A polarization calibration application performance evaluation method, system, device and medium

Abstract

The application relates to a polarimetric calibration application performance evaluation method, system, device and medium, which comprises the following steps: S1, acquiring a pair of polarimetric SAR images before and after polarimetric calibration to construct an experimental data pair; S2, extracting a plurality of polarimetric features from the two images in sequence, and constructing a corresponding polarimetric feature vector for each image; S3, constructing a target detection method based on joint optimization of features and detectors, and jointly optimizing the number of polarimetric feature vectors, a penalty parameter and a kernel function parameter to complete detection; S4, inputting the images before and after polarimetric calibration to be detected into the SVM detector based on joint optimization of features and detectors, outputting corresponding target detection results, comparing and evaluating the detection results, determining the polarimetric application performance of the polarimetric SAR image to be evaluated based on the target detection results, and realizing evaluation of the polarimetric calibration application performance. The application defines a new evaluation index detection performance, and realizes evaluation of the polarimetric calibration application performance from multiple angles.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method, system, device, and medium for evaluating the application performance of polarization calibration. Background Technology

[0002] Compared to single-polarization images, polarimetric SAR images possess richer phase information and higher-quality intensity information, demonstrating significant application potential. However, a large amount of "interference information" exists during the acquisition and application of polarimetric SAR images, including speckle multiplicative noise and polarization distortion caused by antenna transmission and reception. Therefore, polarimetric calibration of the polarimetric SAR system is necessary to obtain accurate polarization information. Polarimetric calibration is a prerequisite for the application of polarimetric SAR images in various fields, and its effectiveness greatly influences the application efficiency of polarimetric SAR images in target detection. However, current research lacks methods for evaluating the quality of polarimetric calibration and quantitatively assessing its subsequent application effects, and reliable indicators are lacking for evaluating the effectiveness of polarimetric calibration.

[0003] Therefore, there is an urgent need for a method, system, equipment, and medium for evaluating the performance of polarization calibration applications, which can quantify the performance evaluation of polarization calibration applications and achieve evaluation of the performance of polarization calibration applications from multiple perspectives. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method, system, equipment, and medium for evaluating the application performance of polarization calibration, thereby resolving the issues of unclear evaluation methods for the application performance of polarization calibration target detection and the difficulty in quantitatively evaluating the application performance of polarization calibration.

[0005] This invention relates to a method for evaluating the effectiveness of polarization calibration applications, comprising the following steps:

[0006] Step S1: Obtain polarimetric SAR image pairs before and after polarimetric calibration, and construct experimental data pairs;

[0007] Step S2: Extract multiple polarization features from the two images sequentially, and construct a corresponding polarization feature vector for each image;

[0008] Step S3: Construct an SVM detector based on joint optimization of feature-detector for target detection. Jointly optimize the three parameters involved in the detection method to complete the detection.

[0009] Step S4: Input the images before and after polarization calibration to be detected into the SVM detector jointly optimized by feature-detector, output the corresponding target detection results, compare and evaluate the detection results, determine the polarization application performance of the polarization SAR image to be evaluated based on the target detection results, and realize the evaluation of the polarization calibration application performance of the polarization SAR image.

[0010] Further, in step S1, the experimental data pair is Y = [y1, y2], where y1 is the image code before polarization calibration and y2 is the image code of the same scene after polarization calibration.

[0011] Further, in step S2, the polarization feature vectors are F_y1 = [f11, f12, ..., f1N] and F_y2 = [f21, f22, ..., f2N], where f11 represents the first feature in the polarization feature vector F_y1, f12 represents the second feature in the polarization feature vector F_y1, f1N represents the first and last feature in the polarization feature vector F_y1, f21 represents the first feature in the polarization feature vector F_y2, f22 represents the second feature in the polarization feature vector F_y2, f2N represents the last feature in the polarization feature vector F_y2, and N represents the total number of features.

[0012] Furthermore, in step S3, the three parameters involved in the detection method are: the number of polarization feature vectors n, the penalty parameter c, and the kernel function parameter g.

[0013] Further, step S3 specifically includes:

[0014] Step S31: Preprocess the image to be detected and label the sample in the image to be detected to adapt to the detection method using the SVM detector jointly optimized by feature-detector.

[0015] Step S32: Establish an evaluation function to adjust parameters based on the number of correctly detected pixels, including target pixels and background pixels;

[0016] Step S33: Determine the effective range of the number of polarization eigenvectors, penalty parameters, and kernel function parameters, and generate initial solutions n0, c0, and g0 for these three parameters in a random manner to initialize the parameters of the joint optimization method;

[0017] Step S34: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they meet the constraints, the initial solution is directly used as the parameters of the SVM detector jointly optimized by the feature-detector to detect the target and output the detection results. Step S3 ends and jumps directly to step S4.

[0018] Step S35: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they do not meet the constraints, search within the defined parameter range to generate new solutions n1, c1 and g1.

[0019] Step S36: Calculate the matching degree of the new solutions n1, c1, and g1. If the matching degree meets the preset threshold, the new solution is used for training. The matching degree is calculated using the normalized inner product of vectors.

[0020] Step S37: Calculate the fitness function for the training results of the new solution that meets the conditions, and enter the new solution into the next loop according to a certain probability p based on the matching degree calculation result; until the solution result meets the constraints, use the obtained solution that meets the conditions as the parameter of the feature-detector joint optimization SVM detector to detect the target, output the detection result, step S3 ends, and proceed to step S4.

[0021] Further, in step S32, the evaluation function is represented by Ev(x), which is used to measure the number of samples that are accurately detected in the training samples.

[0022] Further, step S4 specifically includes:

[0023] Step S41: Statistically analyze the target detection results before and after polarization calibration. The number of correctly detected targets in the SAR data before polarization calibration is N. tr1 The number of false alarm targets in the SAR data before polarization calibration is N. fa1 After polarization calibration, the number of correctly detected targets in the SAR data is N. tr2 After polarization calibration, the number of false alarm targets in the SAR data is N. fa2 The actual target quantity is N. ac ;

[0024] Step S42: Calculate the indicators of target detection results before and after polarization calibration, and calculate the detection rate P of target detection results before and after polarization calibration respectively. d1 =N tr1 / N ac P d2 =N tr2 / N ac False alarm rate P of target detection results before and after polarization calibration f1 =N fa1 / (N ac +N fa1 ), P f2 =N fa2 / (N ac +N fa2 The quality factor of target detection results before and after polarization calibration, FOM1 = N tr1 / (N fa1 +N ac ), FOM2 = N tr2 / (N fa2 +N ac );

[0025] Step S43: Evaluate the application performance of polarization calibration and define an evaluation index for the application performance of polarization calibration based on target detection, namely, detection performance E. det =FOM2 / FOM1.

[0026] The present invention also relates to a polarization calibration application performance evaluation system, wherein the system applies a polarization calibration application performance evaluation method.

[0027] The present invention also relates to a computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor executes the computer program to implement the steps of the above-described polarization calibration application performance evaluation method.

[0028] The present invention also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described polarization calibration application performance evaluation method.

[0029] The beneficial effects of this invention compared to the prior art

[0030] 1. A new evaluation index, detection efficiency E, was defined. det =FOM2 / FOM1 can be used to quantitatively evaluate the effectiveness of polarization calibration applications, filling the gap in quantitative indicators in the existing field of polarization calibration evaluation.

[0031] 2. This method can accurately evaluate the polarization components of interest, or it can accurately evaluate them after replacing the required polarization components. The method is highly flexible and not limited to specific polarization characteristics. It can evaluate the effectiveness of polarization calibration applications from multiple perspectives. Attached Figure Description

[0032] Figure 1a Image y1 is the image before polarization calibration in this invention;

[0033] Figure 1b Image y2 is the polarization-calibrated image of this invention;

[0034] Figure 2a This is a process diagram of the land-sea mask part of the data preprocessing in this invention;

[0035] Figure 2b This is a schematic diagram illustrating the data annotation situation of the present invention;

[0036] Figure 3a This is a diagram showing the target detection results before polarization calibration in this invention;

[0037] Figure 3b This is a diagram showing the target detection results after polarization calibration according to the present invention. Detailed Implementation

[0038] The following describes this embodiment in detail with reference to Figures 1 to 3.

[0039] The following is a detailed description of the implementation process for evaluating the application effectiveness of this method using fully polarimetric images of a pair of RadarSat2 satellites before and after polarimetric calibration in the Qingdao area, including the following steps:

[0040] Step S1: Obtain polarimetric SAR image pairs before and after polarimetric calibration, and construct experimental data pairs Y = [y1, y2], where y1 is the image code before polarimetric calibration, corresponding to... Figure 1a y2 is the image code for the same scene after polarization calibration, corresponding to Figure 1b .

[0041] Step S2: Extract multiple polarization features from the two images sequentially, and construct corresponding polarization feature vectors for each image as F_y1=[f11,f12,……,f1N] and F_y2=[f21,f22,……,f2N], where f11 represents the first feature in polarization feature vector F_y1, f12 represents the second feature in polarization feature vector F_y1, f1N represents the first and last feature in polarization feature vector F_y1, f21 represents the first feature in polarization feature vector F_y2, f22 represents the second feature in polarization feature vector F_y2, and f2N represents the last feature in polarization feature vector F_y2; N represents the total number of features.

[0042] This embodiment describes 10 types of polarization features, that is, the total number of features is N. The specific value of N and the polarization features it represents can be set as needed, and are not limited to the 10 types used in this part.

[0043] The 10 types of polarization features include: Freeman odd-order scattering polarization features, Freeman even-order scattering polarization features, Freeman volume scattering polarization features, Yamaguchi decomposition odd-order scattering polarization features, Yamaguchi decomposition even-order scattering polarization features, Yamaguchi decomposition volume scattering polarization features, Yamaguchi decomposition rotational scattering polarization features, scattering entropy H in HAalpha decomposition, target anisotropy A, and average scattering angle alpha.

[0044] Construct the polarization feature vectors corresponding to the polarization SAR images before and after polarization calibration:

[0045] F_y1=[Freeman_odd1,Freeman_dbl1,Freeman_vol1,Yamaguchi_odd1,Yamaguchi_dbl1,Yamaguchi_vol1,Yamaguchi_hlx1,H1,A1,alpha1];

[0046] F_y2=[Freeman_odd2, Freeman_dbl2, Freeman_vol2, Yamaguchi_odd2, Yamaguchi_dbl2, Yamaguchi_vol2, Yamaguchi_hlx2, H2, A2, alpha2].

[0047] Step S3: Construct an SVM detector based on joint optimization of features and detectors for target detection. Jointly optimize the number of polarization feature vectors n, the penalty parameter c, and the kernel function parameter g to complete the detection.

[0048] Further, step S3 specifically includes:

[0049] Step S31: Preprocess the image to be detected to remove background factors that have a significant impact. This part uses a land-sea mask method, and the template of the land-sea mask is as follows: Figure 2a As shown, the white area represents the region to be detected, and the black area represents the interference region. Sample annotation is performed on the image to be detected to adapt to the detection method using an SVM detector jointly optimized by features and detectors. The annotated samples are as follows: Figure 2b As shown, the target pixel area is circled with an ellipse, and the background pixel area is circled with a rectangle.

[0050] Step S32: Establish an evaluation function Ev(x), where the evaluation function measures the number of correctly detected pixels in the training samples. To fine-tune the parameters, the number of correctly detected pixels is used as a basis, including both target and background pixels.

[0051] Step S33: Determine the effective range of the number of polarization eigenvectors, penalty parameters, and kernel function parameters, and generate initial solutions n0, c0, and g0 for these three parameters in a random manner to initialize the parameters of the joint optimization method; in this embodiment, the value of n is 10, and its solution value is required to be a positive integer between 1 and 10.

[0052] Step S34: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they meet the constraints, the initial solution is directly used as the parameters of the SVM detector jointly optimized by the feature-detector to detect the target and output the detection results. Step S3 ends and jumps directly to step S4.

[0053] Step S35: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they do not meet the constraints, search within the defined parameter range to generate new solutions n1, c1 and g1.

[0054] Step S36: Calculate the matching degree of the new solutions n1, c1, and g1. If the matching degree meets the preset threshold, the new solution is used for training. The matching degree is calculated using the normalized inner product of vectors.

[0055] Step S37: Calculate the fitness function for the training results of the new solution that meets the conditions, and enter the new solution into the next loop according to a certain probability p based on the matching degree calculation result; until the solution result meets the constraints, use the obtained solution that meets the conditions as the parameter of the feature-detector joint optimization SVM detector to detect the target, output the detection result, step S3 ends, and proceed to step S4.

[0056] Step S4: Input the images before and after polarization calibration to be detected into the SVM detector jointly optimized by feature-detector, output the corresponding target detection results, compare and evaluate the detection results, determine the polarization application performance of the polarization SAR image to be evaluated based on the target detection results, and realize the evaluation of the polarization calibration application performance of the polarization SAR image.

[0057] A portion of the detection results before and after polarization calibration are selected and presented as follows: Figure 3a and 3b As shown in the figure, it is clear that the target detection performance is significantly better before and after polarization calibration.

[0058] Further, step S4 specifically includes:

[0059] Step S41: Statistically analyze the target detection results before and after polarization calibration. The number of correctly detected targets in the SAR data before polarization calibration is N. tr1 The number of false alarm targets in the SAR data before polarization calibration is N. fa1 After polarization calibration, the number of correctly detected targets in the SAR data is N. tr2 After polarization calibration, the number of false alarm targets in the SAR data is N. fa2 The actual target quantity is N. ac ;

[0060] Statistical analysis of the detection results shows that the actual number of targets was 47. Before polarization calibration, the number of correctly detected targets by SAR data was 35, and the number of incorrectly detected targets was 1. After polarization calibration, the number of correctly detected targets by SAR data was 42, and the number of incorrectly detected targets was 2.

[0061] Step S42: Calculate the indicators of target detection results before and after polarization calibration, and calculate the detection rate P of target detection results before and after polarization calibration respectively. d1 =N tr1 / N ac P d2 =N tr2 / N acFalse alarm rate P of target detection results before and after polarization calibration f1 =N fa1 / (N ac +N fa1 ), P f2 =N fa2 / (N ac +N fa2 The quality factor of target detection results before and after polarization calibration, FOM1 = N tr1 / (N fa1 +N ac ), FOM2 = N tr2 / (N fa2 +N ac );

[0062] Calculations show that before polarization calibration, the target detection rate was 74.47%, the false alarm rate was 2.08%, and the quality factor was 72.92%. After polarization calibration, the target detection rate was 89.36%, the false alarm rate was 4.08%, and the quality factor was 85.71%. The quality factor, as a comprehensive evaluation indicator, shows that the application performance is better after polarization calibration.

[0063] Step S43: Evaluate the application performance of polarization calibration and define an evaluation index for the application performance of polarization calibration based on target detection, namely, detection performance E. det =FOM2 / FOM1, E det The larger the value, the better the target detection application performance of polarization calibration.

[0064] The detection efficiency value is 1.17. After polarization calibration, the target detection efficiency is increased to 1.17 times the original value, which fully demonstrates that the polarization calibration has a significant improvement on target detection. The application efficiency results are well evaluated.

[0065] The present invention also relates to a polarization calibration application performance evaluation system, wherein the system applies a polarization calibration application performance evaluation method.

[0066] The present invention also relates to a computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor executes the computer program to implement the steps of the above-described polarization calibration application performance evaluation method.

[0067] The present invention also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described polarization calibration application performance evaluation method.

[0068] The above description of the present invention is only a preferred embodiment of the present invention and is not intended to limit the implementation of the present invention. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection claimed in the claims.

Claims

1. A method for evaluating the application effectiveness of polarization calibration, characterized in that, Includes the following steps: Step S1: Obtain polarimetric SAR image pairs before and after polarimetric calibration, and construct experimental data pairs; Step S2: Extract multiple polarization features from the two images sequentially, and construct a corresponding polarization feature vector for each image; Step S3: Construct an SVM detector based on joint optimization of feature-detector for target detection. Jointly optimize the three parameters involved in the detection method to complete the detection. Step S4: Input the images before and after polarization calibration to be detected into the SVM detector jointly optimized by feature-detector, output the corresponding target detection results, compare and evaluate the detection results, determine the polarization application performance of the polarization SAR image to be evaluated based on the target detection results, and realize the evaluation of the polarization calibration application performance of the polarization SAR image. Step S4 is as follows: Step S41: Statistically analyze the target detection results before and after polarization calibration. The number of correctly detected targets in the SAR data before polarization calibration is N. tr1 The number of false alarm targets in the SAR data before polarization calibration is N. fa1 After polarization calibration, the number of correctly detected targets in the SAR data is N. tr2 After polarization calibration, the number of false alarm targets in the SAR data is N. fa2 The actual target quantity is N. ac ; Step S42: Calculate the indicators of target detection results before and after polarization calibration, and calculate the detection rate P of target detection results before and after polarization calibration respectively. d1 = N tr1 / N ac P d2 = N tr2 / N ac False alarm rate P of target detection results before and after polarization calibration f1 = N fa1 / (N ac + N fa1 ), P f2 = N fa2 / (N ac + N fa2 FOM1 = N, the quality factor of target detection results before and after polarization calibration. tr1 / (N fa1 + N ac FOM2 = N tr2 / (N fa2 + N ac ); Step S43: Evaluate the application performance of polarization calibration and define an evaluation index for the application performance of polarization calibration based on target detection, namely, detection performance E. det = FOM2 / FOM1.

2. The polarization calibration application performance evaluation method according to claim 1, characterized in that, In step S1, the experimental data pair is Y=[y1,y2], where y1 is the image code before polarization calibration and y2 is the image code of the same scene after polarization calibration.

3. The polarization calibration application performance evaluation method according to claim 1, characterized in that, In step S2, the polarization feature vectors are F_y1=[f11,f12,……,f1N] and F_y2=[f21,f22,……,f2N], where f11 represents the first feature in polarization feature vector F_y1, f12 represents the second feature in polarization feature vector F_y1, f1N represents the first and last feature in polarization feature vector F_y1, f21 represents the first feature in polarization feature vector F_y2, f22 represents the second feature in polarization feature vector F_y2, f2N represents the last feature in polarization feature vector F_y2, and N represents the total number of features.

4. The polarization calibration application performance evaluation method according to claim 1, characterized in that, In step S3, the three parameters involved in the detection method are: the number of polarization feature vectors n, the penalty parameter c, and the kernel function parameter g.

5. The polarization calibration application performance evaluation method according to claim 1, characterized in that, Step S3 is as follows: Step S31: Preprocess the image to be detected and label the sample in the image to be detected to adapt to the detection method using the SVM detector jointly optimized by feature-detector. Step S32: Establish an evaluation function to adjust parameters based on the number of correctly detected pixels, including target pixels and background pixels; Step S33: Determine the effective range of the number of polarization eigenvectors, penalty parameters, and kernel function parameters, and generate initial solutions n0, c0, and g0 for these three parameters in a random manner to initialize the parameters of the joint optimization method; Step S34: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they meet the constraints, the initial solution is directly used as the parameters of the SVM detector jointly optimized by the feature-detector to detect the target and output the detection results. Step S3 ends and jumps directly to step S4. Step S35: Check whether the initial solution of the parameters meets the constraints. The constraints are determined by the evaluation function value. If they do not meet the constraints, search within the defined parameter range to generate new solutions n1, c1 and g1. Step S36: Calculate the matching degree of the new solutions n1, c1, and g1. If the matching degree meets the preset threshold, the new solution is used for training. The matching degree is calculated using the normalized inner product of vectors. Step S37: Calculate the fitness function for the training results of the new solution that meets the conditions, and enter the new solution into the next loop according to a certain probability p based on the matching degree calculation result; until the solution result meets the constraints, use the obtained solution that meets the conditions as the parameter of the feature-detector joint optimization SVM detector to detect the target, output the detection result, step S3 ends, and proceed to step S4.

6. The polarization calibration application performance evaluation method according to claim 5, characterized in that, In step S32, the evaluation function is represented by Ev(x), which is used to measure the number of samples that are accurately detected in the training samples.

7. A polarization calibration application performance evaluation system, characterized in that, The method described in any one of claims 1 to 6 above.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images