A DWT-based InSAR phase filtering method
Patent Information
- Application Number
- CN202111162827.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-30
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-09-30
AI Technical Summary
The existing InSAR phase filtering method has shortcomings in the changes in edge position and the absence of interference fringes, and the filter performance depends on the window dimension and has low processing efficiency.
Using the InSAR phase filtering method based on DWT, the three-stage wavelet transformation and DPWT transformation are used to distinguish signals and noise using adaptive mask patterns and thresholds, and multiple inverse discrete wavelet transformations and mask growth are performed to remove noise and retain phase details and Striped edges.
Improves processing efficiency, can effectively filter out phase noise, completely retain original phase details and striped edges, and does not rely on window dimensions.
Smart Images

Figure CN113920025B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of image processing, and in particular relates to an InSAR phase filtering method based on DWT. Background Art
[0002] Wavelet transform has been widely used in signal and image processing, especially for the elimination of noise in SAR images and InSAR images. Braunisch H proposed a soft threshold SAR interferogram filtering algorithm based on Mallat wavelet decomposition, but in the soft threshold algorithm based on Mallat wavelet decomposition and reconstruction, it is difficult to select a suitable threshold to make the filtered interferogram achieve satisfactory results. Yue Huanyin et al. proposed a wavelet domain multi-scale filtering algorithm for interferograms, processed the real and imaginary parts of the interferograms separately, determined the probability density function of the signal and noise wavelet coefficients using the Pearson distribution system, and then estimated the wavelet coefficients of the noise-free data using the maximum a posteriori probability criterion. Wang Lucai et al. proposed an InSAR interferogram phase image filtering algorithm based on the combination of wavelet analysis and median filtering to improve the filtering effect. He Ruyun et al. proposed an InSAR interference pattern filtering algorithm based on wavelet transform. This algorithm first uses wavelet transform to perform multi-level decomposition on the interference pattern data to obtain multi-level approximate coefficients and detail coefficients in three directions of the image. Then, the detail coefficients in each direction of each level are respectively detected to see whether they are edges in the corresponding direction. The coefficients at the edge are smoothed with different direction templates according to the direction of the edge, and then median filtering is performed. The coefficients at the non-edge are directly median filtered. Although these methods have improved the wavelet transform, there are still some shortcomings in practical applications, such as changes in edge position and missing interference fringes. Summary of the invention
[0003] The present invention aims at the deficiencies of the prior art and provides a DWT-based InSAR phase filtering method. The performance of the filter in this method does not depend on the window dimension, which improves the processing efficiency. In addition, the adaptive mask image extracted by coherence is taken into account during the inverse wavelet transform process, which can not only filter out the phase noise well, but also retain the original phase details and fringe edges more completely.
[0004] In order to achieve the above object, the technical solution provided by the present invention is a DWT-based InSAR phase filtering method, comprising the following steps:
[0005] Step 1, performing a three-level wavelet transform on the complex interference phase;
[0006] Step 2, for the wavelet subdomain that has undergone the three-level wavelet transform, the pixels of the mask are determined by using the signal quality and the set threshold value to generate the mask;
[0007] Step 3, multiply the real and imaginary parts of the coefficients detected as signals in the mask by 2 respectively, while the noise coefficient remains unchanged;
[0008] Step 4, performing a first-order inverse discrete wavelet transform;
[0009] Step 5, obtaining a mask of the signal coefficients in the new high frequency level;
[0010] Step 6, repeating steps 3 to 5 for n times in total to obtain the complex signal in the original data;
[0011] Step 7, calculate the filtered phase value from the complex data obtained in step 6.
[0012] Moreover, after performing two DWT transformations on the original noisy InSAR image in step 1, a DPWT transformation is performed on it, and then a filtering process is performed to remove the noise. 3 With a fixed growth rate as the basis, in the third decomposition level, wavelet filters are applied to all bands to perform discrete wavelet packet decomposition.
[0013] Moreover, in order to obtain the wavelet signal coefficients in step 2, the signal quality Γ sig Using threshold Th ω To distinguish signal from noise, when Γ sig Greater than the threshold Th ω , then it is the signal coefficient, otherwise it is the noise coefficient. Threshold Th ω The signal quality is calculated based on the mean value of the sub-band dynamics. sig The formula is as follows:
[0014]
[0015] Where i is the wavelet transform series, I ω represents the intensity of the wavelet coefficients in the low-frequency subband, represents I in the noise band d1 ω The corresponding spatial domain noise variance.
[0016] Furthermore, the step 5 of obtaining the mask of the signal coefficients in the new high frequency level includes the following sub-steps:
[0017] Step 5.1, replace 2 i The four masks at the level are merged;
[0018] Step 5.2, double the scale of this merged mask to fit 2 i-1 Subband scale at level;
[0019] Step 5.3, according to 2i-1 The mask is grown with the coherence of 4 pixels on the subband.
[0020] Moreover, in step 5.2, 2 i-1 Instead of systematically classifying the four pixels on the first subband into the same class, a downsampled coherence map is obtained from the original InSAR coherence map. i-1 The sub-bands have the same size.
[0021] Furthermore, in step 5.3, 2 i A point (m,n) on the mask at level 2 i-1 The mask growth of the four points (k, l), (k, l+1), (k+1, l), (k+1, l+1) corresponding to the level obeys the following rules: If the point (m, n) is a signal pixel, then when |ρ(k, l)-ρ p |≤ε c When (p∈{(k,l);(k,l+1);(k+1,l);(k+1,l+1)}, ρ(k,l) is 2 i Pixels at level, ρ p For 2 i-1 Pixels at level, ε c is the set threshold), p is the signal coefficient, and vice versa p is the noise coefficient; if the point (m,n) is a noise pixel, then when |ρ(k,l)-ρ p |≤ε c When (p∈{(k,l);(k,l+1);(k+1,l);(k+1,l+1)}, ρ(k,l) is 2 i Pixels at level, ρ p For 2 i-1 Pixels at level, ε c is the set threshold), p is the noise coefficient, and vice versa p is the signal coefficient. That is to say, if a pixel on the merged mask is a signal, then the four pixels corresponding to it are i-1 If the pixel is still a signal pixel at the same level, it is classified as noise otherwise. This series of generated masks can effectively obtain the signal in the original data.
[0022] Compared with the prior art, the present invention has the following advantages: 1) The processing of the original signal does not require window operations similar to the traditional method, and the performance of the filter does not depend on the window dimension, which improves the processing efficiency. 2) In the process of inverse wavelet transform, the adaptive mask image extracted by coherence is taken into account, downsampled to the resolution of the corresponding wavelet decomposition scale, and the threshold is calculated by 4 equivalent signal bands. The threshold can be set adaptively, which can not only filter out phase noise better, but also retain the original phase details and stripe edges more completely. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 is a flow chart of an embodiment of the present invention.
[0024] Figure 2 This is the process of performing wavelet transform on the complex interference phase image in the embodiment of the present invention.
[0025] Figure 3 The mask growth process of the InSAR coherence map is taken into account for the embodiments of the present invention. DETAILED DESCRIPTION
[0026] The present invention provides an InSAR phase filtering method based on DWT. Firstly, a three-level wavelet transform is performed on the complex interference phase, a mask image is generated for the wavelet subdomain after the three-level wavelet transform using a set threshold value, the real part and the imaginary part of the coefficient detected as a signal in the mask image are multiplied by 2 respectively, the noise coefficient remains unchanged, and then a first-level inverse discrete wavelet transform is performed, and a mask growth rule is designed to update the mask image, and finally the phase value after filtering is obtained by calculating the phase value of the complex data.
[0027] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0028] like Figure 1 As shown, the process of the embodiment of the present invention includes the following steps:
[0029] Step 1: Perform a three-level wavelet transform on the complex interference phase.
[0030] like Figure 2 As shown in Figure 1, the original noisy InSAR image is transformed twice by DWT (Discrete Wavelet Transformation), and then transformed by DPWT (Discrete Parameter Wavelet Transform), and then filtered to remove the noise. In order to obtain the wavelet coefficient intensity of the signal at the first decomposition level, the intensity is calculated by 2 3 With a fixed growth rate of basis, in the third decomposition level, wavelet filters are used for all bands (signal a2+each second decomposition d2) to perform discrete wavelet packet decomposition.
[0031] Step 2: For the wavelet subdomain that has undergone the three-level wavelet transform, the pixels of the mask are determined using the signal quality and the set threshold to generate a mask.
[0032] In order to obtain the wavelet signal coefficients, the signal quality Γ sig Using threshold Th ω To distinguish signal from noise, when Γ sig Greater than the threshold Th ω, then it is the signal coefficient, otherwise it is the noise coefficient. Threshold Th ω The signal quality is calculated based on the mean value of the sub-band dynamics. sig The formula is as follows:
[0033]
[0034] Where i is the wavelet transform series, I ω represents the intensity of the wavelet coefficients in the low-frequency subband (a2 and all d2), represents I in the noise band d1 ω The corresponding spatial domain noise variance.
[0035] In step 3, the real part and the imaginary part of the coefficient detected as the signal in step 2 are multiplied by 2 respectively, while the noise coefficient remains unchanged.
[0036] Step 4: perform a first-level inverse discrete wavelet transform.
[0037] Step 5, obtaining a mask of the signal coefficients in the new high-frequency level, specifically includes the following sub-steps:
[0038] Step 5.1, replace 2 i The four masks at the same level are merged.
[0039] Step 5.2, double the scale of this merged mask to fit 2 i-1 The subband scale at the level.
[0040] This is not the time to i-1 Instead of systematically classifying the four pixels on the first subband into the same class, a downsampled coherence map is obtained from the original InSAR coherence map. i-1 The sub-bands have the same size.
[0041] Step 5.3, according to 2 i-1 The mask is grown with the coherence of 4 pixels on the subband.
[0042] like Figure 3 As shown, 2 i A point (m,n) on the mask at level 2 i-1 The mask growth of the four points (k, l), (k, l+1), (k+1, l), (k+1, l+1) corresponding to the level obeys the following rules: If the point (m, n) is a signal pixel, then when |ρ(k, l)-ρ p |≤ε c When (p∈{(k,l);(k,l+1);(k+1,l);(k+1,l+1)}, ρ(k,l) is 2 i Pixels at level, ρ p For 2i-1 Pixels at level, ε c is the set threshold), p is the signal coefficient, otherwise p is the noise coefficient; if the point (m,n) is a noise pixel, then when |ρ(k,l)-ρ p |≤ε c When (p∈{(k,l);(k,l+1);(k+1,l);(k+1,l+1)}, ρ(k,l) is 2 i Pixels at level, ρ p For 2 i-1 Pixels at level, ε c is the set threshold), p is the noise coefficient, and vice versa p is the signal coefficient. That is to say, if a pixel on the merged mask is a signal, then the four pixels corresponding to it are i-1 If the pixel is still a signal pixel at the same level, it is classified as noise otherwise. This series of generated masks can effectively obtain the signal in the original data.
[0043] Step 6, repeat steps 3 to 5 for a total of 3 times to obtain the complex signal in the original data.
[0044] Step 7, calculate the filtered phase value from the complex data obtained in step 6.
[0045] During specific implementation, the above process can be automatically operated using computer software technology.
[0046] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.
Claims
1. A DWT-based InSAR phase filtering method, characterized in that: The steps include: Step 1, performing a three-level wavelet transform on the complex interference phase; Step 2, for the wavelet subdomain that has undergone the three-level wavelet transform, the pixels of the mask are determined by using the signal quality and the set threshold value to generate the mask; Step 3, multiply the real and imaginary parts of the coefficients detected as signals in the mask by λ respectively, while the noise coefficient remains unchanged; Step 4, performing a first-order inverse discrete wavelet transform; Step 5, obtaining a mask of the signal coefficients in the new high frequency level; Step 5.1, replace 2 i The four masks at the level are merged; Step 5.2, double the scale of this merged mask to fit 2 i-1 Subband scale at level; It is not 2 i-1 Instead of systematically classifying the four pixels on the first subband into the same class, a downsampled coherence map is obtained from the original InSAR coherence map. i-1 The sub-bands have the same dimensions; Step 5.3, according to 2 i-1 The coherence of 4 pixels on the subband is used to grow the mask; 2 i A point (m,n) on the mask at level 2 i-1 The mask growth of the four points (k, l), (k, l+1), (k+1, l), (k+1, l+1) corresponding to the level obeys the following rules: If the point (m, n) is a signal pixel, then when |ρ(k, l)-ρ p |≤ε c When , p is the signal coefficient, otherwise p is the noise coefficient, where ρ(k,l) is 2 i Pixels at level, ρ p is a pixel at level 2i-1, p∈{(k,l);(k,l+1);(k+1,l);(k+1,l+1)}, ε c is the set threshold; if point (m,n) is a noise pixel, then when |ρ(k,l)-ρ p |≤ε c When p is the noise coefficient, p is the signal coefficient; that is, if a pixel on the merged mask is a signal, then the four pixels corresponding to it are i-1 At the same time, it is still a signal pixel at the same level, otherwise it is classified as noise; this series of generated masks can effectively obtain the signal in the original data; Step 6, repeating steps 3 to 5 for n times in total to obtain the complex signal in the original data; Step 7, calculate the filtered phase value from the complex data obtained in step 6.
2. The DWT-based InSAR phase filtering method according to claim 1, characterized in that: In step 1, after performing two DWT transformations on the original noisy InSAR image, it is transformed once with DPWT and then filtered to remove the noise. In order to obtain the wavelet coefficient intensity of the signal at the first decomposition level, the intensity is 2 3 With a fixed growth rate as the basis, in the third decomposition level, wavelet filters are applied to all bands to perform discrete wavelet packet decomposition.
3. The DWT-based InSAR phase filtering method according to claim 2, characterized in that: In step 2, in order to obtain the wavelet signal coefficient, the signal quality Γ sig Using threshold Th ω To distinguish signal from noise, when Γ sig Greater than the threshold Th ω , then it is the signal coefficient, otherwise it is the noise coefficient; threshold Th ω The signal quality is calculated based on the mean value of the sub-band dynamics. sig The formula is as follows: Where i is the wavelet transform series, I ω represents the intensity of the wavelet coefficients in the low-frequency subband, represents I in the noise band d1 ω The corresponding spatial domain noise variance.
