Local sampling and alternate iteration arbitrary polarization channel adaptive inversion method

Abstract

The invention provides an arbitrary polarization channel adaptive inversion method based on local sampling and alternate iteration. The method comprises the following steps: S1, acquiring a polarization channel image; s2, estimating global feature information and obtaining a local sample point set; s3, setting an initial analysis angle; s4, performing matrix transformation to obtain a multi-polarization channel matrix, an initial polarization angle matrix and an intermediate analysis angle matrix; s5, updating the middle polarization angle matrix according to the multi-polarization channel matrix and the middle analysis angle matrix, and solving to obtain a middle polarization angle; s6, judging whether a convergence condition is met or not, and if yes, turning to S7; if not, alternately iterating the step S4 and the step S5; and S7, calculating a final analysis angle, and reconstructing according to the final analysis angle matrix and the multi-polarization channel matrix to obtain a polarization degree image. The method has the beneficial effects that accurate direction angle inversion and efficient reconstruction of polarization characteristics can be realized, and the accuracy and robustness of polarization information analysis capability of different application scenes are improved.

Description

Technical Field

[0001] The invention relates to the technical field of polarization imaging, and in particular to an arbitrary polarization channel adaptive inversion method of local sampling and alternating iteration. Background Art

[0002] In modern optical imaging and remote sensing technology, polarization image acquisition and information analysis have become important means to improve imaging quality and target recognition capabilities. Polarization imaging technology can provide richer multi-dimensional feature information of the target than traditional intensity imaging by detecting the polarization state of the incident light wave, and has broad application prospects in industrial detection, biomedicine, and atmospheric monitoring. As the key to target information acquisition and feature deconstruction, the polarization imaging system usually obtains light field information under multiple polarization channels (no less than three), that is, images corresponding to different polarization angles (such as 0°, 60°, 120° or 0°, 45°, 90°, 135°, etc.), and then combines the back-end analysis algorithm to achieve the purpose of deconstructing the target polarization characteristics (such as polarization degree, polarization angle, etc.).

[0003] At present, according to the working principle of the spectroscopic element and the acquisition method of the polarization image, the polarization imaging system can be divided into four categories: time-sharing type, amplitude-sharing type, aperture-sharing type and focal plane-sharing type. Among them, the amplitude-sharing type, aperture-sharing type and focal plane-sharing type polarization imaging systems usually rely on preset polarization channels and can only obtain polarization state information at three or four fixed directions and angles, which greatly limits the flexibility and adaptability of the system when applied in complex environments. In addition, for the aperture-sharing type and focal plane-sharing type imaging systems, the structural arrangement mode of the fixed channels will cause the loss of spatial resolution, and it is difficult to meet the precise reconstruction of the scene polarization characteristics. In contrast, the time-sharing polarization imaging system can arbitrarily obtain polarization channel images at different directions and angles by rotating and scanning the polarization optical device. It has the advantages of free tuning, flexible adaptation and high spatial resolution, and is expected to achieve more abundant polarization state information acquisition in the scene and target detection with the optimal polarization channel combination in complex and changing environments. Among them, as one of the core components of the time-sharing polarization imaging system, the positioning accuracy of the rotating scanning polarization filter module is the key to determining the measurement accuracy and reliability of the system. However, due to the influence of factors such as calibration error, mechanical vibration, assembly error and module performance, polarization imaging systems often have inevitable rotation angle errors during application, and rotation errors may cause the angles corresponding to the polarization channels to have the following two types of deviations: 1) Absolute angle error, that is, the actual absolute angle of a channel deviates from the theoretical value during the rotation of the polarization optical element; 2) Relative angle error, that is, the actual relative angle between two adjacent channels deviates from the theoretical value during the rotation of the polarization optical element. Regardless of the type of rotation error, this will cause the direction angles corresponding to the different polarization channels actually collected to deviate from the theoretical value, thereby reducing the reconstruction accuracy of the polarization characteristics. Therefore, how to quickly and accurately invert the actual corresponding angle position of the image information obtained by the polarization imaging system when the actual direction angle of the rotating scanning polarization filter module is unknown, so as to achieve more accurate polarization feature reconstruction has become a technical problem that needs to be solved urgently in the current research field.

[0004] In order to solve the problem of rotation angle error in the time-sharing polarization imaging system, Wibowo et al. ( IEEE Access 2019，7：28651 ) proposed a correction method by analyzing and studying the errors, and realized the calibration and compensation of the rotation error of the polarization imaging system by building a calibration device. Although this method has achieved good results in the laboratory, it relies on expensive and precise calibration equipment and complicated calibration procedures, and is difficult to apply to complex and changeable field environments. In recent years, Mu Yankui's research group at Xi'an Jiaotong University ( Opt. Lett. 2020, 45(1): 57) inspired by the phase-shift interferometry technique, the polarization angle of the scene target and the analysis angle of the polarization optical device are calculated by least squares iteration between the polarization channel images until the iterative error converges. This is method 1. Subsequently, Lu Xiang et al. ( IEEE Trans. Instrum. Meas. 2024, 73: 4505708 ) On this basis, a nonlinear least squares iterative method based on Levenberg-Marquart is proposed to achieve more accurate polarization channel angle extraction and polarization feature reconstruction, which is method 2. Although the above method has the advantages of high accuracy and convenient operation, it has the following two shortcomings in practical applications due to the influence of unstable lighting environment and noise interference: 1) There is a problem of non-uniformity in the non-polarization intensity value and polarization intensity value between pixels in different polarization channel images and in the same polarization channel image, which seriously affects the accuracy of angle inversion and feature reconstruction; 2) Selecting a suitable area to invert the corresponding angle of the global polarization channel image, that is, the strategy of "generalizing the whole from the particular" can significantly reduce the amount of calculation, but the existing method lacks appropriate sampling strategy considerations, resulting in long iteration time, large amount of calculation and low inversion accuracy.

[0005] In summary, although the existing research has analyzed and studied the rotation angle error of the system, and the proposed iterative method can realize the polarization channel angle inversion when the rotation direction of the rotating scanning polarization filter module is unknown, due to the consistency assumption between channels and pixels and the iterative calculation of the pixel values ​​contained in the entire image, it lacks versatility and robustness in practical applications, and it is difficult to meet the needs of high-precision and high-efficiency polarization channel angle inversion and polarization feature reconstruction. Therefore, an arbitrary polarization channel adaptive inversion method is proposed to achieve accurate inversion of the corresponding angle of the high-efficiency polarization channel under realistic conditions such as environmental changes, noise interference, and uneven distribution of polarization features, which is of great significance to reducing the measurement error of the imaging system and improving the reconstruction accuracy of polarization features. Summary of the invention

[0006] The technical problem to be solved by the present invention is to achieve accurate azimuth inversion and efficient reconstruction of polarization characteristics, and to improve the accuracy and robustness of polarization information analysis capabilities in different application scenarios. In order to overcome the defects of the above-mentioned prior art (or related technology), the present invention provides an arbitrary polarization channel adaptive inversion method with local sampling and alternating iteration.

[0007] The present invention provides an arbitrary polarization channel adaptive inversion method with local sampling and alternating iteration, comprising the following steps: Step S1, for any scene target, using a polarization camera to collect a plurality of polarization channel images of different directions and angles containing the scene target; Step S2, estimating global feature information according to each of the polarization channel images, performing cluster analysis using the K-Means method to obtain a clustering result image, and selecting a local sample point set according to the clustering result image; Step S3, setting an initial analysis angle according to the number of each polarization channel image; Step S4, based on the least squares method, performing matrix transformation on each of the polarization channel images according to the local sample point set and the initial analysis angle to obtain a multi-polarization channel matrix, and solving the multi-polarization channel matrix to obtain an initial polarization angle matrix and an intermediate analysis angle matrix; Step S5, based on the least square method, updating the intermediate polarization angle matrix according to the multi-polarization channel matrix and the intermediate analysis angle matrix and solving to obtain the intermediate polarization angle; Step S6, determine whether the preset convergence condition is met: If yes, go to step S7; If not, then the step S4 and the step S5 are iterated alternately; Step S7, obtaining a final analysis angle according to the final analysis angle matrix and the final polarization angle matrix after alternating iterations, obtaining the Stokes vector of the incident light according to the final analysis angle matrix and the multi-polarization channel matrix, and reconstructing a polarization degree image according to the Stokes vector of the incident light.

[0008] Compared with the prior art, the method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration of the present invention has the following advantages: 1) The present invention does not rely on additional hardware optimization and complex calibration procedures. It adopts an alternating iteration method to achieve accurate azimuth inversion and efficient reconstruction of polarization characteristics when the azimuth angle of the polarization image is unknown and the initial analysis angle is randomly set. This method is simple and efficient, has strong applicability, and can cope with a variety of polarization image analysis scenarios; 2) The local sampling strategy adopted in the present invention does not require global estimation of the polarization channel image, but starts from local samples with consistent similarity, performs inversion on the local area, and applies the result to the reconstruction of the corresponding angle of the global image. This strategy makes the method more robust in complex scenes and significantly improves the reconstruction accuracy and computational efficiency; 3) Compared with the traditional global optimization method, the present invention can significantly reduce the computational burden and resource consumption by using alternating iterative optimization calculations in local areas. In addition, the method is suitable for various types of polarization images and complex scene conditions, has strong adaptability, and can achieve efficient and accurate polarization information inversion and reconstruction in a variety of practical applications, greatly improving the scalability and practical application value of the algorithm.

[0009] In a possible implementation manner, in step S1, the polarization camera used integrates a rotatable scanning polarization module, and the polarization module is driven to rotate to The polarization channel images at different direction angles are obtained at the positions of the angles.

[0010] In a possible implementation, step S2 includes: Step S21, based on the polarization channel images and in combination with the correlation between polarization channels, obtaining a maximum intensity image, a minimum intensity image, an estimated intensity image and an estimated degree of polarization image as the global feature information; Step S22, obtaining a feature vector of each pixel in each polarization channel image according to the global feature information combination, setting the number of clusters to randomly generate multiple cluster centers, counting the minimum distance from each feature vector to each cluster center, and performing cluster analysis using the K-Means method to obtain a cluster label of each pixel to form the cluster result image; Step S23, obtaining a sample point with the most significant polarization feature and an optimal cluster label to which the sample point belongs according to the estimated polarization degree image; Step S24: obtaining the in-cluster samples corresponding to the optimal cluster label as the local sample point set.

[0011] In a possible implementation, in step S21, the global feature information is obtained by the following calculation formula: ; in, represents the maximum intensity image; express The measured intensity value corresponding to the mth pixel position at angle, ; represents the minimum intensity image; representing the estimated intensity image; represents the estimated polarization degree image.

[0012] In a possible implementation, step S4 includes: Step S41, obtaining the corresponding multi-polarization channel matrix according to the local sample point set and the initial analysis angle; Step S42, based on the least square method, performing matrix transformation according to the multi-polarization channel matrix and the initial analysis angle matrix corresponding to the initial analysis angle to obtain the initial polarization angle matrix; Step S43, based on the least square method, matrix transformation is performed according to the intermediate polarization angle matrix after the multi-polarization channel matrix and the initial polarization angle matrix are changed to obtain the intermediate analysis angle matrix.

[0013] In a possible implementation, in step S41, the multi-polarization channel matrix is ​​obtained by the following calculation formula: ; in, represents the multi-polarization channel matrix; An intensity vector representing the polarization channel image; A cosine vector representing the polarization angle corresponding to the polarization channel image; A sine vector representing the polarization angle corresponding to the polarization channel image; a cosine vector representing the initial analysis angle; Represents the sine vector of the initial analysis angle.

[0014] In a possible implementation manner, the convergence condition in step S6 is: ; in, represents the intermediate polarization angle of the kth iteration; represents the intermediate polarization angle of the k-1th iteration; Indicates the preset convergence accuracy.

[0015] In a possible implementation, step S7 includes: Step S71, obtaining the final analysis angle according to the final analysis angle matrix and the final polarization angle matrix after alternating iterations; Step S72, performing matrix transformation based on the multi-polarization channel matrix and the final analysis angle matrix to obtain the Stokes vector of the incident light based on the least squares method; Step S73, reconstructing the polarization degree image according to the Stokes vector.

[0016] In a possible implementation, in step S71, the final analysis angle is obtained by the following calculation formula: ; in, represents the final analysis angle; represents the final polarization angle matrix; represents the final analysis angle matrix; represents the second column elements of the final analysis angle matrix; Represents the third column elements of the final analysis angle matrix.

[0017] In a possible implementation, in step S73, the polarization degree image is obtained by the following calculation formula: ; ; in, representing the polarization degree image; The Stokes vector representing the incident light; a first parameter representing the Stokes vector; a second parameter representing the Stokes vector; A third parameter representing the Stokes vector; represents the multi-polarization channel matrix; Denotes the final analysis angle matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the steps of the present invention; Figure 2 It is a schematic diagram of the image acquisition process of the present invention; Figure 3 It is a schematic diagram of polarization channel images at different directions and angles of the present invention; Figure 4 It is a schematic diagram of the global characteristic image of the present invention; Figure 5 It is a schematic diagram of the clustering result image of the present invention; Figure 6 It is a schematic diagram of a local sample point set mapping image and a Malus fitting curve of the present invention; Figure 7 A schematic diagram of a reconstructed Stokes parameter image and polarization degree image of the present invention; Figure 8 Schematic diagram of residual image for reconstructing polarization characteristics using different methods of the present invention. DETAILED DESCRIPTION

[0019] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present invention, and are not intended to limit the protection scope of the embodiments of the present invention. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.

[0020] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] See also Figure 1 The embodiment of the present invention discloses an arbitrary polarization channel adaptive inversion method with local sampling and alternating iteration, comprising: Step S1, for any scene target, using a polarization camera to collect a plurality of polarization channel images of different directions and angles containing the scene target; Step S2, estimating global feature information according to each of the polarization channel images, performing cluster analysis using the K-Means method to obtain a clustering result image, and selecting a local sample point set according to the clustering result image; Step S3, setting an initial analysis angle according to the number of each polarization channel image; Step S4, based on the least squares method, performing matrix transformation on each of the polarization channel images according to the local sample point set and the initial analysis angle to obtain a multi-polarization channel matrix, and solving the multi-polarization channel matrix to obtain an initial polarization angle matrix and an intermediate analysis angle matrix; Step S5, based on the least square method, updating the intermediate polarization angle matrix according to the multi-polarization channel matrix and the intermediate analysis angle matrix and solving to obtain the intermediate polarization angle; Step S6, determine whether the preset convergence condition is met: If yes, go to step S7; If not, then the step S4 and the step S5 are iterated alternately; Step S7, obtaining a final analysis angle according to the final analysis angle matrix and the final polarization angle matrix after alternating iterations, obtaining the Stokes vector of the incident light according to the final analysis angle matrix and the multi-polarization channel matrix, and reconstructing a polarization degree image according to the Stokes vector of the incident light.

[0022] See also Figure 2In the embodiment of the present invention, an experimental data set is constructed and illustrated for a submarine model, wherein the imaging system mentioned in the present invention is a polarization imaging camera integrated with a rotating scanning polarization filter module, which can achieve 180 polarization channel image acquisitions with a minimum interval of 1° in the range of 0°~180°; the imaging detector is a CCD black-and-white industrial camera with an imaging resolution of 960×1280 and a spectral response range of 380~1050nm; in the multi-polarization channel image acquisition process, indoor ambient light is used for illumination, and the submarine model and the camera are kept still; in order to suppress the influence of random noise on image quality and establish a reliable real data set, 50 images are collected in the same polarization channel direction in the same scene and averaged.

[0023] In step S1, a polarization camera is used to collect four polarization channel images containing scene targets, wherein the polarization camera is a time-sharing system integrated with a rotating scanning polarization module, which drives the polarization optical device to rotate to , , and angular position (this embodiment introduces an angle deviation of ±5° on the basis of the traditional 0° / 45° / 90° / 135°, which is used as the unknown rotation angle in the actual working process; for any other angle deviation, the method proposed in the present invention is applicable), the obtained 4 polarization channel images with different directions and angles are , , and .

[0024] See also Figure 3 , which are 4 polarization channel images at different directions and angles obtained in the embodiment of the present invention. It can be seen that: with the change of the direction angle of the polarization optical element, the light intensity in the local area of ​​the submarine model shows a trend of changing from bright to dark, which indicates that the polarization imaging process has a modulation effect on the reflected light intensity, and the area with obvious light intensity change has more significant polarization characteristics.

[0025] To further illustrate the principle of the method of the present invention, combined with the Stokes-Muller matrix method, it is assumed in step S1 that the Stokes vector It can represent the polarization state of light. For polarization imaging optical systems, represents the Stokes vector of the incident light, Represents the Stokes vector of the outgoing light. When the light beam passes through the imaging system, the modulation effect of the system on the light can be expressed by the Mueller matrix, that is: ; According to the Mueller matrix of an ideal linear polarization device, the above formula can be expressed as: ; In the linear polarization imaging system, the photosensitive pixel unit of the imaging detector is only sensitive to the intensity of the incident light. N The light intensity of the corresponding pixel of the polarization channel image at different directions and angles It can be expressed as: ; Where n represents the polarization channel sequence, is the total number of polarization channels collected; m represents the pixel sequence in the image, is the total number of pixels in each polarization channel image, ; Indicates the polarization direction The measured intensity value corresponding to the mth pixel position at angle, ; Assume that the polarization degree corresponding to the mth pixel of the scene target in the detector is , the polarization angle is , then the Stokes vector of the incident light is It can be expressed as: ; Among them, the circular polarization parameter is ignored in the process of linear polarization imaging, that is, ; Therefore, combined with the above formula, the polarization channel images of different directions and angles are obtained It can also be expressed as: ; After simplification, we can get: .

[0026] Further, step S2 is described in steps, and step S2 includes: Step S21, estimating global feature information according to any polarization channel image; In step S1, a multi-polarization channel image sequence set is obtained. Based on the correlation between polarization channels, the global feature information is estimated, including: maximum intensity image , minimum intensity image , estimated intensity image and the estimated polarization image ,in, and It can be estimated by traversing the pixel values ​​contained in the image, that is: ; ; Degree of Polarization (DoP) is an indicator used to describe the degree of polarization of light, which can be measured by the maximum intensity image. and the minimum intensity image To calculate, therefore, the estimated polarization image It can be expressed as: ; in The value range is between [0,1], 1 represents completely polarized light, and 0 represents unpolarized light; Estimated intensity image It can be approximated by the average intensity value of the polarization image of multiple polarization channels, that is, calculating the average value of all polarization channels, that is: ; According to the above steps, the maximum intensity image of the embodiment of the present invention is , minimum intensity image , estimated intensity image and the estimated polarization image See also Figure 4 As shown, arranged from left to right, it can be seen that due to the inherent polarization properties of the material itself and the light field effect between the interface and the environment, these four images can fully display the differentiated feature information of different regions, which provides a reliable data basis for subsequent global feature analysis and regional sampling; Step S22: Collecting statistics on global feature information and setting the number of clusters for cluster analysis; Based on the global feature image estimated in step S21, it is combined into a feature vector of each pixel, that is: ; Among them, F is m ×4 matrix; In order to count the sum of the squares of the distances from each eigenvector in F to its cluster center, the K-Means method is used for cluster analysis. , randomly generated Cluster Center , where the minimized distance from each eigenvector to the cluster center can be expressed as: ; in, represents the indicator function, if belongs to cluster q, then ;otherwise ; After completing the cluster analysis using the K-Means method, the cluster label of pixel m can be obtained. , and form the clustering result image: ; This clustering result image represents the final classification of the clustering analysis of the global samples of the image based on the global feature information; S23: Searching for the sample point with the most significant polarization feature, and obtaining the corresponding optimal cluster number; The estimated polarization degree image obtained in step S21 Based on this, the sample points with the most significant polarization characteristics and their coordinate positions can be expressed as: ; Therefore, according to the clustering result image of step S22 , the sample point with the most significant polarization characteristics The optimal cluster label of (coordinate position: x=6; y=678) is , see Figure 5 As shown; S24: Obtain a local sample point set according to the optimal intra-cluster sample distribution; The clustering result image in step S22 and the optimal cluster label in step S23 are On the basis of , the in-cluster samples corresponding to the optimal cluster label, that is, the set of local sample points obtained is: ; in, S =713 is the total number of local sample points, and .

[0027] According to the content in step S22, in the clustering result image of the embodiment of the present invention, the polarization characteristics are related to the surface properties of the material, the angle of the incident light, and the angle of the surface normal; the sample points in different areas of the global image can also be clustered into the same cluster; the sample points in the local area can also present different polarization characteristics due to the influence of the target surface curvature, etc. Therefore, the local sampling method proposed in the present invention can effectively search for sample points with similar features globally, which provides a basis for subsequent angle inversion and feature reconstruction.

[0028] According to the content in step S24, the local sample point set mapping image and the Malus fitting curve of the embodiment of the present invention refer to Figure 6 As shown, the red area is the local sample point obtained, and the five-pointed stars of different colors represent the analysis points selected in each area. It can be seen that the sampled local data have similar polarization characteristics and the light intensity distribution curve has good consistency.

[0029] Furthermore, the initial analysis angle of any direction preset in step S3 is ,Right now: ,in, , N=4 is the number of polarization channels.

[0030] Further, in step S4, in combination with the local sample points obtained in step S24 and the initial analysis angle preset in step S3, in step S41, the expression of the channel polarization image can be rewritten as follows using the matrix method: ; Among them, matrix I represents the local samples obtained in step S24 corresponding to S × N Multi-polarization channel matrix; represents the intensity vector of the polarization channel image, ; Represents the cosine vector of the polarization angle corresponding to the polarization channel image, ; Represents the sine vector of the polarization angle corresponding to the polarization channel image, ; represents a unit vector; represents the cosine vector of the preset initial analysis angle, ; represents the sine vector of the preset initial analysis angle, ; Step S42, taking into account the multi-polarization channel matrix I and the initial analysis angle matrix As a known quantity, matrix transformation is performed through the least squares principle to solve the initial polarization angle matrix ,Right now: ; Step S43, based on the known multi-polarization channel matrix I and the intermediate polarization angle matrix ( is the initial polarization angle matrix) and the matrix transformation is performed through the least squares principle to solve the intermediate analysis angle matrix ,Right now: ; in, k represents the number of alternating iterations, k=0,1,2,..., K .

[0031] Further, step S5 is described in steps, and step S5 includes: S51: using the least square method, updating the intermediate polarization angle matrix; According to the known multi-polarization channel matrix I and the intermediate analysis angle matrix , matrix transformation is performed through the least squares principle to update the intermediate polarization angle matrix ,Right now: ; S52: Calculate the intermediate polarization angle; The intermediate polarization angle matrix obtained in step S51 And the definition of polarization angle, intermediate polarization angle It can be expressed as: ; in, and Represent the initial polarization angle matrix The second and third column elements of ; S53: Calculate the intermediate polarization angle matrix; Considering that updating the polarization angle matrix V can bring about a disturbance that meets the desired solution characteristics, achieve better and faster analytical convergence, and avoid falling into an infinite loop, therefore, in combination with the intermediate polarization angle obtained in step S52 As well as the definition of the polarization angle matrix, the calculated intermediate polarization angle matrix It can be expressed as: .

[0032] Furthermore, in step S6, when the convergence condition is met, the cross iteration stops, and the convergence condition is: ; in, represents the convergence accuracy, For the k The intermediate polarization angle calculated by the iteration, For the k - The intermediate polarization angle calculated for 1 iteration.

[0033] Further, step S7 is described in steps, and step S7 includes: S71: Calculate the final analysis angle; According to the intermediate analysis angle matrix and analysis angle definition solved in steps S5-S6, the final analysis angle can be expressed as: ; in and Respectively represent the second and third column elements of the final analysis angle matrix. The final analysis angle solved in this embodiment is: .

[0034] S72: Calculate Stokes parameter image; Combined with the final analysis angle obtained in step S71 And the definition of the analysis angle matrix, and finally the analysis angle matrix It can be expressed as: ; According to the known multi-polarization channel matrix I and the final analysis angle matrix , the matrix transformation is performed by the least squares method to calculate the Stokes vector of the incident light ,Right now: ; S73: Calculate polarization image; Stokes vector obtained from S72 , the polarization image can be reconstructed: ; According to the above steps, the Stokes parameters reconstructed by the embodiment of the present invention are Image, Stokes parameters , Image Stokes parameters Image and polarization Image as Figure 7 Arranged from left to right.

[0035] In order to verify the effectiveness and advancement of the technical solution of the present invention, the two methods described in the background technology are used as a comparison. Figure 1 The scenes were inverted at different polarization channel angles, and the results are shown in Table 1 below: Table 1 Comparison of polarization channel inversion angle results of different methods ; Among them, due to the complexity of the test atlas, the two methods in the background technology fell into the local minimum during the test, resulting in the inability to effectively converge the iterative error; the analysis angle inverted by Method 1 and Method 2 is solved on the basis of the local sampling method proposed in this method. When the analysis angle is arbitrarily selected, the average error between the analysis angle inverted by the present invention and the true analysis angle is only 0.64°, while Method 1 and Method 2 are difficult to realize iterative inversion of the analysis angle due to the complexity of the test scene, and the average errors based on the local adoption method proposed in the present invention are 2.61° and 6.34°, respectively.

[0036] In addition, in order to more intuitively compare the quality of polarization feature images reconstructed by different methods, this embodiment further compares the deviations occurring in the reconstruction process by calculating the difference between the reconstructed image and the real image. Figure 8 The residual images of polarization characteristics reconstructed by different methods show that the residual corresponding to the Stokes parameter image reconstructed by the method proposed in the present invention and the polarization degree image is the smallest, followed by method 1, and method 2 is the worst.

[0037] Finally, the image quality is quantitatively evaluated using indicators such as Root Mean Square Error (RMSE), Peak Signal to Noise Ratio (PSNR) and Gradient Magnitude Similarity Deviation (GMSD). RMSE is used to evaluate the error between the reference image and the reconstructed image. The smaller the RMSE value, the more similar the two images are and the higher the quality of the reconstructed image. PSNR is an indicator to measure the global difference between the reconstructed image and the reference image. The larger the PSNR value, the more similar the two images are and the higher the quality of the reconstructed image. GMSD is used to evaluate the difference between images based on the gradient information of the images. The smaller the GMSD value, the more similar the two images are, as shown in Table 2 below: Table 2 Comparison of evaluation index results of polarization feature image reconstruction by different methods ; Evaluation index results and results of different methods for reconstructing polarization feature images Figure 8 The visualization results are consistent with those of , that is, the method proposed in the present invention is superior to the existing methods in both the reconstructed Stokes parameter image and the polarization degree image, which further verifies the superiority of the technical solution of the present invention.

[0038] In the description of the present invention, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0039] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration, characterized in that: The following steps are involved: Step S1, for any scene target, using a polarization camera to collect a plurality of polarization channel images of different directions and angles containing the scene target; Step S2, estimating global feature information according to each of the polarization channel images, performing cluster analysis using the K-Means method to obtain a clustering result image, and selecting a local sample point set according to the clustering result image; Step S3, setting an initial analysis angle according to the number of each polarization channel image; Step S4, based on the least squares method, performing matrix transformation on each of the polarization channel images according to the local sample point set and the initial analysis angle to obtain a multi-polarization channel matrix, and solving the multi-polarization channel matrix to obtain an initial polarization angle matrix and an intermediate analysis angle matrix; Step S5, based on the least square method, updating the intermediate polarization angle matrix according to the multi-polarization channel matrix and the intermediate analysis angle matrix and solving to obtain the intermediate polarization angle; Step S6, determine whether the preset convergence condition is met: If yes, go to step S7; If not, then the step S4 and the step S5 are iterated alternately; Step S7, obtaining a final analysis angle according to the final analysis angle matrix and the final polarization angle matrix after alternating iterations, obtaining the Stokes vector of the incident light according to the final analysis angle matrix and the multi-polarization channel matrix, and reconstructing a polarization degree image according to the Stokes vector of the incident light.

2. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 1, characterized in that: In step S1, the polarization camera used integrates a rotatable scanning polarization module, and the polarization module is driven to rotate to The polarization channel images at different direction angles are obtained at the positions of the angles.

3. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 1, characterized in that: The step S2 comprises: Step S21, based on the polarization channel images and in combination with the correlation between polarization channels, obtaining a maximum intensity image, a minimum intensity image, an estimated intensity image and an estimated degree of polarization image as the global feature information; Step S22, obtaining a feature vector of each pixel in each polarization channel image according to the global feature information combination, setting the number of clusters to randomly generate multiple cluster centers, counting the minimum distance from each feature vector to each cluster center, and performing cluster analysis using the K-Means method to obtain a cluster label of each pixel to form the cluster result image; Step S23, obtaining a sample point with the most significant polarization feature and an optimal cluster label to which the sample point belongs according to the estimated polarization degree image; Step S24: obtaining the in-cluster samples corresponding to the optimal cluster label as the local sample point set.

4. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 3 is characterized in that: In step S21, the global feature information is obtained by the following calculation formula: ； in, represents the maximum intensity image; express The measured intensity value corresponding to the mth pixel position at angle, ; represents the minimum intensity image; representing the estimated intensity image; represents the estimated polarization degree image.

5. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 1, characterized in that: The step S4 comprises: Step S41, obtaining the corresponding multi-polarization channel matrix according to the local sample point set and the initial analysis angle; Step S42, based on the least square method, performing matrix transformation according to the multi-polarization channel matrix and the initial analysis angle matrix corresponding to the initial analysis angle to obtain the initial polarization angle matrix; Step S43, based on the least square method, matrix transformation is performed according to the intermediate polarization angle matrix after the multi-polarization channel matrix and the initial polarization angle matrix are changed to obtain the intermediate analysis angle matrix.

6. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 5, characterized in that: In step S41, the multi-polarization channel matrix is ​​obtained by the following calculation formula: ； in, represents the multi-polarization channel matrix; An intensity vector representing the polarization channel image; A cosine vector representing the polarization angle corresponding to the polarization channel image; A sine vector representing the polarization angle corresponding to the polarization channel image; a cosine vector representing the initial analysis angle; Represents the sine vector of the initial analysis angle.

7. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 1, characterized in that: The convergence condition in step S6 is: ； in, represents the intermediate polarization angle of the kth iteration; represents the intermediate polarization angle of the k-1th iteration; Indicates the preset convergence accuracy.

8. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 1, characterized in that: The step S7 comprises: Step S71, obtaining the final analysis angle according to the final analysis angle matrix and the final polarization angle matrix after alternating iterations; Step S72, performing matrix transformation based on the multi-polarization channel matrix and the final analysis angle matrix to obtain the Stokes vector of the incident light based on the least squares method; Step S73, reconstructing the polarization degree image according to the Stokes vector.

9. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 8, characterized in that: In step S71, the final analysis angle is obtained by the following calculation formula: ； in, represents the final analysis angle; represents the final polarization angle matrix; represents the final analysis angle matrix; represents the second column elements of the final analysis angle matrix; Represents the third column elements of the final analysis angle matrix.

10. The method for adaptive inversion of arbitrary polarization channels with local sampling and alternating iteration according to claim 8, characterized in that: In step S73, the polarization degree image is obtained by the following calculation formula: ； ； in, representing the polarization degree image; The Stokes vector representing the incident light; a first parameter representing the Stokes vector; a second parameter representing the Stokes vector; A third parameter representing the Stokes vector; represents the multi-polarization channel matrix; Denotes the final analysis angle matrix.

Images