A fuzzy comprehensive evaluation method for construction effect of artificial electromagnetic environment based on SAR image and signal

Abstract

The application provides a comprehensive evaluation method for SAR artificial electromagnetic environment construction effect, and belongs to the SAR image similarity evaluation field.The method comprises two parts of index analysis selection and fuzzy comprehensive evaluation, image sample augmentation, index calculation and discretization processing are carried out on the image index of SAR, so that attribute reduction of the image index is realized, the key image index is obtained, and the commonly used SAR signal evaluation index is added, the evaluation indexes are comprehensively used, the two-stage fuzzy comprehensive evaluation method is used, the comprehensive evaluation of the SAR artificial electromagnetic environment construction effect is carried out, the similarity between the original image and the image constructed by the artificial electromagnetic environment is evaluated, and finally the required evaluation grade is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of SAR image similarity evaluation, and in particular to a fuzzy comprehensive evaluation method for the construction effect of artificial electromagnetic environment applicable to SAR images and signals. BACKGROUND

[0002] Since synthetic aperture radar (SAR) can be immune to weather, and can complete high-resolution imaging observation on the terrain, facilities, fixed and low-speed targets in the survey area, it has important military application value. In order to cover important military facilities and combat equipment of our own side, and to conceal military operations, it is urgently needed to study the artificial electromagnetic environment construction technology effective against SAR, i.e. the technology of interfering with enemy SAR to prevent information of our own facilities and equipment and military operations from being acquired by enemy SAR. At present, this technology is divided into two methods: suppression type and deception type. The deception type method is to calculate the echo signal of a false target through a certain method according to the in-pulse parameters of the SAR signal intercepted by reconnaissance, and to forward the signal to the SAR, so that the SAR signal acquired by the enemy contains false target information after imaging, thereby covering the important target information of our own side. Since the false target is coherent in range and azimuth, it can obtain the same processing gain as the real target echo signal, and the required interference power of the SAR deception type artificial electromagnetic environment construction technology is low, so it has become a research hotspot in SAR countermeasures.

[0003] In order to study the SAR effective deception type artificial electromagnetic environment construction technology, it is necessary to evaluate the construction effect of SAR deception type artificial electromagnetic environment, and to promote the development of artificial electromagnetic environment construction technology through accurate evaluation.

[0004] The existing SAR artificial electromagnetic environment construction effect evaluation mainly has the following two problems: first, it only evaluates the features of SAR images, and does not consider the features of SAR two-dimensional signals. Peak side lobe ratio, integral side lobe ratio and spatial resolution to some extent reflect SAR signal energy and image definition information, and these feature indexes have important significance in SAR image similarity evaluation, and the lack of these feature indexes will affect the accuracy of the evaluation result. Second, the selection of indexes is not analyzed: the existing evaluation method mainly has the following two ways in the selection of indexes: one is to subjectively select one or several indexes, which may lead to insufficient information that the selected indexes can represent, so that the evaluation angle is not comprehensive enough, and the accuracy of the final evaluation result is affected. The second is to select too many evaluation indexes, which may contain redundant indexes. Since each index has its corresponding weight, the weight of the redundant index will cause the information reflected by the index to be superimposed in the weight, resulting in a decrease in the accuracy of the evaluation result. SUMMARY

[0005] In order to overcome the above SAR image comprehensive evaluation method, the present application proposes a kind of fuzzy comprehensive evaluation method based on SAR image and signal artificial electromagnetic environment construction effect, the advantage of this method is: first, on the basis of considering the image feature, six important indexes of signal feature are added, so that the angle of evaluation is more rich, and the evaluation result is more accurate;Second, the attributes of many commonly used indexes in the index of image feature are reduced, which can guarantee the comprehensiveness of the information represented by the index, eliminate redundant attributes, select the evaluation index suitable for the sample group, guarantee the accuracy of the information weight represented by the index, and then improve the accuracy of the evaluation result.The evaluation method is divided into index selection module and fuzzy comprehensive evaluation module, and the total process is as shown in Figure 1 The index selection module is selected in the index of image, and the index of each group of input original image and simulation image is screened, the redundant index is eliminated, the index reduction is realized, the accuracy of index selection is improved on the basis of guaranteeing the comprehensiveness of index, and six commonly used indexes of SAR signal feature are added.The index after reduction is used, the similarity between original image and signal and artificial electromagnetic environment construction image and signal is evaluated systematically and comprehensively by using fuzzy comprehensive evaluation method, and the required evaluation grade is obtained.

[0006] Figure 2 The flow chart of index selection module includes image sample augmentation, image index calculation, data discretization and attribute reduction.The input of the module is a original image formed by SAR according to actual target and a simulation image constructed by artificial electromagnetic environment.

[0007] Image sample augmentation: Since attribute reduction requires the analysis of various indicators of multiple image samples, the number of input image samples is insufficient to perform subsequent attribute reduction, so image sample augmentation is needed. Traditional image sample augmentation methods, such as horizontal or vertical flipping, scaling, cropping, and translation, can hardly distinguish the augmented samples from the original samples in terms of statistical and geometric features, and thus cannot analyze the differences between the sample data. Therefore, a deep learning method is needed to augment the image samples to generate sample data similar to the original samples for subsequent reduction steps. The image sample augmentation network uses the SinGAN network proposed by Tamar Rott Shaham et al. in the paper “SinGAN: Learning a Generative Model from a Single Natural Image”. The basic principle of using SinGAN for image sample augmentation is that the network can learn the internal distribution of an image from a single image data and generate image data similar to the input image, which can achieve image sample augmentation for a single SAR image. The generator of the network can extract the internal features of the input SAR image and output image data with the features, and the discriminator is used to distinguish the true and false of the generated image data. In the process of “game” between the generator and the discriminator, the generator is constantly improved to generate more realistic data, and the discriminator is constantly improved to improve its ability to distinguish the true and false of the image data. Finally, when the discriminator cannot judge whether the output image of the generator is true or false, the output image of the generator is realistic enough, and this image can be used as an image sample augmentation image. The specific steps are as follows: input a simulated image into the SinGAN network, and after training, generate image sample data with the same style type and similar content but not the same as the input image sample. In order to calculate simply and accurately, the most similar m image samples need to be manually selected from the augmented generated image samples.

[0008] Image indicator calculation: In order to perform the subsequent attribute reduction step, the data of the statistical and geometric feature indicators of the image samples need to be calculated. The statistical feature indicators include: mean, variance, entropy, dynamic range, equivalent number of views, radiation resolution, and average gradient of image gray scale. The geometric feature indicators include: area, perimeter, area-perimeter ratio, minimum circumscribed rectangle perimeter, minimum circumscribed rectangle area, shape parameter, circularity, aspect ratio, elongation, sphericity, and equivalent circle diameter of image target.

[0009] Data discretization: In order to carry out the subsequent attribute reduction step, the calculated index data needs to be assigned to discrete levels according to size. Set the number of discrete intervals n, first sort the same index data of m images in ascending order, calculate the difference value of adjacent data, select the endpoints of the largest n-1 difference values as the interval breakpoints, and divide into n discrete intervals. Assign discrete levels from 1 to n to each interval in size order. Discretize the 17 indicators of image statistical features and geometric features respectively.

[0010] Attribute reduction: There are a large number of evaluation indicators for the evaluation of the construction effect of the artificial electromagnetic environment of SAR images. In selecting the final evaluation indicators, the redundant indicators in the SAR image evaluation are removed by reducing the indicators of the SAR image. According to the concept and method introduced in the paper "Minimum Attribute Reduction Algorithm Based on 0-1 Programming" by Jane Wanrong and Yu Hai, the reduction problem is gradually transformed into a 0-1 programming problem, and the attribute reduction is realized by solving the problem. The basic principle of attribute reduction is: the discretized data is used as the information system in rough set theory, and the reduction of the information system is defined as the minimal attribute set AT that does not change the indistinguishable relationship ind(AT) of the non-empty finite set. The attributes in the reduction are taken from the non-empty elements in the discrimination matrix M, and the repeated elements and empty sets in the discrimination matrix are removed to obtain the minimal discrimination set MS. To find the minimum attribute reduction R, it needs to satisfy that R has a non-empty intersection with any set S in the minimal discrimination set, and the set after removing any attribute a in R has no non-empty intersection with at least one set in the minimal discrimination set. The above problem can be transformed into an optimal solution problem (formula 1):

[0011]

[0012] Then according to the mathematical theory, the problem is transformed into the corresponding 0-1 programming problem (formula 2):

[0013]

[0014] Where x is a 0-1 column vector corresponding to the minimum attribute reduction R, and the i-th attribute a i is selected into the minimum reduction R, the element x i in this row in x is 1, otherwise it is zero. c represents an m-dimensional column vector with all components being 1. The matrix P = (p1, p2, …, p l ) T , where p i = (p i1 , p i2 , …, p im ), i = 1,2, …, l, j = 1,2, …, m. q is an l-dimensional column vector with all components being 1. Solving this 0-1 programming problem, the minimum attribute reduction is obtained according to the result, the redundant attributes are removed, and the output result of the index selection module is obtained, that is, the evaluation index of the image selected in the fuzzy comprehensive evaluation module. The specific steps of attribute reduction are as follows: first, the discernibility matrix M is calculated; second, the minimum discernibility set MS = {S1, S2, …, S l} is calculated; third, the matrix P is calculated from the minimum discernibility set MS; fourth, the 0-1 programming (formula 2) is solved to obtain a minimum attribute reduction, and the corresponding attributes are found, that is, the final evaluation index is determined.

[0015] The fuzzy comprehensive evaluation module is the result output by the index selection module, that is, the statistical feature index and the geometric feature index of the image part after reduction, which is selected as the evaluation index of the image part, and then the six indexes of the signal part are added, and the two-level fuzzy comprehensive evaluation method is used to evaluate the similarity of the SAR image and the signal. The first level is to allocate the weight among the three types of indexes of statistical features, geometric features, and signal features, and the second level is to allocate the weight of each evaluation index within each feature. The two-level fuzzy comprehensive evaluation can more accurately control and allocate the weight of each index, thereby improving the accuracy of the evaluation result. In addition to the reduced image index, the input part of the fuzzy comprehensive evaluation module also includes the calculation values of the reduced image index of the initial input original image and the simulation image constructed by the artificial electromagnetic environment, and the calculation values of each item of the signal index of the two images.

[0016] The fuzzy comprehensive evaluation method converts qualitative evaluation into quantitative evaluation according to the membership degree theory of fuzzy mathematics, that is, it uses fuzzy mathematics to make a general evaluation of things or objects subject to multiple factors. It has the characteristics of clear results and strong systematization, can better solve fuzzy, difficult to quantify, and non-deterministic problems, and is suitable for similarity evaluation of the simulation image of the SAR artificial electromagnetic environment construction and the original image. The basic principle of fuzzy comprehensive evaluation is to determine the evaluation set of several levels from good to bad of the evaluated object and the weight of the evaluation index from the factors affecting the problem, make a corresponding fuzzy evaluation of each index, determine the membership function, form a fuzzy judgment matrix, and perform fuzzy operation with the weight matrix to obtain a quantitative comprehensive evaluation result. According to the methods of Hu Liping et al. in “Vehicle Target SAR Simulation Image Evaluation Method Based on Fuzzy Comprehensive Evaluation” and He Fengbiao in “MATLAB Implementation of Fuzzy Comprehensive Method”, the fuzzy comprehensive evaluation module can be divided into the following steps: determining the selected index, determining the evaluation set, calculating the fuzzy relationship matrix, determining the weight vector, fuzzy synthesis, and outputting the evaluation result. The overall flowchart of the module is as shown in Figure 3 .

[0017] Determination of selected indicators: evaluation indicators are divided into image indicators and signal indicators. Image indicators are the results output by the indicator selection module, that is, the reduced statistical feature indicators and geometric feature indicators. Signal feature indicators include: azimuth and range peak-to-sidelobe ratio, azimuth and range integrated sidelobe ratio, azimuth and range spatial resolution. These commonly used SAR signal feature indicators represent different characteristics, so they are not reduced.

[0018] Determination of comment set: the comment set is the final evaluation level set, which is used to represent the similarity degree of the SAR simulation image and the original image. The comment set is defined as five levels, that is, {very good, better, medium, worse, very bad}.

[0019] Fuzzy relation matrix calculation: in order to perform subsequent fuzzy synthesis, the fuzzy relation matrix of each feature needs to be calculated according to the membership function of each feature. The determination of the membership function can refer to the previous experience. The statistical features adopt the half-lower trapezoidal to calculate the membership degree, and the geometric features and signal features adopt the triangle and half-lower trapezoidal to calculate the membership degree. According to the method of He Fengbiao in the book "Fuzzy Comprehensive Method MATLAB Implementation", the ratios of the statistical feature indicators, geometric feature indicators and signal feature indicators of the original image and the simulation image are calculated respectively, and are brought into the corresponding membership functions to obtain the fuzzy relation matrix R1 of the statistical features, the fuzzy relation matrix R2 of the geometric features and the fuzzy relation matrix R3 of the signal features.

[0020] Determination of weight vector: in order to perform subsequent fuzzy synthesis, the weight vector of each feature needs to be calculated according to the judgment matrix of each feature. The weight vector represents the weight of each indicator, and is also the eigenvector of the artificially constructed judgment matrix. According to the method of Yan Zhiqiang et al. in the paper "Research and Practice of Automatic Evaluation of 5G Cell Performance Based on AHP Two-level Fuzzy Comprehensive Evaluation Method", the relative importance of each indicator is determined by using 1-9 scale method through pairwise comparison. The judgment matrices A1, A2 and A3 of the statistical features, geometric features and signal features of the second layer are constructed respectively, and the overall judgment matrix A of the statistical features, geometric features and signal features of the first layer is constructed. The random consistency ratio CR of the judgment matrices A1, A2, A3 and A is calculated respectively, and consistency test is performed. If CR<0.1, the test is not passed, and the judgment matrix needs to be reconstructed and the weight vector needs to be determined again.

[0021] After consistency test, the eigenvectors W1, W2, W3 and W of each judgment matrix are calculated. The weight vector of each feature is the eigenvector of the corresponding judgment matrix, so the weight vectors of the second layer are W1, W2 and W3, which correspond to the indicators of the statistical features, geometric features and signal features respectively. The weight vector of the first layer is W, which is the weight vector of the three features as a whole.

[0022] Fuzzy synthesis: in order to obtain the FCE result vector, the weight vector of the statistical feature, the geometric feature and the signal feature needs to be combined with the fuzzy relation matrix corresponding to each feature to obtain the result vector. The fuzzy evaluation vector of the statistical feature is B1=W1*R1, the fuzzy evaluation vector of the geometric feature is B2=W2*R2, the fuzzy evaluation vector of the signal feature is B3=W3*R3, the fuzzy relation matrix of the first layer is R=[B1,B2,B3], the weight vector of the first layer is W, and the FCE result vector is B=W*R=W*[B1,B2,B3] (the fuzzy operator. is a dot product operation).

[0023] Output evaluation result: for the FCE result vector after fuzzy synthesis, the evaluation result level corresponding to the maximum membership degree principle in the evaluation set is output, that is, the level corresponding to the maximum value in the FCE result vector is selected as the final evaluation result level.

[0024] Beneficial effects

[0025] The application provides a comprehensive evaluation method for SAR artificial electromagnetic environment construction effect, which enriches the evaluation angle and makes the evaluation result more accurate by adding six important indexes of signal features on the basis of image features; secondly, the attributes of many commonly used indexes in the image features are reduced, so that the comprehensive information of the indexes is ensured, redundant attributes are removed, evaluation indexes suitable for the sample set are screened, the accuracy of the information weight of the indexes is ensured, and the accuracy of the evaluation result is improved; and the SAR artificial electromagnetic environment construction effect can be systematically and comprehensively evaluated by combining fuzzy comprehensive evaluation, so that the required evaluation level is obtained, and an effective method for solving the SAR artificial electromagnetic environment construction effect evaluation problem is provided. DETAILED DESCRIPTION

[0026] Figure 1 The application implements a general flowchart.

[0027] Figure 2 The flowchart for index selection.

[0028] Figure 3 The flowchart for fuzzy comprehensive evaluation.

[0029] Figure 4 The original image of SAR according to the actual target.

[0030] Figure 5 The simulation image of artificial electromagnetic environment construction.

[0031] Figure 6 (a)-(i) are nine image samples selected from the augmented generated image samples.

[0032] Figure 7 Data of statistical feature indexes for 9 image samples.

[0033] Figure 8 Data of geometric feature indexes for 9 image samples.

[0034] Figure 9 Data of statistical feature indexes after discretization.

[0035] Figure 10 Data of geometric feature indexes after discretization.

[0036] Figure 11 Bar chart of evaluation results of the embodiment. DETAILED DESCRIPTION

[0037] The present application will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings.

[0038] Figure 1 is the overall flowchart of a fuzzy comprehensive evaluation method for artificial electromagnetic environment construction effect based on SAR images and signals proposed by the present application, which includes an index selection module and a fuzzy comprehensive evaluation module.

[0039] Figure 2 is the flowchart of index selection, which includes image sample augmentation, image index calculation, data discretization, and attribute reduction.

[0040] Image sample augmentation: In order to perform image attribute reduction later, data of multiple image samples are needed. Since the input image is a single original image ( Figure 4 ) formed by SAR according to an actual target and a simulation image ( Figure 5 ) constructed by artificial electromagnetic environment, the number of image samples does not meet the quantity requirement of attribute reduction, so image sample augmentation is needed. The simulation image ( Figure 5 ) constructed by artificial electromagnetic environment is input into the SinGAN network, and after training, image sample data consistent with the style type of the input image sample and very similar but not identical in image content are generated. In order to calculate conveniently and ensure accuracy, the closest 9 image samples need to be manually selected from the augmented generated image samples, as shown in Figure 6 .

[0041] Image index calculation: In order to carry out the subsequent attribute reduction step, the data of each index of the image sample statistical characteristics and geometric characteristics need to be calculated. The indexes of statistical characteristics include: mean, variance, entropy, dynamic range, equivalent number of views, radiation resolution, average gradient of image gray scale, and the indexes of geometric characteristics include: area, perimeter, area-perimeter ratio, minimum circumscribed rectangle perimeter, minimum circumscribed rectangle area, shape parameter, circularity, aspect ratio, stretch length, sphericity, equivalent circle diameter of image target. The data of statistical characteristic indexes of 9 image samples are shown in Table 1, and the data of geometric characteristic indexes are shown in Table 2. Figure 7 Figure 8

[0042] Data discretization: In order to carry out the subsequent attribute reduction step, the calculated data of each index needs to be assigned to a discrete level according to the size. Set the number of discrete intervals to 4, first sort the same index data of 9 images in ascending order, calculate the difference value of adjacent data, select the end points of the largest 3 difference values as the interval breakpoints, and divide into 4 discrete intervals. Assign discrete levels from 1 to 4 to each interval in order of size. A total of 17 indexes of image statistical characteristics and geometric characteristics are discretized, and the discretized data of statistical characteristics are shown in Table 3, and the discretized data of geometric characteristics are shown in Table 4. Figure 9 Figure 10

[0043] Attribute reduction: The discretized data can be used as an information system in rough set theory. According to the concepts and methods introduced in the paper “Minimum attribute reduction algorithm based on 0-1 programming” by Zhan Wanrong and Yu Hai, the reduction problem is gradually transformed into a 0-1 programming problem, and the attribute reduction is realized by solving the problem. The specific steps are as follows: first, the discernibility matrix M is calculated; second, the minimum discernibility set MS is calculated; third, the matrix P is calculated from the minimum discernibility set MS; fourth, the 0-1 programming (formula 2) is solved to obtain a minimum attribute reduction, and the corresponding attributes are found to determine the final evaluation index. In this example, the vector of statistical characteristics after attribute reduction is [0 0 0 1 0 1 0], and in the reduction result vector of the optimal solution of statistical characteristics, the 4th and 6th elements are 1, and the other elements are 0. Therefore, the indexes of statistical characteristics after reduction are the dynamic range corresponding to the 4th element and the radiation resolution corresponding to the 6th element. Similarly, the vector of geometric characteristics after attribute reduction is [0 0 0 0 0 1 0 1 00], and the remaining indexes of geometric characteristics after attribute reduction are stretch length and minimum circumscribed rectangle area. Therefore, the output results of the index selection module are: the indexes of statistical characteristics are dynamic range and radiation resolution, and the indexes of geometric characteristics are stretch length and minimum circumscribed rectangle area.

[0044] ​​​​The fuzzy comprehensive evaluation module comprises determining selected indexes, determining an evaluation set, calculating a fuzzy relation matrix, determining a weight vector, fuzzy synthesis, and outputting an evaluation result.

[0045] Determining selected indexes: The evaluation indexes are divided into image index part and signal index part. The image aspect indexes are the results output by the index selection module, i.e., the statistical feature indexes and the geometric feature indexes after reduction. In this example, the statistical feature indexes are dynamic range and radiation resolution, and the geometric feature indexes are elongation and minimum circumscribed rectangle area. The signal feature indexes include: azimuth and range peak side lobe ratio, azimuth and range integrated side lobe ratio, azimuth and range spatial resolution.

[0046] Determining an evaluation set: The evaluation set is a set of final evaluation levels, used to represent the similarity degree of the SAR simulation image and the original image. The evaluation set is defined as five levels, i.e., {very good, better, medium, worse, very bad}.

[0047] Calculating a fuzzy relation matrix: In order to perform subsequent fuzzy synthesis, the fuzzy relation matrix of each feature needs to be calculated according to the membership function of each feature. The determination of the membership function can refer to previous experience. The statistical feature adopts the half-lower trapezoidal to calculate the membership degree, and the geometric feature and the signal feature adopt the triangle and the half-lower trapezoidal to calculate the membership degree. According to the method of He Fengbiao in the book "MATLAB Implementation of Fuzzy Comprehensive Method", the ratio of the statistical feature indexes, the geometric feature indexes, and the signal feature indexes of the original image and the simulation image are calculated respectively, and are brought into the corresponding membership function to obtain the fuzzy relation matrix of the statistical feature The fuzzy relation matrix of the geometric feature The fuzzy relation matrix of the signal feature

[0048] Determining a weight vector: In order to perform subsequent fuzzy synthesis, the weight vector of each feature needs to be calculated according to the judgment matrix of each feature. According to the method and steps of Yan Zhiqiang et al. in the paper "Research and Practice of Automatic Evaluation of 5G Cell Performance Based on AHP Two-level Fuzzy Comprehensive Evaluation Method", the 1-9 scale method is used to determine the relative importance of each index through pairwise comparison. The judgment matrix of the statistical feature indexes, the geometric feature indexes, and the signal feature indexes of the second layer is constructed and the judgment matrix of the statistical feature, the geometric feature, and the signal feature of the first layer The random consistency ratio CR of the judgment matrices A1, A2, A3 and A is less than 0.1, satisfying the consistency test. Then the eigenvectors W1 = [0.6 0.4], W2 = [0.6667 0.3333], W3 = [0.1667 0.1667 0.1667 0.1667 0.1667 0.1667] and W = [0.3333 0.3333 0.3333] are obtained for the respective judgment matrices. Each eigenvector is the eigenvector of the corresponding judgment matrix, so the second layer weight vectors are W1, W2 and W3, which correspond to the statistical features, geometric features and signal features respectively, and the first layer weight vector is W, which is the weight vector of the three features as a whole.

[0049] Fuzzy synthesis: In order to obtain the FCE result vector, the weight vectors of the statistical features, geometric features and signal features need to be combined with the fuzzy relation matrix corresponding to each feature to obtain the result vector. The fuzzy evaluation vector of the statistical features is B1 = W1·R1 = [0.6 0.2402 0 0 0.4], the fuzzy evaluation vector of the geometric features is B2 = W2·R2 = [1 0 0 0 0], the fuzzy evaluation vector of the signal features is B3 = W3·R3 = [0.9453 0.0547 0 0 0], and the fuzzy relation matrix of the first layer is: The first layer weight vector is W = [0.3333 0.3333 0.3333], and the FCE result vector is B = W·R = W·[B1, B2, B3] = [0.8484 0.0983 0 0 0.1333] (the fuzzy operator·is a dot product operation).

[0050] Output evaluation result: For the FCE result vector after fuzzy synthesis, the evaluation result level corresponding to the maximum membership degree in the evaluation set is output, that is, the level corresponding to the maximum value in the FCE result vector is selected as the final evaluation result level. In the FCE result vector B = [0.8484 0.0983 0 0 0.1333] of this example, the maximum value 0.8484 is located at the position of the first element, so the level "very good" corresponding to the maximum value in the FCE result vector is selected as the final evaluation result level, and the evaluation result bar chart is shown in Figure 11

[0051] It should be understood that although the present specification is described in terms of embodiments, each embodiment does not necessarily contain only one independent technical method, and the specification is described in this way only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined and implemented according to the understanding of those skilled in the art.

[0052] ​The above detailed description merely illustrates preferred non-limiting embodiments of the application, and it is not intended to limit the scope of the application. Rather, it is intended to cover all equivalents within the spirit and scope of the application.

Claims

1. A fuzzy comprehensive evaluation method for the construction effect of artificial electromagnetic environment based on SAR images and signals, characterized in that: The index selection module and the fuzzy comprehensive evaluation module are specifically implemented, The index selection module is used for screening indexes of the input original image and the to-be-evaluated simulation image, and eliminating redundant indexes, and includes image sample augmentation, image index calculation, data discretization and attribute reduction; the image sample augmentation is used for generating multiple image samples which are consistent with the style type of the to-be-evaluated simulation image and have very similar but not identical image contents; the image index calculation is used for calculating data of indexes of image sample statistical features and geometric features; the data discretization is used for subsequent attribute reduction steps, and each index data calculated is assigned to a discrete level according to the size; the attribute reduction takes the data processed by discretization as an information system in rough set theory, and uses a minimum attribute reduction algorithm to obtain screened indexes of statistical features and geometric features, and completes index selection. The fuzzy comprehensive evaluation module respectively constructs a judgment matrix of statistical features, a judgment matrix of geometric features and a judgment matrix of signal features according to the selected evaluation indexes, and constructs an overall judgment matrix of the three features, obtains a weight vector of each feature according to the judgment matrix, then determines a comment set, calculates a fuzzy relation matrix, and obtains a result vector through fuzzy synthesis, and finally obtains the final evaluation level according to the maximum membership degree principle. The fuzzy comprehensive evaluation module includes determining the selected indexes, determining the comment set, calculating the fuzzy relation matrix, determining the weight vector, fuzzy synthesis, and outputting the evaluation result. The comment set is a set of final evaluation levels, and is used to represent the similarity degree of the to-be-evaluated simulation image and the original image.

2. The method according to claim 1, wherein the method is characterized by: The sample augmentation is performed by using a SinGAN network.

3. The method according to claim 1, wherein the method is characterized by: The indexes of the statistical features include mean value, variance, entropy, dynamic range, equivalent number of views, radiation resolution and average gradient of image gray scale; and the indexes of the geometric features include area, perimeter, area-perimeter ratio, minimum circumscribed rectangle perimeter, minimum circumscribed rectangle area, shape parameter, circularity, length-width ratio, elongation, sphericity and equivalent circle diameter of an image target.

4. The method according to claim 1 or 3, wherein the method is characterized by: The selected indexes are used for determining evaluation indexes, and the evaluation indexes are divided into image index parts and signal index parts; the image indexes are the results output by the index selection module, that is, the reduced statistical feature indexes and geometric feature indexes. The indexes of the signal features include azimuth and distance peak side lobe ratio, azimuth and distance integral side lobe ratio, and azimuth and distance spatial resolution.

5. The method of claim 1, wherein the method is characterized by: The calculation process of the fuzzy relation matrix is as follows: the ratio of the original image to the to-be-evaluated simulation image on the statistical feature indexes, the geometric feature indexes and the signal feature indexes is calculated respectively, and the membership degree functions corresponding to the indexes are brought in to obtain the fuzzy relation matrix R1 of the statistical features, the fuzzy relation matrix R2 of the geometric features and the fuzzy relation matrix R3 of the signal features respectively.

6. The method of claim 5, wherein the method is characterized by: The statistical features adopt a descending half trapezoidal calculation membership degree, and the geometric features and the signal features adopt a triangular and a descending half trapezoidal calculation membership degree.

7. The method of claim 1, wherein the method is characterized by: The weight vector is determined according to the judgment matrix of each feature. The fuzzy comprehensive evaluation module adopts a two-level fuzzy comprehensive evaluation method. The first level is to distribute the weight among the three types of indexes, i.e., statistical features, geometric features and signal features. The second level is to distribute the weight among the indexes of each feature. Specifically, the relative importance of each index is determined by using the 1-9 scale method and a pairwise comparison method. The judgment matrix A1, A2 and A3 are constructed for the indexes of the statistical features, the indexes of the geometric features and the indexes of the signal features, respectively. The overall judgment matrix A is constructed for the three features, i.e., the statistical features, the geometric features and the signal features. The random consistency ratios CR of the judgment matrix A1, A2, A3 and A are calculated, respectively. Consistency test is performed. If CR<0.1, the test is not passed. The judgment matrix needs to be re-constructed and the weight vector needs to be re-determined until the consistency test is passed. After the consistency test, the eigenvectors W1, W2, W3 and W of the judgment matrices are calculated, respectively. The weight vector of each feature is the eigenvector of the corresponding judgment matrix. Therefore, the second-level weight vectors are W1, W2 and W3, which correspond to the indexes of the statistical features, the geometric features and the signal features, respectively. The first-level weight vector is W, which is the weight vector of the overall three features.

8. The fuzzy comprehensive evaluation method of the construction effect of the artificial electromagnetic environment based on the SAR image and signals according to claim 7, characterized in that: The fuzzy synthesis includes calculating the fuzzy evaluation vectors of the statistical features, the fuzzy evaluation vectors of the geometric features and the fuzzy evaluation vectors of the signal features, respectively, i.e., B1=W1·R1, B2=W2·R2 and B3=W3·R3. The fuzzy relation matrix of the first level is R=[B1,B2,B3]. The first-level weight vector is W. The FCE result vector is B=W·R=W·[B1,B2,B3].

9. The fuzzy comprehensive evaluation method of the construction effect of the artificial electromagnetic environment based on the SAR image and signals according to claim 8, characterized in that: The evaluation result is output. According to the maximum membership degree principle, the evaluation result level corresponding to the maximum value in the FCE result vector is selected as the final evaluation result level. ​

Images