An Enhanced Directioning Method for Underwater Confined Polarization Modes

By employing techniques such as adaptive low-pass filtering, exponential operation, and nonlocal sparse coding, a biomimetic polarization enhancement orientation method was constructed, which solved the orientation accuracy problem of underwater polarization compasses under weak polarization and noise modes, and achieved high-precision heading calculation.

CN118038256BActive Publication Date: 2026-11-17ZHONGBEI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410139947.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-31
Publication Date
2026-11-17
Estimated Expiration
2044-01-31

AI Technical Summary

Technical Problem

In underwater weak polarization and noise modes, the orientation accuracy of the polarization compass decreases, leading to inaccurate heading calculations.

Method used

A biomimetic polarization enhancement orientation method is constructed by employing techniques such as adaptive low-pass filtering, exponential operation, similar block operation, nonlocal sparse coding, and spatial high-pass filtering. Through adaptive gain control and denoising optimization, polarization image details are restored and orientation accuracy is improved.

Benefits of technology

The polarization compass has improved orientation accuracy and enhanced anti-disturbance capability in underwater environments, optimized feature pixel processing speed, and broadened its high-precision application in underwater environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118038256B_ABST
    Figure CN118038256B_ABST
Patent Text Reader

Abstract

This invention discloses an enhanced orientation method for underwater constrained polarization modes. It constructs a biomimetic polarization mode enhancement algorithm by drawing inspiration from the enhanced perception mechanism of insect-like visual neural pathways under low-light conditions, thereby enhancing weak polarization information under underwater polarization transmission constraints. Further optimization is achieved using a nonlocal sparse coding denoising module. Considering the similarity of the detailed structure of the effective zenith region of the polarized image, and relying on the stable symmetrical distribution characteristics of the polarization angle, the distribution of positive and negative feature points is traversed and allocated to accurately eliminate invalid polarized pixels. This invention effectively solves the orientation error problem caused by underwater constrained polarization modes, reconstructs the symmetry of partially damaged polarization angle modes, reduces external interference errors, autonomously corrects the fitted heading, effectively improves the underwater environmental adaptability of the biomimetic polarization compass, and enhances the orientation accuracy of the polarization compass.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to biomimetic polarization orientation technology, specifically to an enhanced orientation method for underwater confined polarization modes. Background Technology

[0002] In aquatic environments with satellite rejection and multi-source interference, underwater polarization information can provide reliable heading feedback. However, the low illumination and weak refraction caused by cloud depolarization in underwater polarization environments can lead to a loss of detail in polarization images, affecting refraction and polarization caused by light scattering, weakening the polarization capability of aerosols, and causing deviations in polarization mode calculations when the aerosols are refracted into the water and detected by a polarization compass lens. To a certain extent, the "∞" symmetry of the polarization angle image tends to diverge, with positive and negative feature points clustered sparsely, which is not conducive to the accurate extraction of the solar meridian. Furthermore, given that in certain turbid water scenarios, the degree of polarization decreases and becomes indistinct as turbidity increases, [further details may be needed]. The distribution of feature point pixels can be disrupted by various irregular noise particles, directly affecting the feature point fitting and solving process, and severely reducing orientation accuracy. Therefore, it is necessary to invent a feasible polarization orientation method under the constrained conditions of underwater scenarios to solve the problem of large errors in polarization compasses in underwater confined environments. Summary of the Invention

[0003] The technical problem to be solved by this invention is: in order to solve the problem of decreased orientation accuracy of polarization compasses in underwater weak polarization and noise modes, this invention proposes an enhanced orientation method for underwater restricted polarization modes, which can improve the orientation accuracy of polarization compasses in underwater environments.

[0004] Technical solution: An enhanced orientation method for underwater confined polarization modes, comprising the following steps:

[0005] Step 1: Obtain the original underwater polarization image pixels Based on the average brightness intensity of the polarized image, the original image pixels are... Classify the data and then process the classified data. Adaptive low-pass filtering is performed, and pixel values ​​are obtained by using nonlinear adaptive gain for variable gain control. The adaptive low-pass filter contains a split feedback loop, which uses the signal obtained at the previous time step. Using the square root function as the denominator in the splitting feedback loop to simulate steady state, the filtered underwater polarization image pixels are obtained. , The calculation formula is:

[0006] (1)

[0007] In the formula, The gain factor for the gain mapping; For adaptive angular frequency; Indicates frame rate; These are the results of the first and second stage adaptive low-pass filtering. This is a low-pass filter function; This is the output result of the third-stage adaptive low-pass filter;

[0008] Step 2: Filter the underwater polarization image pixels Perform exponential operations and then the Naka-Rushton equation transformation to obtain the transformed underwater polarization image pixels. , The calculation formula is:

[0009] (2)

[0010] In the formula: It is the pixel value output by the exponential operation feedback loop. Indicates exponential sensitivity. It is a positive offset of empirical selection. This is the output result of the fourth-stage adaptive low-pass filter;

[0011] Step 3: For the transformed underwater polarization image pixels Given any point in an underwater noise polarization image The vertices are selected from n overlapping neighborhoods of size m [Y]. l , ... Y n Define set S i for The similarity measure is obtained from the matrix of similar block sets:

[0012] (3)

[0013] In the formula, Indicates a fixed threshold. A represents the vectorized neighborhood of the corresponding point j in the search window. i Represents the matrix norm. It is S i The number of elements in, a j A represents i In the j-th row, A i The number of non-zero rows in set S is limited. i Performing similar block operations in the middle will yield similar estimates, where (p,q) is an empirical value;

[0014] Step 4: Perform dictionary learning on the similarity block set matrix obtained in Step 3. Use polarization image blocks constructed from similar underwater pixels as solutions with the same sparse coefficients. Optimize using a nonlocal sparse coding denoising method to obtain the optimized denoised polarization image I. out2,

[0015] (4)

[0016] In the formula, D is the learning dictionary, ε i For threshold parameters, For any given polarization pattern, It is a sparse coefficient matrix;

[0017] Step 5: For the denoised polarization pattern I obtained in Step 4 out2 Remove the normals m estimated from the training spatiotemporal data and standardize them to obtain the data after removal. The calculation formula is:

[0018] (5)

[0019] Step Six: Process the removed data Spatial high-pass filtering is performed, and the adaptive time constant during the filtering process is calculated using the Naka-Rushton transform adaptive factor to obtain the filtering output result. The calculation formula is:

[0020] (6)

[0021] In the formula, Scaling factor This is the output of the fifth-stage adaptive low-pass filter;

[0022] Step 7: The result of filtering in Step 6 A hyperbolic tangent function is used for saturated nonlinear processing to obtain the processed underwater polarization pixel image W(t). The degree of polarization and polarization angle of W(t) are calculated. Based on the strong inverse symmetry and stability of the polarization angle, the effective polarization angle image pixels within a threshold range of approximately ±90° are fitted to obtain the meridian position. Given the carrier coordinate axis, the angle between the carrier direction and the solar meridian in the carrier coordinate axis is obtained based on the carrier coordinate axis and the meridian position. The solar azimuth angle is obtained based on the local solar altitude angle, azimuth angle, latitude of the observation point, solar declination angle, and solar hour angle. Calculate the absolute heading angle φ:

[0023] (7).

[0024] Furthermore, in step one, the original image pixels... The classification process includes: setting the average brightness intensity as the classification threshold, and classifying the original image pixels in different regions of the polarized image. The brightness intensity is compared with the classification threshold to divide the polarization image into high brightness area and low brightness area.

[0025] Furthermore, in step one, the adaptive methods for nonlinear adaptive gain and adaptive angular frequency during the adaptive filtering process include: selecting different gain factors for the high-brightness region and the low-brightness region, respectively. The gain factor in the low-brightness region is greater than that in the high-brightness region; different filter angular frequencies are set according to the noise level contained in the polarization image signal intensity, and the calculation formula is as follows:

[0026] (8)

[0027] In the formula, I nr for Bit depth normalized value; This is the maximum gain value; and These are the maximum and minimum fit rates of the filter, which depend on the signal noise intensity of the polarization compass sensor.

[0028] Compared to existing technologies, the enhanced orientation method for underwater confined polarization modes provided by this invention has the following beneficial effects: This invention constructs an integrated underwater biomimetic polarization enhancement and descrambling system model, simulating the enhanced perception mechanism found in insect-like visual neural pathways. It performs adaptive low-pass filtering and variable gain control on the raw data, compressing the dynamic range by amplifying lower rather than higher pixel values, thereby enhancing the information content extracted from underwater weak polarization modes. This achieves external adjustment of underwater pixel illumination changes and reconstructs the "∞" symmetry of typical polarization angle images, a fundamental characteristic for detecting solar meridians. Furthermore, considering the pollution noise in turbid water environments encountered when processing underwater polarization images, this invention optimizes and embeds a nonlocal sparse coding denoising optimization module. Combining the similarity of the detailed structure of the effective zenith region of the underwater polarization image, and relying on the stable symmetrical distribution characteristics of the polarization angle, it completes denoising compensation and removes interfering pixels. Therefore, this invention can compensate for the lack of orientation accuracy of polarization compasses in underwater confined modes, has robustness against disturbances, can improve the regularity of feature pixels, improve the accuracy of heading calculation, and optimize the processing speed of feature point areas. It is of great significance for broadening the high-precision application of polarization compasses in underwater environments. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the underwater polarization image enhancement sensing mechanism model proposed in this invention;

[0030] Figure 2 This is the experimental processing result of the present invention under weak polarization and noise mode. Detailed Implementation

[0031] The present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments.

[0032] An enhanced orientation method for underwater confined polarization modes, such as Figure 1 As shown, it includes the following steps:

[0033] Step 1: Obtain the original underwater polarization image pixels The average brightness intensity is set as the classification threshold. In different regions of the polarized image, the original image pixels are classified. The brightness intensity is compared with the classification threshold to divide the polarization image into high brightness area and low brightness area.

[0034] The classified data is subjected to adaptive low-pass filtering, and pixel values ​​are obtained by using nonlinear adaptive gain for variable gain control. The adaptive low-pass filter contains a split feedback loop, which uses the signal obtained at the previous time step. Using the square root function as the denominator in the splitting feedback loop to simulate steady state, the filtered underwater polarization image pixels are obtained. , The calculation formula is:

[0035] (1)

[0036] In the formula, The gain factor for the gain mapping; For adaptive angular frequency; Indicates frame rate; These are the results of the first and second stage adaptive low-pass filtering. This is a low-pass filter function; This is the output of the third-stage adaptive low-pass filter.

[0037] The adaptive low-pass filtering in this step mainly includes nonlinear adaptive gain and adaptive angular frequency. Nonlinear adaptive gain: Different gain factors are selected for the high-brightness and low-brightness regions respectively. The gain factor in the low-brightness region is greater than that in the high-brightness region. By using a larger gain when the brightness is relatively low, the difference in brightness over a wide range is reduced, thereby reducing the brightness difference. To improve the signal-to-noise ratio, different filter angular frequencies are set according to the noise level contained in the polarized image signal intensity. The calculation formula is as follows:

[0038] (2)

[0039] In the formula, for Bit depth normalized value; This is the maximum gain value; and These are the maximum and minimum fit rates of the filter, which depend on the signal noise intensity of the polarization compass sensor. Step 2: Filter the underwater polarization image pixels. Perform exponential operations and then the Naka-Rushton equation transformation to obtain the transformed underwater polarization image pixels. , The calculation formula is:

[0040] (3)

[0041] In the formula: It is the pixel value output by the exponential operation feedback loop. Indicates exponential sensitivity. It is a positive offset of empirical selection. This is the output of the fourth-stage adaptive low-pass filter;

[0042] Step 3: For the transformed underwater polarization image pixels Given any point in an underwater noise polarization image The vertices are selected from n overlapping neighborhoods of size m [Y]. l , ... Y n Define set S i for The similarity measure is obtained from the matrix of similar block sets:

[0043] (4)

[0044] In the formula, Indicates a fixed threshold. A represents the vectorized neighborhood of the corresponding point j in the search window. i Represents the matrix norm. It is S i The number of elements in, a j A represents i In the j-th row, A i The number of non-zero rows in set S is limited. i Performing similar block operations in the middle will yield similar estimates, where (p,q) is an empirical value;

[0045] Step 4: Perform dictionary learning on the similarity block set matrix obtained in Step 3. Use polarization image blocks constructed from similar underwater pixels as solutions with the same sparse coefficients. Optimize using a nonlocal sparse coding denoising method to obtain the optimized denoised polarization image I. out2 ,

[0046] (5)

[0047] In the formula, Regularization ensures that all groups have the same weights, D is the learning dictionary, and ε i For threshold parameters; For any given polarization pattern, i.e., the training sample, It is a sparse coefficient matrix.

[0048] Step 5: For the optimized solution I obtained in Step 4 out2 Remove the normals m estimated from the training spatiotemporal data and standardize them to obtain the data after removal. The calculation formula is:

[0049] (6)

[0050] Step Six: For the removed data Spatial high-pass filtering is performed, and the adaptive time constant in the filtering process is calculated using the Naka-Rushton transform adaptive factor to obtain the filtering output result. The calculation formula is:

[0051] (7)

[0052] In the formula, Scaling factor This is the fifth-stage adaptive low-pass filter output.

[0053] Step 7: The result of filtering in Step 6 The hyperbolic tangent function is used for saturated nonlinear processing to obtain the processed underwater polarization pixel image W(t). The degree of polarization and polarization angle of W(t) obtained in step six are calculated. Based on the strong inverse symmetry and stability of the polarization angle (∞), the effective polarization angle image pixels within a threshold range of approximately ±90° are fitted to obtain the meridian position. Given the carrier coordinate axes, the angle between the carrier direction and the solar meridian in the carrier coordinate axes is obtained based on the carrier coordinate axes and the meridian position. The solar azimuth angle is obtained based on the local solar altitude angle, azimuth angle, latitude of the observation point, solar declination angle, and solar hour angle. Calculate the absolute heading angle φ:

[0054] (8)

[0055] The following experiments verify the effectiveness of the enhanced orientation method for underwater confined polarization modes in this embodiment. By simulating real polarization images in an aquatic environment, the processing effect of the integrated insect-inspired visual enhancement and de-interference system model on polarization images under underwater confined conditions is evaluated. The key test compass equipment, including a custom biomimetic polarization compass supported by an integrated satellite and inertial navigation reference system (SPAN-KVH1750), was placed in a water tank. The biomimetic polarization compass consists of a wide-angle lens, a PHX050S-PC polarization camera, and an NVIDIA TX2 processor. Under underwater low-light conditions, cloud depolarization refraction, and turbidity, weak polarization and noisy modes formed by irregularly distributed feature pixels significantly reduce fitting accuracy. Figure 2 It can be seen that the feature lines distinguished by the polarization angle diagram fitting under low illumination are unclear, the details of the polarization mode are lost under the depolarization refraction of thin clouds, and the polarization angle diagram under turbidity is even more problematic because the critical ∞ symmetry is destroyed. All of these are detrimental to the processing of the fitting algorithm. The enhanced orientation method proposed in this embodiment retains or recovers some zenith region feature points in the effective polarization region, removes noise, and adaptively compensates for the distribution characteristics of feature points, thus highlighting polarization details. The ∞ symmetry distribution of the polarization angle diagram along the solar meridian is largely preserved. Therefore, the output polarization solution image can be partially or completely recovered from the weak polarization mode and the noisy mode, thereby improving the orientation accuracy of the polarization compass.

[0056] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes can be made to the invention in terms of weather conditions and algorithm parameters without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. An enhanced orientation method for underwater confined polarization modes, characterized in that, Includes the following steps: Step 1: Obtain the original underwater polarization image pixels Based on the average brightness intensity of the polarized image, the original image pixels are... Classify the data and then process the classified data. Adaptive low-pass filtering is performed, and pixel values ​​are obtained by using nonlinear adaptive gain for variable gain control. The adaptive low-pass filter contains a split feedback loop, which uses the signal obtained at the previous time step. Using the square root function as the denominator in the splitting feedback loop to simulate steady state, the filtered underwater polarization image pixels are obtained. , The calculation formula is: (1) In the formula, The gain factor for the gain mapping; For adaptive angular frequency; Indicates frame rate; These are the results of the first and second stage adaptive low-pass filtering. This is a low-pass filter function; This is the output result of the third-stage adaptive low-pass filter; Step 2: Filter the underwater polarization image pixels Perform exponential operations and then the Naka-Rushton equation transformation to obtain the transformed underwater polarization image pixels. , The calculation formula is: (2) In the formula: It is the pixel value output by the exponential operation feedback loop. Indicates exponential sensitivity. It is a positive offset of empirical selection. This is the output result of the fourth-stage adaptive low-pass filter; Step 3: For the transformed underwater polarization image pixels Given any point in an underwater noise polarization image The vertices are selected from n overlapping neighborhoods of size m [Y]. l , ... Y n Define set S i for From the set matrix of similar blocks, the similarity measure is obtained as follows: (3) In the formula, Indicates a fixed threshold. A represents the vectorized neighborhood of the corresponding point j in the search window. i Represents the matrix norm. It is S i The number of elements in, a j A represents i In the j-th row, A i The number of non-zero rows in set S is limited. i Performing similar block operations in the middle will yield similar estimates, where (p,q) is an empirical value; Step 4: Perform dictionary learning on the similarity block set matrix obtained in Step 3. Use polarization image blocks constructed from similar underwater pixels as solutions with the same sparse coefficients. Optimize using a nonlocal sparse coding denoising method to obtain the optimized denoised polarization image I. out2 , (4) In the formula, D is the learning dictionary, ε i For threshold parameters, For any given polarization pattern, It is a sparse coefficient matrix; Step 5: For the denoised polarization pattern I obtained in Step 4 out2 Remove the normals m estimated from the training spatiotemporal data and standardize them to obtain the data after removal. The calculation formula is: (5) Step Six: Process the removed data Spatial high-pass filtering is performed, and the adaptive time constant during the filtering process is calculated using the Naka-Rushton transform adaptive factor to obtain the filtering output result. The calculation formula is: (6) In the formula, Scaling factor This is the output of the fifth-stage adaptive low-pass filter; Step 7: The result of filtering in Step 6 A hyperbolic tangent function is used for saturated nonlinear processing to obtain the processed underwater polarization pixel image W(t). The degree of polarization and polarization angle of W(t) are calculated. Based on the strong inverse symmetry and stability of the polarization angle, the meridion position is obtained by fitting the effective polarization angle image pixels within a threshold range of approximately ±90°. Given the carrier coordinate axis, the angle between the carrier direction and the solar meridion is obtained from the carrier coordinate axis and the meridion position. The solar azimuth angle is obtained based on the local solar altitude angle, azimuth angle, latitude of the observation point, solar declination angle, and solar hour angle. Calculate the absolute heading angle φ: (7)。 2. The enhanced orientation method for underwater confined polarization modes according to claim 1, characterized in that, In step one, the original image pixels are processed. The classification process includes: setting the average brightness intensity as the classification threshold, and classifying the original image pixels in different regions of the polarized image. The brightness intensity is compared with the classification threshold to divide the polarization image into high-brightness and low-brightness regions.

3. The enhanced orientation method for underwater confined polarization modes according to claim 1 or 2, characterized in that, In step one, the adaptive methods for nonlinear adaptive gain and adaptive angular frequency during the adaptive filtering process include: selecting different gain factors for the high-brightness region and the low-brightness region, respectively. The gain factor in the low-brightness region is greater than that in the high-brightness region; different filter angular frequencies are set according to the noise level contained in the polarization image signal intensity, and the calculation formula is as follows: (8) In the formula, for Bit depth normalized value; This is the maximum gain value; and These are the maximum and minimum fit rates of the filter, which depend on the signal noise intensity of the polarization compass sensor.