 | 2024-11-08 | 南京海汇装备科技有限公司 | Multi-parameter anomaly calibration system and method applied to thermal infrared imager |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105740991B (en) | Climate change prediction method and system based on improved BP neural network fitting of multiple climate modes | |
| CN111310968A (en) | LSTM neural network circulation hydrological forecasting method based on mutual information | |
| Torrielli et al. | Extreme wind speeds from long-term synthetic records | |
| CN107436963B (en) | O-shaped rubber sealing ring service life prediction method based on Copula function multi-degradation | |
| CN112285664B (en) | Method for evaluating countermeasure simulation confidence of radar-aircraft system | |
| CN112883497B (en) | Space valve reliability assessment method based on multi-source information fusion | |
| CN104899135A (en) | Software defect prediction method and system | |
| CN116308958A (en) | Carbon emission online detection and early warning system and method based on mobile terminal | |
| CN108154271A (en) | A kind of surface air temperature method of quality control based on spatial coherence and surface fitting | |
| CN116137012B (en) | Online teaching quality supervision and management system based on Internet education | |
| CN106599367A (en) | Method for detecting abnormal state of spacecraft | |
| CN114264915A (en) | Power distribution network cable joint operation condition assessment early warning device and method | |
| CN114359695A (en) | Insulator breakage identification method based on uncertainty estimation | |
| CN116305705A (en) | Imaging system environment adaptability assessment method based on infrared simulation | |
| CN111178756A (en) | Multiple linear regression fire risk assessment method based on environmental big data | |
| CN110889207A (en) | System combination model credibility intelligent evaluation method based on deep learning | |
| CN118362321B (en) | Automobile seat comfort evaluation management system based on artificial intelligence | |
| CN108052953B (en) | Sample expansion method based on feature correlation | |
| CN107066786A (en) | Aerosol optical depth inversion algorithm based on neutral net | |
| CN117352088A (en) | Prediction method of spatial pollutant distribution based on convolutional neural network | |
| CN113553708B (en) | Method and device for tracing key influence factors of simulation model | |
| CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic neural network | |
| CN115983329A (en) | Method, device, equipment and storage medium for predicting air quality and meteorological conditions | |
| CN115907178A (en) | Clean ecosystem CO 2 Method for predicting exchange amount | |
| CN114563771A (en) | Double-threshold laser radar cloud layer detection algorithm based on cluster analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |