Method and system for evaluating filter performance of phase filter model
By constructing an evaluation method of test data sets and phase filtering models, the lack of the entire process of InSAR phase filtering evaluation in the prior art is solved, and the effectiveness comparison of different phase filtering methods is achieved, and the accuracy and efficiency of the evaluation are improved.
Patent Information
- Application Number
- CN202410942665.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-15
- Publication Date
- 2024-11-12
AI Technical Summary
The lack of effective evaluation scheme for the entire InSAR phase filtering process in the prior art makes it difficult to compare the effectiveness of different phase filtering methods under the same data set.
The test data set is constructed, including noise interference phase maps and noise-free interference phase maps, filtered through the phase filtering model, and filter performance is determined using preset evaluation rules, including phase simulation modules, phase filtering modules and filter performance evaluation modules.
Qualitative and quantitative evaluation of different phase filtering methods is realized, the gap in the existing technology is filled, and the effectiveness of different phase filtering methods can be better compared.
Smart Images

Figure CN118938216A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of radar interferometry technology, and in particular to a method, system, electronic device, non-transitory computer-readable storage medium, and computer program product for evaluating filtering performance of a phase filtering model. Background Art
[0002] Synthetic aperture radar (InSAR) interferometry has become one of the commonly used earth observation technologies due to its advantages of all-weather, all-day, wide monitoring range and high accuracy. This technology obtains the interference phase by differential interference of SAR images taken twice at different times in the same area, and further derives the surface elevation or deformation data using the interference phase.
[0003] The steps of processing InSAR data include image registration, interferogram generation, phase filtering, ground flattening and phase unwrapping. The purpose of these processing steps is to eliminate interference factors, improve the quality and accuracy of the interferogram, so as to make the subsequent surface deformation monitoring results more accurate. Among them, the quality of the interferogram is a key factor affecting the accuracy of InSAR technology. However, due to the influence of system thermal noise, sensor imaging geometric characteristics, time and space baseline decorrelation and other factors in the InSAR interferometric measurement process, the interferogram has complex noise, which affects the subsequent interferometric measurement results. Therefore, in the process of interferometric processing, using phase filtering method to remove the noise phase of the interferogram is a key step.
[0004] Currently, in the existing technologies of InSAR phase filtering, there are many filtering methods that can effectively suppress phase noise, including spatial domain filtering methods, transform domain filtering methods, and filtering methods based on deep learning. However, the existing technologies generally only focus on phase filtering methods, and there is no solution to effectively evaluate phase filtering methods from the perspective of the entire process of InSAR phase filtering. Summary of the invention
[0005] The present disclosure provides a method, system, electronic device, non-transitory computer-readable storage medium and computer program product for evaluating the filtering performance of a phase filtering model, so as to address the deficiencies in the prior art.
[0006] The present disclosure provides a method for evaluating filtering performance of a phase filtering model, wherein the phase filtering model is used to filter an interferometric synthetic aperture radar phase map, and the method comprises: Constructing a test data set for evaluating the phase filtering model; wherein the test data set includes at least one set of test data pairs, each of the at least one set of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; Inputting the noise interference phase image in each group of test data pairs into a preset phase filtering model, and outputting a corresponding filtered interference phase image via the preset phase filtering model; Based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined using preset filtering performance evaluation rules.
[0007] The present disclosure also provides a system for evaluating filtering performance of a phase filtering model, wherein the phase filtering model is used to filter an interferometric synthetic aperture radar phase map, and the device comprises: A phase simulation module is configured to: construct a test data set for evaluating the phase filtering model; wherein the test data set includes at least one set of test data pairs, each of the at least one set of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; The phase filtering module is configured to: input the noise interference phase image in each group of test data pairs into a preset phase filtering model, and output a corresponding filtered interference phase image through the preset phase filtering model; The filtering performance evaluation module is configured to determine the filtering performance of the preset phase filtering model based on the noise interference phase diagram, the filtered interference phase diagram and the corresponding noise-free interference phase diagram using preset filtering performance evaluation rules.
[0008] The present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for evaluating the filtering performance of a phase filtering model as described in any one of the above is implemented.
[0009] The present disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for evaluating the filtering performance of a phase filtering model as described in any one of the above is implemented.
[0010] The present disclosure also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above methods for evaluating the filtering performance of a phase filtering model.
[0011] As described above, the present disclosure proposes a method for evaluating the filtering performance of a phase filtering model (each phase filtering model corresponds to an InSAR phase filtering method) from the perspective of the entire process of InSAR phase filtering. Specifically, the method first produces a test (simulation) data set to provide test data for evaluating the effects of different phase filtering methods; then, the test data set is applied with different phase filtering methods respectively, and the corresponding filtering results are saved; finally, different filtering results are qualitatively and quantitatively evaluated, so that the effectiveness of different phase filtering methods can be better compared under the same data set, filling the gap in the prior art in solving the problem of evaluating the filtering capabilities of different phase filtering methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] In order to more clearly illustrate the technical solutions in the present disclosure or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0013] Figure 1 It is one of the flowcharts of the method for evaluating the filtering performance of the phase filtering model provided by the embodiment of the present disclosure; Figure 2 This is a second flow chart of a method for evaluating filtering performance of a phase filtering model provided by an embodiment of the present disclosure; Figure 3 is a schematic diagram of the DnCNN model structure provided in an embodiment of the present disclosure; Figure 4 is a schematic diagram of the improved CNN model structure provided by an embodiment of the present disclosure; Figure 5 is a schematic diagram of the residual block structure provided by an embodiment of the present disclosure; Figure 6 This is a third flow chart of a method for evaluating filtering performance of a phase filtering model provided by an embodiment of the present disclosure; Figure 7a is the noise interference phase diagram and the corresponding 128th row cross-sectional diagram provided by an embodiment of the present disclosure; Figure 7b is the noise-free interference phase diagram provided by the embodiment of the present disclosure and the corresponding 128th row cross-sectional diagram; Figure 7c is the filtered interference phase diagram (corresponding to Lee filtering) and the corresponding 128th row cross-sectional diagram provided by an embodiment of the present disclosure; Figure 7dis the filtered interference phase diagram (corresponding to Goldstein filtering) provided in an embodiment of the present disclosure and the corresponding 128th row cross-sectional diagram; Figure 7e It is the filtered interference phase image (corresponding to DnCNN filtering) provided by the embodiment of the present disclosure and the corresponding 128th row cross-sectional image; Figure 7f It is the filtered interference phase image (corresponding to the improved CNN filtering) and the corresponding 128th row cross-sectional image provided by the embodiment of the present disclosure; Figure 8 is a schematic diagram for comparing the number of phase residual points provided by an embodiment of the present disclosure; Fig. 9 is a schematic diagram of root mean square error comparison provided by an embodiment of the present disclosure; Fig.10 is a structural schematic diagram of a system for evaluating filtering performance of a phase filtering model provided by an embodiment of the present disclosure; Fig.11 It is a schematic diagram of the structure of an electronic device provided by the present disclosure. DETAILED DESCRIPTION
[0014] In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the technical solutions in the present disclosure will be clearly and completely described below in conjunction with the drawings in the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.
[0015] Summary of the invention concept: In order to evaluate the effectiveness of the InSAR phase filtering method more quickly and efficiently, it is necessary to conduct research from the perspective of the entire InSAR phase filtering process.
[0016] In view of this, the inventors of the present disclosure will study a solution from the perspective of the entire process of InSAR phase filtering. The solution may include a phase simulation module, a phase filtering module, and a filtering evaluation module, which are respectively used to produce simulation data, perform phase filtering on test data, and evaluate filtered data, so as to provide an effective means for comprehensively evaluating the performance of various InSAR phase filtering methods, and the solution can be further promoted to practical applications. The specific implementation details of the solution are shown in the following embodiments.
[0017] Example: The following describes the implementation details of the solution for evaluating the filtering performance of the phase filtering model disclosed in the present invention in conjunction with the accompanying drawings.
[0018] Figure 11 is a flow chart of a method for evaluating the filtering performance of a phase filtering model provided by an exemplary embodiment of the present disclosure. This embodiment can be applied to electronic devices (for example, servers or cloud computing platforms), such as Figure 1 As shown, the method for evaluating the filtering performance of a phase filtering model (the phase filtering model is used to filter an interferometric synthetic aperture radar phase image) comprises the following steps: S110, constructing a test data set for evaluating the phase filtering model.
[0019] The test data set includes at least one set of test data pairs, each of the at least one set of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image.
[0020] S120, inputting the noise interference phase image in each group of test data pairs into a preset phase filtering model, and outputting a corresponding filtered interference phase image via the preset phase filtering model.
[0021] S130. Based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined by using preset filtering performance evaluation rules.
[0022] As described above, the present disclosure proposes a method for evaluating the filtering performance of a phase filtering model (each phase filtering model corresponds to an InSAR phase filtering method) from the perspective of the entire process of InSAR phase filtering. Specifically, the method first produces a test (simulation) data set to provide test data for evaluating the effects of different phase filtering methods; then, the test data set is applied with different phase filtering methods respectively, and the corresponding filtering results are saved; finally, different filtering results are qualitatively and quantitatively evaluated, so that the effectiveness of different phase filtering methods can be better compared under the same data set, filling the gap in the prior art in solving the problem of evaluating the filtering capabilities of different phase filtering methods.
[0023] exist Figure 1 Based on the embodiment, as an optional implementation method, refer to Figure 2 Step S110 "constructing a test data set for evaluating the phase filter model" may include the following steps: S1110 , obtaining a sample digital elevation model of a selected sample area and satellite parameters associated with determining the sample digital elevation model.
[0024] Among them, Digital Elevation Model (DEM) is a technology that realizes digital simulation of ground terrain through limited terrain elevation data, that is, digital expression of terrain surface morphology.
[0025] The present disclosure does not limit the selection of the sample area, and the sample area can be selected according to actual needs.
[0026] The satellite parameters may include, for example, the main satellite altitude, satellite baseline length, slant range, ground point altitude, the angle between two satellite baselines and the horizontal direction, wavelength, incident angle, and vertical baseline.
[0027] The present disclosure does not limit the method of "obtaining". For example, the "sample digital elevation model and satellite parameters associated with determining the sample digital elevation model" may be stored in a data storage server, and the electronic device (e.g., a server or a cloud computing platform) that is the subject of step S1110 may communicate with the data storage server to obtain the "sample digital elevation model and satellite parameters associated with determining the sample digital elevation model".
[0028] S1120, based on the sample digital elevation model and the satellite parameters, using preset interference phase simulation rules, generate multiple noise-free interference phase images.
[0029] As an optional example, the multiple noise-free interference phase images may be determined using the following calculation formula (1); (1) in, represents the noise-free interferometric phase; H Indicates the main satellite altitude; B Indicates the spatial baseline between the primary satellite and the secondary satellite; r 0 represents the slant distance between the primary satellite and the ground point; h Indicates the elevation of a ground point; represents the angle between the spatial baseline and the horizontal direction; Indicates the radar wavelength.
[0030] Specifically, using the sample digital elevation model and the satellite parameters in a certain sample area, multiple sets of different or related noise-free interferometric phase , and then based on the multiple sets of noise-free interference phase , multiple noise-free interferometric phase images can be constructed.
[0031] Alternatively, as another optional example, the multiple noise-free interference phase images may be determined using the following calculation formula (2); (2) in, Represents the angle between the slant range of the main satellite relative to the ground point and the ground normal; represents the vertical baseline component obtained by separating the spatial baseline at the slant distance.
[0032] S1130: Based on the multiple noise-free interference phase images, determine the phase noise corresponding to each noise-free interference phase image using interference phase noise simulation rules.
[0033] As an optional example, first, based on each of the noise-free interference phase images, the probability density function of the corresponding phase noise may be determined using the following calculation formula (3).
[0034] (3) in, represents the first i phases; express The mathematical expectation of represents the i-th coherence corresponding to the noise-free interferometric phase pattern.
[0035] The phase noise may then be determined based on the probability density function.
[0036] Optionally, the noise phase may be generated according to the interference noise characteristic model by combining the simulated interference phase (ie, the noise-free interference phase diagram described above), the probability density function, and the simulated coherence diagram.
[0037] S1140, adding each of the noise-free interference phase images to the corresponding phase noise to obtain a plurality of the noise interference phase images.
[0038] As an optional example, the noise interference phase map may be calculated using the following calculation formula (4); (4) in, represents the noise phase, represents the noise-free interferometric phase; represents the noise interference phase.
[0039] Specifically, the noise interference phase diagram can be obtained by adding the noise-free interference phase in each noise-free interference phase diagram to the corresponding phase noise using calculation formula (4).
[0040] S1150, obtaining the test data set based on the multiple noise-free interference phase images and the multiple noise interference phase images.
[0041] It should be noted that, considering the current lack of public noise-free phase data sets and noisy phase data sets for InSAR, in order to enable the disclosed embodiment to have a reliable test data set, the test data set for evaluating the phase filtering model is constructed using the implementation method of step S110.
[0042] exist Figure 1 , Figure 2 Based on the embodiment, as an optional implementation mode, the preset phase filtering model includes a spatial filtering model, a frequency domain filtering model and a deep learning filtering model.
[0043] As an optional example, the spatial domain filtering model may be a filtering algorithm directly applied to the noise interference phase image; for example, it may include but is not limited to the Lee filtering algorithm.
[0044] Specifically, the Lee filter is a filtering algorithm based on a directional window. This method uses 16 windows (9×9 pixels) with different fringe directions, selects the window closest to the interference fringe direction, and then filters the interference phase within this window, so as to maintain the continuity of the phase fringe information as much as possible.
[0045] As another optional example, the frequency domain filtering model can implement a frequency domain filtering algorithm, which is particularly suitable for processing dense stripe areas; for example, it may include but is not limited to the Goldstein filtering algorithm. The essence of the Goldstein filtering algorithm is to perform weighted processing on the two-dimensional spectrum of the interference phase based on the characteristics that the useful signal power spectrum in the interference phase is concentrated in the low-frequency part, while the noise phase is randomly distributed, so as to remove the noise phase from the frequency domain and obtain the interference phase after noise reduction.
[0046] As another optional example, the deep learning filtering model includes a first convolutional neural network model and a second convolutional neural network model.
[0047] Optionally, the first convolutional neural network model may be a DnCNN model, and the second convolutional neural network model may be an improved CNN model.
[0048] Reference Figure 3 , the DnCNN model may include a cascaded first input unit L1, multiple first feature learning units (L2~L16) and a first output unit L17.
[0049] Among them, the first input unit includes a cascaded convolution layer Conv and an activation function layer ReLU, each of the first feature learning units includes a cascaded convolution layer, an activation function layer ReLU and a batch normalization layer BN, and the first output unit includes a cascaded convolution layer.
[0050] It should be noted that the DnCNN model is an end-to-end single convolutional neural network model, which is modified based on VGG. Its advantage is that it can directly learn from the input and map to the output, and can automatically learn complex features in the task. Figure 3 As shown in the figure, the DnCNN model can include 15 convolution blocks, L1 is the input layer, L2~L16 are feature learning layers, and each convolution block includes a convolution layer Conv, a ReLU activation layer, and a batch normalization layer BN. The batch normalization layer BN added to each convolution block can improve the stability of the training process, accelerate convergence, and reduce the risk of overfitting; the convolution kernel size is 3×3, and the number of input and output channels between each neuron is kept consistent. Finally, the convolution layer is used to output the filtered interference phase.
[0051] Reference Figure 4 and Figure 5 The improved CNN model may include a cascaded second input unit L1, multiple second feature learning units (L2~L7), one of the first feature learning units L8 and a second output unit L9.
[0052] Among them, the second input unit includes a cascaded convolution layer Conv, an activation function layer ReLU and a batch normalization layer BN, and each of the second feature learning units includes a cascaded residual block structure, a convolution layer Conv, an activation function layer ReLU and a batch normalization layer BN; the second output unit includes a cascaded convolution layer and an activation layer Sigmoid.
[0053] It should be noted that, referring to Figure 4 , the residual block structure L2 and the convolution block L3 (including the convolution layer Conv, the activation function layer ReLU and the batch normalization layer BN) constitute a second feature learning unit; similarly, the residual block structure L4 and the convolution block L5 (including the convolution layer Conv, the activation function layer ReLU and the batch normalization layer BN) constitute a second feature learning unit; the residual block structure L6 and the convolution block L7 (including the convolution layer Conv, the activation function layer ReLU and the batch normalization layer BN) constitute a second feature learning unit.
[0054] Among them, the first feature learning unit L8 can refer to Figure 3 The content of the example is not repeated here.
[0055] Among them, refer to Figure 5 , each of the residual block structures includes a cascaded input terminal Input(x), two first convolution blocks and an output terminal Output(x), and the input terminal is jump-connected with the outputs of the two first convolution blocks ( Figure 5Each of the first convolution blocks includes a cascaded convolution layer Conv, an activation function layer ReLU and a batch normalization layer BN.
[0056] The convolution kernel size of the convolution layer is 3×3.
[0057] It needs to be explained that an improved convolutional neural network model, namely the CNN model, is designed based on the DnCNN model structure. Figure 4 As shown in Figure 2, the CNN model retains the main structure of the DnCNN model, increases the depth of some convolutional blocks while reducing the number of convolutional blocks, and adds a residual block structure to improve the denoising performance. The residual block structure is as follows: Figure 5 As shown in the figure. In the improved CNN model, L1 is the input layer, L2~L8 are feature learning layers, and each convolution block contains a convolution layer, a ReLU activation layer, and a batch normalization layer. The ReLU activation layer and the batch normalization layer are used to ensure numerical stability and accelerate convergence during model training. The convolution kernel size is set to 3, and the number of input and output channels of each neuron is kept consistent. Each residual block also contains a convolution layer, a ReLU activation layer, and a batch normalization layer. At the same time, a skip connection is introduced to realize a residual connection. Five convolution blocks and three residual blocks are set to fully learn the overall and local features of the image. Finally, the L9 output layer uses the convolution layer and the Sigmoid activation layer to output the filtered component to obtain the filtered interference phase. The network also uses the MSE loss function for reverse gradient propagation to update the network parameters.
[0058] Based on the above embodiment, as an optional implementation mode, refer to Figure 6 Step S130 "determine the filtering performance of the preset phase filtering model based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image using a preset filtering performance evaluation rule" may include the following steps: S1310. Based on the noise interference phase image and the corresponding filtering interference phase image, a first performance indicator of the preset phase filtering model is determined using a preset qualitative evaluation rule.
[0059] As an optional example, a preset filtering effect identification method can be used to perform noise identification on the noise interference phase image to obtain a first noise identification result; a preset filtering effect identification method can be used to perform noise identification on the filtered interference phase image to obtain a second noise identification result; and based on the difference between the first noise identification result and the second noise identification result, a noise retention index is determined.
[0060] As another optional example, a first cross-sectional image of a preset position in the noise interference phase image can be obtained; a second cross-sectional image of the preset position in the filtered interference phase image can be obtained; a profile line fluctuation index can be determined based on the difference between the first cross-sectional image and the second cross-sectional image; and the first performance index can be determined based on the noise retention index and the profile line fluctuation index.
[0061] It should be noted that the qualitative assessment or evaluation can be based on the analysis of the filtering effect diagram and the profile diagram to judge the quality of the filtering effect.
[0062] Specifically, for the “Lee filter, Goldstein filter, DnCNN model, and improved CNN model” mentioned in the above embodiments, the corresponding first performance indicators can be determined using the above S1310 association examples. Figure 7a~7f , select a set of test data pairs (including a noise interference phase map and a noise-free interference phase map) from the test data set, input the noise interference phase map into Lee filtering, Goldstein filtering, DnCNN model and improved CNN model respectively to perform filtering, and obtain the filtered interference phase map output by each model. Among them, the filtered interference phase map output by Lee filtering is recorded as filtered Figure 1 , the filtered interference phase map output by Goldstein filtering is recorded as filtered Figure 2 , the filtered interference phase map output by the DnCNN model is recorded as filtered Figure 3 , the filtered interference phase map output by the improved CNN model is recorded as filtered Figure 4 The noise interference phase diagram, the noise-free interference phase diagram, and the filtering Figure 1 ~Filter Figure 4 Then, the noise interference phase diagram, the noise-free interference phase diagram and the filtered phase diagram are compared. Figure 1 ~Filter Figure 4 The filtering effect, that is, the "noise recognition" in the above example, determines the filtering Figure 1 ~Filter Figure 4 Then, the noise interference phase diagram, the noise-free interference phase diagram and the filtered phase diagram are compared. Figure 1 ~Filter Figure 4 The profile of the 128th row of data determines the profile fluctuation index of the aforementioned example.
[0063] Finally, a comprehensive analysis of the filtered interference phase image and the cross-sectional image corresponding to the 128th row of data revealed that the noise level in the noise interference phase image was high, resulting in significant fluctuations in the cross-sectional line in the corresponding cross-sectional image. Although Lee filtering can partially remove the noise in the noise interference phase image, the cross-sectional line in the corresponding cross-sectional image still shows significant fluctuations, that is, the filtering Figure 1The Goldstein filter can effectively remove most of the noise in the noise interference phase diagram, but the filter Figure 2 Some residual noise can still be observed in the image. After filtering by the DnCNN model and the improved CNN model, the profile lines in the corresponding profiles are closer to the profile lines corresponding to the noise-free interferometric phase image, and the overall performance is smoother and more stable. However, from the perspective of details, the profile lines in the corresponding profiles after filtering by the improved CNN model are more consistent with the profile lines corresponding to the noise-free interferometric phase image.
[0064] S1320. Based on the filtered interference phase image and the corresponding noise-free interference phase image, a second performance indicator of the preset phase filtering model is determined using a preset quantitative evaluation rule.
[0065] As an optional example, S1320 may include the following steps: 1) Based on the filtered interferometric phase image, the number of phase residual points in the filtered interferometric phase image is calculated using a preset phase residual point calculation rule as a residual indicator.
[0066] Among them, the number of residual points is an important indicator for evaluating the noise suppression ability of the filtering algorithm. The more phase residual points there are, the worse the noise suppression ability of the filtering algorithm is, and vice versa.
[0067] Specifically, the number of phase residual points in the filtered interferometric phase image can be calculated using the following calculation formulas (5) to (6).
[0068] (5) (6) in, represents the winding operator, which is used to wrap the operator to the range [-π, +π]; represents the phase of the pixel located in the i-th row and j-th column in the filtered interference phase image; represents the phase of the pixel located in the i-th row and j+1-th column in the filtered interference phase image; represents the phase of the pixel located in the i+1th row and jth column in the filtered interference phase image; represents the phase of the pixel located in the i+1th row and j+1th column in the filtered interference phase image; express and The pixel phase difference between express and The pixel phase difference between express and The pixel phase difference between express and The pixel phase difference between represents the phase closed loop integral sum. Where, if , then this point is the phase residual point; when When , it is a positive residual point; when , it is a negative residual point.
[0069] Reference Figure 8 The number of phase residual points is an important indicator for evaluating the phase filtering capability. In this example, 20 test data pairs can be selected from the test data set to calculate the number of residual points before and after filtering. Figure 8 It can be seen that all four methods can suppress the number of residual points. Among them, the Lee filter has a general effect, and there are still many residual points after filtering; Goldstein filtering can further effectively remove residual points compared with Lee filtering, and the filtering effect is more obvious; DnCNN filtering has better performance and can suppress residual points better than Goldstein filtering; and the number of residual points is further suppressed after improved CNN filtering, and for most test data, the number of residual points can even be reduced to single digits. From the perspective of the number of residual points, the improved CNN filtering has the best effect.
[0070] 2) Using a preset root mean square error calculation rule, calculate the root mean square error of the corresponding phase value between the filtered interference phase image and the corresponding noise-free interference phase image as an error indicator.
[0071] Among them, the root mean square error (RMSE) is a statistical indicator used to measure the difference between two data. In the field of image processing, RMSE is usually used to measure the difference between two images and evaluate the effect of image processing algorithms. The smaller the value, the higher the similarity between the two images.
[0072] Optionally, in step 2), the root mean square error may be calculated using the following calculation formula (7).
[0073] (7) in, represents the phase of the pixel located in the i-th row and j-th column in the filtered interference phase image; represents the pixel phase of the i-th row and j-th column in the noise-free phase image corresponding to the filtered interference phase image; N represents the number of pixels in the sliding window.
[0074] 3) Determine the second performance indicator based on the residual indicator and the error indicator.
[0075] S1330. Determine the filtering performance of the preset phase filtering model based on the first performance indicator and the second performance indicator.
[0076] Reference Fig. 9 , the difference between the filtered phase image and the noise-free phase image is measured by calculating the root mean square error index, and the root mean square error of the test set data is also calculated. Fig. 9 It can be seen that all four methods can achieve denoising effect and reduce the difference between the filtered phase image and the noise-free phase; among them, the Lee filtering effect is general, and the RMSE mean after filtering only drops to 1.410; the Goldstein filtering effect is better than the Lee filtering, and the RMSE mean can be reduced to 1.188; the DnCNN filtering effect is also very obvious, which is more concentrated in the box plot, and the RMSE mean is 0.391, which is more advantageous than the traditional filtering method; the improved CNN filtering effect is also very obvious, and its box plot is also relatively concentrated. After calculation, its RMSE mean is 0.237, which is more advantageous than the DnCNN filtering method, indicating that the phase image after filtering by this method is closer to the noise-free phase image.
[0077] As described above, the present disclosure proposes a method for evaluating the filtering performance of a phase filtering model (each phase filtering model corresponds to an InSAR phase filtering method) from the perspective of the entire process of InSAR phase filtering. Specifically, the method first produces a test (simulation) data set to provide test data for evaluating the effects of different phase filtering methods; then, the test data set is applied with different phase filtering methods respectively, and the corresponding filtering results are saved; finally, different filtering results are qualitatively and quantitatively evaluated, so that the effectiveness of different phase filtering methods can be better compared under the same data set, filling the gap in the prior art in solving the problem of evaluating the filtering capabilities of different phase filtering methods.
[0078] The system for evaluating the filtering performance of a phase filtering model provided by the present disclosure is described below. The system for evaluating the filtering performance of a phase filtering model described below and the method for evaluating the filtering performance of a phase filtering model described above can be referenced to each other.
[0079] Fig.10 FIG. 1 is a schematic diagram of a system for evaluating the filtering performance of a phase filtering model provided by an exemplary embodiment of the present disclosure. Fig.10 As shown, a system for evaluating the filtering performance of a phase filtering model (the phase filtering model is used to filter an interferometric synthetic aperture radar phase image) comprises: The phase simulation module 101 is configured to: construct a test data set for evaluating the phase filtering model; wherein the test data set includes at least one group of test data pairs, each group of test data pairs in the at least one group of test data pairs includes a noise interference phase map and a noise-free interference phase map, and the noise interference phase map is obtained by adding phase noise to the noise-free interference phase map; the phase filtering module 102 is configured to: input the noise interference phase map in each group of test data pairs into a preset phase filtering model, and output the corresponding filtered interference phase map via the preset phase filtering model; the filtering performance evaluation module 103 is configured to: based on the noise interference phase map, the filtered interference phase map and the corresponding noise-free interference phase map, use the preset filtering performance evaluation rules to determine the filtering performance of the preset phase filtering model.
[0080] Fig.11 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.11 As shown, the electronic device may include: a processor (processor) 1110, a communication interface (Communications Interface) 1120, a memory (memory) 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call the logic instructions in the memory 1130 to execute a method for evaluating the filtering performance of a phase filtering model, the method comprising: constructing a test data set for evaluating the phase filtering model; wherein the test data set comprises at least one set of test data pairs, each set of test data pairs in the at least one set of test data pairs comprises a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; the noise interference phase image in each set of test data pairs is input into a preset phase filtering model, and a corresponding filtered interference phase image is output via the preset phase filtering model; based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined using preset filtering performance evaluation rules.
[0081] In addition, the logic instructions in the above-mentioned memory 1130 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0082] On the other hand, the present disclosure also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method provided by the above-mentioned methods for evaluating the filtering performance of a phase filtering model, the method including: constructing a test data set for evaluating the phase filtering model; wherein the test data set includes at least one group of test data pairs, each group of test data pairs in the at least one group of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; the noise interference phase image in each group of test data pairs is input into a preset phase filtering model, and the corresponding filtered interference phase image is output through the preset phase filtering model; based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined using preset filtering performance evaluation rules.
[0083] On the other hand, the present disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the method provided by the above-mentioned methods for evaluating the filtering performance of a phase filtering model, the method comprising: constructing a test data set for evaluating the phase filtering model; wherein the test data set comprises at least one group of test data pairs, each group of test data pairs in the at least one group of test data pairs comprises a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; inputting the noise interference phase image in each group of test data pairs into a preset phase filtering model, and outputting a corresponding filtered interference phase image via the preset phase filtering model; based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, using preset filtering performance evaluation rules, determining the filtering performance of the preset phase filtering model.
[0084] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0085] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure.
Claims
1. A method for evaluating the filtering performance of a phase filtering model, characterized in that: The phase filtering model is used to filter an interferometric synthetic aperture radar phase image, and the method comprises: Constructing a test data set for evaluating the phase filtering model; wherein the test data set includes at least one set of test data pairs, each of the at least one set of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; Inputting the noise interference phase image in each group of test data pairs into a preset phase filtering model, and outputting a corresponding filtered interference phase image via the preset phase filtering model; Based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined using preset filtering performance evaluation rules.
2. The method according to claim 1, characterized in that The constructing of a test data set for evaluating the phase filtering model comprises: Obtaining a sample digital elevation model of a selected sample area and satellite parameters associated with determining the sample digital elevation model; Based on the sample digital elevation model and the satellite parameters, a plurality of noise-free interferometric phase images are generated using a preset interferometric phase simulation rule; Based on the multiple noise-free interference phase images, using interference phase noise simulation rules, determining the phase noise corresponding to each noise-free interference phase image; Adding each of the noise-free interference phase images to the corresponding phase noise to obtain a plurality of the noise interference phase images; The test data set is obtained based on a plurality of the noise-free interference phase images and a plurality of the noise interference phase images.
3. The method according to claim 2, characterized in that Based on the sample digital elevation model and the satellite parameters, a plurality of noise-free interferometric phase images are generated using preset interferometric phase simulation rules, including: Using the following calculation formula (1), the plurality of noise-free interference phase images are determined; (1) in, represents the noise-free interferometric phase; H Indicates the main satellite altitude; B Indicates the spatial baseline between the primary satellite and the secondary satellite; r 0 represents the slant distance between the primary satellite and the ground point; h Indicates the elevation of a ground point; represents the angle between the spatial baseline and the horizontal direction; represents the radar wavelength; or, Using the following calculation formula (2), the plurality of noise-free interference phase images are determined; (2) in, Represents the angle between the slant range of the main satellite relative to the ground point and the ground normal; represents the vertical baseline component obtained by separating the spatial baseline at the slant distance.
4. The method according to claim 2, characterized in that: Based on the multiple noise-free interference phase images, using interference phase noise simulation rules, determining the phase noise corresponding to each noise-free interference phase image, including: Based on each of the noise-free interference phase images, the corresponding probability density function of the phase noise is determined using the following calculation formula (3); (3) in, represents the i-th phase in the noise-free interferometric phase diagram; express The mathematical expectation of represents the i-th coherence corresponding to the noise-free interferometric phase pattern; Based on the probability density function, the phase noise is determined.
5. The method according to claim 1, characterized in that The preset phase filtering model includes a spatial filtering model, a frequency domain filtering model and a deep learning filtering model; The deep learning filtering model includes a first convolutional neural network model and a second convolutional neural network model.
6. The method according to claim 1, characterized in that The first convolutional neural network model includes a cascaded first input unit, a plurality of first feature learning units, and a first output unit; The first input unit includes a cascaded convolutional layer and an activation function layer ReLU, each of the first feature learning units includes a cascaded convolutional layer, an activation function layer ReLU and a batch normalization layer BN, and the first output unit includes a cascaded convolutional layer; The second convolutional neural network model includes a cascaded second input unit, a plurality of second feature learning units, one first feature learning unit and a second output unit; The second input unit comprises a cascaded convolution layer, an activation function layer ReLU and a batch normalization layer BN, and each of the second feature learning units comprises a cascaded residual block structure, a convolution layer, an activation function layer ReLU and a batch normalization layer BN; the second output unit comprises a cascaded convolution layer and an activation layer Sigmoid; The residual block structure includes a cascaded input end, two first convolution blocks and an output end, and the input end is jump-connected with the outputs of the two first convolution blocks; each of the first convolution blocks includes a cascaded convolution layer, an activation function layer ReLU and a batch normalization layer BN; The convolution kernel size of the convolution layer is 3×3.
7. The method according to claim 1, characterized in that Based on the noise interference phase image, the filtered interference phase image and the corresponding noise-free interference phase image, the filtering performance of the preset phase filtering model is determined by using a preset filtering performance evaluation rule, including: Based on the noise interference phase image and the corresponding filtered interference phase image, using a preset qualitative evaluation rule, determining a first performance indicator of the preset phase filtering model; Based on the filtered interference phase image and the corresponding noise-free interference phase image, using a preset quantitative evaluation rule, determining a second performance indicator of the preset phase filtering model; Based on the first performance indicator and the second performance indicator, the filtering performance of the preset phase filtering model is determined.
8. The method according to claim 7, characterized in that Based on the noise interference phase image and the corresponding filtering interference phase image, a first performance indicator of the preset phase filtering model is determined by using a preset qualitative evaluation rule, including: Using a preset filtering effect identification method, performing noise identification on the noise interference phase image to obtain a first noise identification result; Using a preset filtering effect identification method, noise identification is performed on the filtered interference phase image to obtain a second noise identification result; determining a noise retention index based on a difference between the first noise recognition result and the second noise recognition result; Acquire a first cross-sectional view at a preset position in the noise interference phase map; Acquire a second cross-sectional image of the preset position in the filtered interference phase image; determining a profile line fluctuation index based on a difference between the first profile and the second profile; The first performance indicator is determined based on the noise persistence indicator and the profile fluctuation indicator.
9. The method according to claim 7, characterized in that: Based on the filtered interference phase image and the corresponding noise-free interference phase image, a second performance indicator of the preset phase filtering model is determined using a preset quantitative evaluation rule, including: Based on the filtered interferometric phase image, using a preset phase residual point calculation rule, the number of phase residual points in the filtered interferometric phase image is calculated as a residual indicator; Using a preset root mean square error calculation rule, calculate the root mean square error of the corresponding phase value between the filtered interference phase image and the corresponding noise-free interference phase image as an error indicator; The second performance indicator is determined based on the residual indicator and the error indicator.
10. A system for evaluating the filtering performance of a phase filtering model, characterized in that: The phase filtering model is used to filter an interferometric synthetic aperture radar phase image, and the device comprises: A phase simulation module is configured to: construct a test data set for evaluating the phase filtering model; wherein the test data set includes at least one set of test data pairs, each of the at least one set of test data pairs includes a noise interference phase image and a noise-free interference phase image, and the noise interference phase image is obtained by adding phase noise to the noise-free interference phase image; The phase filtering module is configured to: input the noise interference phase image in each group of test data pairs into a preset phase filtering model, and output a corresponding filtered interference phase image through the preset phase filtering model; The filtering performance evaluation module is configured to determine the filtering performance of the preset phase filtering model based on the noise interference phase diagram, the filtered interference phase diagram and the corresponding noise-free interference phase diagram using preset filtering performance evaluation rules.
