Non-uniformity correction method for uncooled long-wave infrared split focal plane polarized detector

Abstract

The application provides a non-uniformity correction method for a non-refrigeration long-wave infrared focal plane polarization detector, and belongs to the technical field of infrared polarization imaging. In view of the problems that the existing correction method ignores noise heteroscedasticity and joint solving coupled iteration leads to poor stability, the application establishes a super-pixel linear response model, constructs a correction function containing a calibration matrix and an offset vector, weights each sample time-domain noise variance reciprocal, constructs a weighted least squares objective function, realizes differentiated weighting of parameter estimation, adopts a step-by-step optimization strategy to decouple the bivariate joint problem into two independent sub-problems, derives a closed-form analytical solution, and finally completes correction through linear transformation of the correction function. The application can adaptively suppress heteroscedastic noise interference, avoid numerical instability of iteration, and improve correction accuracy and robustness.

Description

Technical Field

[0001] This application belongs to the field of infrared polarization imaging technology, specifically relating to a method and system for correcting the non-uniformity of an uncooled long-wave infrared focal plane polarization detector. Background Technology

[0002] Infrared focal plane polarization detectors achieve synchronous acquisition of intensity and polarization information through micro-polarizer arrays. However, due to factors such as inconsistent pixel response, micro-polarizer defects, and temperature drift, the images have fixed pattern noise and stripe noise, which require non-uniformity correction.

[0003] Most existing calibration methods employ ordinary least squares (OLS) to solve for parameters, implicitly assuming that sample noise satisfies homoscedasticity. However, in practical engineering, due to factors such as variations in radiation intensity, focal plane temperature fluctuations, and high-frequency dark-field correction, the noise levels of different samples often differ, exhibiting heteroscedasticity. OLS cannot distinguish sample quality and is susceptible to interference from samples with low signal-to-noise ratios, leading to a decrease in calibration accuracy. Furthermore, existing methods jointly optimize the calibration matrix and offset vector, resulting in strong coupling between the two. The solution requires iterative or approximate processing, leading to limitations in calibration accuracy and poor numerical stability.

[0004] Therefore, there is an urgent need for a non-uniformity correction method that can adaptively handle sample heteroscedasticity and improve correction accuracy and numerical stability. Summary of the Invention

[0005] In a first aspect, embodiments of this application provide a method for correcting the non-uniformity of an uncooled long-wave infrared focal plane polarization detector, comprising the following steps: S1. Establish a linear response model for the superpixel to obtain the ideal response; S2. Construct a correction function to linearly transform the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; S3. Based on the dataset containing multiple independent measurements and the ideal response, a weighted least squares objective function is constructed, and a step-by-step optimization strategy is used to solve for the calibration matrix and offset vector; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the calibration function and the ideal response; S4. Use the solved correction function to perform a linear transformation on the actual response of the superpixel to be corrected, and output the corrected response.

[0006] Furthermore, the specific steps of step S1 are as follows: S11. The Stokes vector of the incident light is modulated by the Mueller matrix of the micro-polarizer array and converted into outgoing light, which then illuminates the focal plane array; S12. For the ideal polarization direction in a superpixel, it is... pixels Based on the gain coefficient of that pixel Bias coefficient and analysis vectors Establish pixel values Linear response model with respect to the incident Stokes vector: , in, For outgoing light; Mueller matrix for integrating micro-polarizers into detectors; The incident light Stokes vector; S13. Combine pixels with four different bias detection directions within a single superpixel to construct a linear response model for the superpixel: , in, Represents superpixel values. , , , These represent pixel values ​​for four different detection directions; The analysis matrix represents the superpixel. , , , These represent the analysis vectors for four different bias detection directions; This represents the bias vector of a superpixel. , , , These represent the bias coefficients for four different bias detection directions.

[0007] Furthermore, in the linear response model of the superpixel, the ideal bias vector of the superpixel is set to zero, thus obtaining the ideal response of the superpixel. : , in, This is the ideal analysis matrix.

[0008] Furthermore, the specific steps of step S2 are as follows: Constructing the correction function for superpixels This is used to perform a linear transformation on the actual response to obtain the corrected Stokes vector estimate: , in, For calibration matrix, This is the offset vector.

[0009] Furthermore, in step S3, the weighted least squares objective function is constructed based on the dataset containing multiple independent measurements and the ideal response. The specific steps are as follows: S31. Obtain a dataset containing N independent measurements. ,in The actual response of the kth independent measurement. Let be the Stokes vector of the incident light in the kth independent measurement; The dataset mentioned In the middle, the noise variance estimate of each superpixel of the measurement sample The variance of the four pixels within each superpixel over time , , , The mean was calculated as follows: ; S32. Construct the objective function that minimizes the weighted sum of squared residuals. : , in, Let be the sample weights for the k-th independent measurement; the objective function is used to solve for the optimal calibration matrix. and offset vector .

[0010] Furthermore, the sample weight mentioned in step S32 is specifically the reciprocal of the sample noise variance estimate: .

[0011] Furthermore, the step-by-step optimization strategy described in step S3 for solving the calibration matrix and offset vector is as follows: S33. Adjust the calibration matrix Treating it as a fixed parameter, the offset vector in the weighted least squares objective function Taking the partial derivative and setting it to zero, we get about The conditionally optimal solution is: , in, The weighted average of the ideal response. , This represents the ideal response value for the k-th measurement. This is the weighted average of the actual responses. , The actual response value of the k-th measurement; W is the sum of all weights. ; S34. Substitute the conditional optimal solution into the objective function. Eliminate offset vector This transforms the bivariate coupled optimization problem into one that depends only on the calibration matrix. Univariate objective function: , in, For the ideal response after decentralization, ; The actual response after decentralization ; S35. The univariate objective function is centered: The ideal response after decentralization The actual response after decentralization Perform horizontal splicing separately to obtain and ; Introducing a weight square root diagonal matrix, the objective function... Written in matrix form: , in, This is a diagonal matrix of weight square roots. ; tr(·) represents the trace of the matrix; A decentralized ideal response global matrix; The global matrix is ​​a decentralized actual response; S36. Take the matrix derivative of the objective function in matrix form and set it to zero to obtain the optimal solution of the calibration matrix. : , in, for and The weighted cross-covariance matrix between them; for Its own weighted autocovariance matrix; S37. Will Substituting the values ​​into the conditional optimal solution described in step S33, we obtain the optimal estimate of the offset vector. : .

[0012] Further, in step S36, the weighted cross-covariance matrix The calculation method is as follows: ; The weighted autocovariance matrix The calculation method is as follows: .

[0013] Furthermore, the specific steps of step S4 are as follows: S41. Based on the optimal solution of the calibration matrix Optimal estimation of offset vector The corrected function is obtained after solving: , in, This represents the superpixel value to be corrected. This represents the corrected Stokes vector estimate; S42. Obtain a dataset containing multiple independent measurements through a blackbody experiment. and ideal analysis matrix ; S43. The actual response of each superpixel output by the detector is linearly transformed using the solved correction function to obtain the corrected Stokes vector estimate, thus completing the non-uniformity correction of the image.

[0014] Furthermore, step S42 specifically includes the following steps: An infrared focal plane polarization detector is positioned directly against a high-temperature blackbody radiation source. It is rotated at predetermined angular intervals within the polarization detection angle range, with different blackbody temperatures set at each detection angle. Multiple frames of detector response data are collected for each state, and the pixel grayscale value of each superpixel is used as the actual response for that state. ; Based on the multi-frame detector response data continuously acquired under the same measurement state, the variance of the four pixels in the detection direction within each superpixel is calculated over time. The mean of the variances is used as the noise variance estimate for that superpixel, and the reciprocal of the noise variance estimate is used as the sample weight. ; Based on the radiation intensity value of the blackbody temperature and the current polarization analysis angle of the polarizer, the theoretical Stokes vector of the incident light under each measurement state is calculated, and used as... ; Based on the four ideal analysis directions of the superpixel (0°, 45°, 90°, and 135°), determine the ideal analysis vector corresponding to each direction. The four ideal analysis vectors are then combined to construct the ideal analysis matrix.

[0015] Secondly, embodiments of this application also provide a non-uniformity correction system for an uncooled long-wave infrared focal plane polarization detector, comprising: The response model building module is used to build a linear response model for superpixels and obtain the ideal response. A correction function construction module is used to construct a correction function for linearly transforming the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; The objective function construction module constructs a weighted least squares objective function based on a dataset containing multiple independent measurements and the ideal response, and solves for the calibration matrix and offset vector using a step-by-step optimization strategy; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the correction function and the ideal response; The calibration execution module is used to perform a linear transformation on the actual response of the superpixel to be calibrated using the solved calibration function, and output the calibrated response.

[0016] As can be seen from the above technical solutions, this application has the following advantages: The non-uniformity correction method and system for uncooled long-wave infrared focal plane polarization detectors provided in this application independently estimates the temporal noise variance for each measurement sample and constructs a weighted least squares objective function using the inverse of the variance as weights. This automatically suppresses the contribution of low signal-to-noise ratio samples and enhances the contribution of high signal-to-noise ratio samples during the solution of correction parameters, thereby adaptively matching the heteroscedasticity distribution of sample noise in actual engineering and avoiding the problem of decreased correction accuracy caused by equal-weight fitting in existing methods. Simultaneously, by decoupling the joint optimization problem of the calibration matrix and the offset vector into a step-by-step solution strategy, first finding the conditionally optimal solution of the offset vector with respect to the calibration matrix and then substituting it back to solve the closed-form analytical solution of the calibration matrix, no iteration or approximation is required. This avoids the numerical instability and convergence difficulties caused by coupled iteration in existing joint optimization methods, improving the accuracy and robustness of correction parameter estimation. Furthermore, this method is not dependent on specific hardware and has good versatility and industrial applicability. Attached Figure Description

[0017] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the non-uniformity correction method for the uncooled long-wave infrared focal plane polarization detector of the present invention.

[0019] Figure 2 This is a schematic diagram of the superpixel structure of the present invention.

[0020] Figure 3 This is a schematic diagram of the detector imaging process of the present invention. Detailed Implementation

[0021] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0022] This embodiment provides a non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector. It establishes a superpixel response model and constructs a correction function containing a calibration matrix and an offset vector. A weighted least-squares objective function is constructed using the reciprocal of the sample noise variance, allowing the correction parameter solution process to be weighted differently based on the reliability of each measurement sample. A step-by-step optimization strategy decouples the joint optimization problem of coupled variables into two independent sub-problems, deriving closed-form analytical solutions for the calibration matrix and offset vector. Finally, the solved correction function is used to linearly transform the actual response, suppressing the interference of heteroscedastic noise on the correction parameter estimation and improving the accuracy and robustness of the correction parameter estimation.

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 The diagram shows a flowchart of a non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector in a specific embodiment. The method includes the following steps: S1. Establish a linear response model for the superpixel to obtain the ideal response; It should be noted that establishing a superpixel linear response model transforms the complex nonlinear degradation problem into an analytical ideal benchmark, providing a clear target reference for subsequent correction; separating the inherent error of the system from random disturbances makes the correction direction based on evidence, avoids blind fitting, ensures the physical rationality and generalization ability of the correction, and improves the interpretability of the overall algorithm. S2. Construct a correction function to linearly transform the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; It should be noted that this step quantizes the correction target as an approximation of the ideal response. The calibration matrix in the correction function is used to compensate for the response differences between pixels, and the offset vector is used to compensate for the baseline shift caused by dark current and background radiation. S3. Based on the dataset containing multiple independent measurements and the ideal response, a weighted least squares objective function is constructed, and a step-by-step optimization strategy is used to solve for the calibration matrix and offset vector; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the calibration function and the ideal response; It should be noted that the weighted least squares and step-by-step optimization strategy is adopted to make full use of multiple independent measurement data and suppress the interference of outliers and noise on parameter estimation. The step-by-step solution reduces the optimization difficulty of coupled parameters and ensures convergence stability and computational efficiency. The weighting mechanism can adaptively allocate influence according to the measurement confidence, thereby obtaining a more robust and accurate calibration matrix and offset vector, and improving the calibration accuracy. S4. Use the solved correction function to perform a linear transformation on the actual response of the superpixel to be corrected, and output the corrected response; It should be noted that this step applies the obtained calibration matrix and offset vector to each superpixel in the image to be corrected. The non-uniformity correction of the entire image can be completed with a single linear transformation, which is computationally efficient and meets the requirements of real-time processing.

[0025] This embodiment avoids the numerical instability problem of joint iterative optimization in existing methods by establishing a superpixel response model, constructing an objective function by weighting with the inverse of the sample variance, and deriving a closed analytical solution by adopting a step-by-step optimization strategy. It achieves adaptive suppression of heteroscedastic noise and improves the correction accuracy and algorithm robustness.

[0026] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0027] S1. Establish a linear response model for the superpixel to obtain the ideal response; The specific steps of step S1 are as follows: S11. For example Figure 3 As shown, the Stokes vector of the incident light is modulated by the Mueller matrix of the micro-polarizer array and converted into outgoing light, which then illuminates the focal plane array. S12. For the ideal polarization direction in a superpixel, it is... pixels Based on the gain coefficient of that pixel Bias coefficient and analysis vectors Establish pixel values Linear response model with respect to the incident Stokes vector: , in, For outgoing light; Mueller matrix for integrating micro-polarizers into detectors; The incident light Stokes vector; S13. Combine pixels with four different bias detection directions within a single superpixel to construct a linear response model for the superpixel: , in, Represents superpixel values. , , , These represent the pixel values ​​in four different polarization directions. A schematic diagram of the superpixel structure is shown below. Figure 2 As shown; The analysis matrix represents the superpixel. , , , These represent the analysis vectors for four different bias detection directions; This represents the bias vector of a superpixel. , , , These represent the bias coefficients for four different detection directions; In the linear response model of the superpixel, the ideal bias vector of the superpixel is set to zero, thus obtaining the ideal response of the superpixel. : , in, This is the ideal analysis matrix.

[0028] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0029] S2. Construct a correction function to linearly transform the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; The specific steps of step S2 are as follows: Constructing the correction function for superpixels This is used to perform a linear transformation on the actual response to obtain the corrected Stokes vector estimate: , in, For calibration matrix, This is the offset vector.

[0030] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0031] S3. Based on the dataset containing multiple independent measurements and the ideal response, a weighted least squares objective function is constructed, and a step-by-step optimization strategy is used to solve for the calibration matrix and offset vector; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the calibration function and the ideal response; Step S3 involves constructing a weighted least squares objective function based on a dataset containing multiple independent measurements and the ideal response. The specific steps are as follows: S31. Obtain a dataset containing N independent measurements. ,in The actual response of the kth independent measurement. Let be the Stokes vector of the incident light in the kth independent measurement; For example, data is acquired using an infrared polarization calibration device, which includes an infrared polarization detector, a rotating platform equipped with a polarizer, and a blackbody radiation source. The polarization detector is an uncooled long-wave infrared focal plane polarization detector with a pixel response band of 7.5-14 μm and a resolution of 640×512 pixels. The rotating platform is an OptoSigma PH-50-ARS, and the polarizer mounted on it is a Thorlabs WP50H-Z with an extinction ratio of approximately 300:1. The blackbody radiation source is a DY-HT200M-100 type surface source blackbody with a temperature resolution of 0.1℃. Data acquisition process: The infrared polarization detector is positioned directly in front of the high-temperature blackbody radiation source. The polarizer is fixed to the rotating platform and placed between the infrared polarization detector and the blackbody. The blackbody radiation source is set to different temperatures, with a set temperature interval of 10° within the range of 30~60°. The polarizer detection angle is set between 0° and 180°, rotating 15° each time. 105 detector response data are collected under each specific temperature and polarizer detection angle combination. 100 of these data are used to calculate correction parameters, and 5 are used for testing. In the experiment, the detector's field of view does not exceed 2 / 3 of the central region of the blackbody source to ensure that the blackbody radiation is as uniform as possible. The dataset mentioned In the middle, the noise variance estimate of each superpixel of the measurement sample The variance of the four pixels within each superpixel over time , , , The mean was calculated as follows: ; S32. Construct the objective function that minimizes the weighted sum of squared residuals. : , in, Let be the sample weights for the k-th independent measurement; the objective function is used to solve for the optimal calibration matrix. and offset vector ; The sample weights mentioned in step S32 are specifically the reciprocals of the sample noise variance estimates: ; Step S3 describes the use of a step-by-step optimization strategy to solve for the calibration matrix and offset vector. The specific steps are as follows: S33. Adjust the calibration matrix Treating it as a fixed parameter, the offset vector in the weighted least squares objective function Taking the partial derivative and setting it to zero, we get about The conditionally optimal solution is: , in, The weighted average of the ideal response. , This represents the ideal response value for the k-th measurement. This is the weighted average of the actual responses. , The actual response value of the k-th measurement; W is the sum of all weights. ; S34. Substitute the conditional optimal solution into the objective function. Eliminate offset vector This transforms the bivariate coupled optimization problem into one that depends only on the calibration matrix. Univariate objective function: , in, For the ideal response after decentralization, ; The actual response after decentralization ; S35. The univariate objective function is centered: The ideal response after decentralization The actual response after decentralization Perform horizontal splicing separately to obtain and ; Introducing a weight square root diagonal matrix, the objective function... Written in matrix form: , in, This is a diagonal matrix of weight square roots. ; tr(·) represents the trace of the matrix; A decentralized ideal response global matrix; The global matrix is ​​a decentralized actual response; S36. Take the matrix derivative of the objective function in matrix form and set it to zero to obtain the optimal solution of the calibration matrix. : , in, for and The weighted cross-covariance matrix between them; for Its own weighted autocovariance matrix; In step S36, the weighted cross-covariance matrix The calculation method is as follows: ; The weighted autocovariance matrix The calculation method is as follows: ; S37. Will Substituting the values ​​into the conditional optimal solution described in step S33, we obtain the optimal estimate of the offset vector. : .

[0032] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0033] S4. Use the solved correction function to perform a linear transformation on the actual response of the superpixel to be corrected, and output the corrected response; The specific steps of step S4 are as follows: S41. Based on the optimal solution of the calibration matrix Optimal estimation of offset vector The corrected function is obtained after solving: , in, This represents the superpixel value to be corrected. This represents the corrected Stokes vector estimate; S42. Obtain a dataset containing multiple independent measurements through a blackbody experiment. and ideal analysis matrix ; Step S42 includes the following specific steps: An infrared focal plane polarization detector is positioned directly against a high-temperature blackbody radiation source. It is rotated at predetermined angular intervals within the polarization detection angle range, with different blackbody temperatures set at each detection angle. Multiple frames of detector response data are collected for each state, and the pixel grayscale value of each superpixel is used as the actual response for that state. ; Based on the multi-frame detector response data continuously acquired under the same measurement state, the variance of the four pixels in the detection direction within each superpixel is calculated over time. The mean of the variances is used as the noise variance estimate for that superpixel, and the reciprocal of the noise variance estimate is used as the sample weight. ; Based on the radiation intensity value of the blackbody temperature and the current polarization analysis angle of the polarizer, the theoretical Stokes vector of the incident light under each measurement state is calculated, and used as... ; Based on the four ideal analysis directions of the superpixel (0°, 45°, 90°, and 135°), determine the ideal analysis vector corresponding to each direction. The four ideal analysis vectors are then combined to construct the ideal analysis matrix. S43. The actual response of each superpixel output by the detector is linearly transformed using the solved correction function to obtain the corrected Stokes vector estimate, thus completing the non-uniformity correction of the image.

[0034] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0035] The following are embodiments of the non-uniformity correction system for an uncooled long-wave infrared focal plane polarization detector provided in this disclosure. This system and the non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the non-uniformity correction system for an uncooled long-wave infrared focal plane polarization detector, please refer to the embodiments of the non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector described above.

[0036] The system includes: The response model building module is used to build a linear response model for superpixels and obtain the ideal response. A correction function construction module is used to construct a correction function for linearly transforming the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; The objective function construction module constructs a weighted least squares objective function based on a dataset containing multiple independent measurements and the ideal response, and solves for the calibration matrix and offset vector using a step-by-step optimization strategy; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the correction function and the ideal response; The calibration execution module is used to perform a linear transformation on the actual response of the superpixel to be calibrated using the solved calibration function, and output the calibrated response.

[0037] This embodiment uses a response model building module to physically model superpixels, a correction function construction module to define a linear transformation including a calibration matrix and an offset vector, and an objective function construction module to construct a weighted least squares objective function using the inverse of the sample noise variance as weights, automatically suppressing the contribution of low signal-to-noise ratio samples. A step-by-step parameter solution module decouples the joint optimization into two independent sub-problems, solving them sequentially to obtain a closed-form analytical solution. The correction execution module performs a linear transformation on the actual response to complete the correction. These modules work collaboratively to achieve adaptive suppression of heteroscedastic noise, avoiding the numerical instability problems of joint iterative optimization and improving correction accuracy and robustness.

[0038] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for correcting the non-uniformity of an uncooled long-wave infrared focal plane polarization detector, characterized in that, Includes the following steps: S1. Establish a linear response model for the superpixel to obtain the ideal response; S2. Construct a correction function to linearly transform the actual response of a superpixel to an ideal response; the correction function includes a calibration matrix and an offset vector; S3. Based on the dataset containing multiple independent measurements and the ideal response, a weighted least squares objective function is constructed, and a step-by-step optimization strategy is used to solve for the calibration matrix and offset vector; the objective function is used to minimize the weighted sum of squared residuals between the Stokes vector estimate output by the calibration function and the ideal response; S4. Use the solved correction function to perform a linear transformation on the actual response of the superpixel to be corrected, and output the corrected response; The specific steps of step S1 are as follows: S11. The Stokes vector of the incident light is modulated by the Mueller matrix of the micro-polarizer array and converted into outgoing light, which then illuminates the focal plane array; S12. For the ideal polarization direction in a superpixel, it is... pixels Based on the gain coefficient of that pixel Bias coefficient and analysis vectors Establish pixel values Linear response model with respect to the incident Stokes vector: ， in, For outgoing light; Mueller matrix for integrating micro-polarizers into detectors; The incident light Stokes vector; S13. Combine pixels with four different bias detection directions within a single superpixel to construct a linear response model for the superpixel: ， in, Represents superpixel values. , , , These represent pixel values ​​for four different detection directions; The analysis matrix represents the superpixel. , , , These represent the analysis vectors for four different bias detection directions; This represents the bias vector of a superpixel. , , , These represent the bias coefficients for four different detection directions; The specific steps of step S2 are as follows: Constructing the correction function for superpixels This is used to perform a linear transformation on the actual response to obtain the corrected Stokes vector estimate: , in, For calibration matrix, It is the offset vector; Step S3 involves constructing a weighted least squares objective function based on a dataset containing multiple independent measurements and the ideal response. The specific steps are as follows: S31. Obtain a dataset containing N independent measurements. ,in The actual response of the kth independent measurement. Let be the Stokes vector of the incident light in the kth independent measurement; The dataset mentioned In the middle, the noise variance estimate of each superpixel of the measurement sample The variance of the four pixels within each superpixel over time , , , The mean was calculated as follows: ； S32. Construct the objective function that minimizes the weighted sum of squared residuals. : ， in, Let be the sample weights for the k-th independent measurement; the objective function is used to solve for the optimal calibration matrix. and offset vector .

2. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 1, characterized in that, In the linear response model of the superpixel, the bias vector of the superpixel is set. Using a zero vector, we obtain the ideal response of the superpixel. : ， in, This is the ideal analysis matrix.

3. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 1, characterized in that, The sample weights mentioned in step S32 are specifically the reciprocals of the sample noise variance estimates: .

4. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 1, characterized in that, Step S3 describes the use of a step-by-step optimization strategy to solve for the calibration matrix and offset vector. The specific steps are as follows: S33. Adjust the calibration matrix Treating it as a fixed parameter, the offset vector in the weighted least squares objective function Taking the partial derivative and setting it to zero, we get about The conditionally optimal solution is: ， in, The weighted average of the ideal response. , This represents the ideal response value for the k-th measurement. This is the weighted average of the actual responses. , The actual response value of the k-th measurement; W is the sum of all weights. ; S34. Substitute the conditional optimal solution into the objective function. Eliminate offset vector This transforms the bivariate coupled optimization problem into one that depends only on the calibration matrix. Univariate objective function: ， in, For the ideal response after decentralization, ; The actual response after decentralization ; S35. The univariate objective function is centered: The ideal response after decentralization The actual response after decentralization Perform horizontal splicing separately to obtain and ; Introducing a weight square root diagonal matrix, the objective function... Written in matrix form: ， in, This is a diagonal matrix of weight square roots. ; tr(·) represents the trace of the matrix; A decentralized ideal response global matrix; The global matrix is ​​a decentralized actual response; S36. Take the matrix derivative of the objective function in matrix form and set it to zero to obtain the optimal solution of the calibration matrix. : ， in, for and The weighted cross-covariance matrix between them; for Its own weighted autocovariance matrix; S37. Will Substituting the values ​​into the conditional optimal solution described in step S33, we obtain the optimal estimate of the offset vector. : 。 5. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 4, characterized in that, In step S36, the weighted cross-covariance matrix The calculation method is as follows: ； The weighted autocovariance matrix The calculation method is as follows: 。 6. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 4, characterized in that, The specific steps of step S4 are as follows: S41. Based on the optimal solution of the calibration matrix Optimal estimation of offset vector The corrected function is obtained after solving: ， in, This represents the superpixel value to be corrected. This represents the corrected Stokes vector estimate; S42. Obtain a dataset containing multiple independent measurements through a blackbody experiment. and ideal analysis matrix ; S43. The actual response of each superpixel output by the detector is linearly transformed using the solved correction function to obtain the corrected Stokes vector estimate, thus completing the non-uniformity correction of the image.

7. The non-uniformity correction method for an uncooled long-wave infrared focal plane polarization detector according to claim 6, characterized in that, Step S42 includes the following specific steps: An infrared focal plane polarization detector is positioned directly against a high-temperature blackbody radiation source. It is rotated at predetermined angular intervals within the polarization detection angle range, with different blackbody temperatures set at each detection angle. Multiple frames of detector response data are collected for each state, and the pixel grayscale value of each superpixel is used as the actual response for that state. ; Based on the multi-frame detector response data continuously acquired under the same measurement state, the variance of the four pixels in the detection direction within each superpixel is calculated over time. The mean of the variances is used as the noise variance estimate for that superpixel, and the reciprocal of the noise variance estimate is used as the sample weight. ; Based on the radiation intensity value of the blackbody temperature and the current polarization analysis angle of the polarizer, the theoretical Stokes vector of the incident light under each measurement state is calculated, and used as... ; Based on the four ideal analysis directions of the superpixel (0°, 45°, 90°, and 135°), determine the ideal analysis vector corresponding to each direction. Then, the four ideal analysis vectors are combined to construct the ideal analysis matrix.

Images