Dynamic image enhancement method and system based on liquid crystal polarization adaptive modulation

Abstract

The application provides a dynamic image enhancement method and system based on liquid crystal polarization adaptive modulation, and relates to the technical field of optical imaging and information processing. The method comprises the following steps: collecting target polarization observation images under illumination light of different polarization states, and constructing a polarization image sequence; establishing a mathematical model of a voltage-phase delay response model and a modulation intensity model containing a liquid crystal polarization modulator; fitting model parameters of the modulation intensity model according to the current polarization image sequence, and judging convergence by using a sampling uncertainty evaluation function; if convergence is achieved, the optimal model parameters of the modulation intensity model are determined; if convergence is not achieved, new target polarization observation images are collected and re-evaluated until convergence is achieved; and according to the optimal model parameters, a spatio-temporal joint constraint-based enhancement reconstruction model is used to reconstruct the current polarization image sequence, so as to generate an enhanced image and a dynamic enhancement video stream. The application can separate target signals and noise components in real time and with high precision, and realizes dynamic image enhancement.

Description

Technical Field

[0001] This invention relates to the field of optical imaging and information processing technology, and in particular to a dynamic image enhancement method and system based on liquid crystal polarization adaptive modulation. Background Technology

[0002] During imaging, the presence of suspended particles and dissolved substances (mostly underwater or in the air) causes significant absorption and scattering of light during propagation, leading to reduced contrast, blurred details, and decreased signal-to-noise ratio in the imaging results. In particular, the background light component introduced by multiple scattering during propagation severely interferes with the effective acquisition of the target signal, thus limiting the performance of traditional imaging methods in complex environments.

[0003] In recent years, polarization imaging technology has gradually become an important research direction for image enhancement due to its ability to distinguish the polarization characteristics of reflected light from a target from scattered background light. Existing techniques typically employ fixed polarization angles (e.g., 0°, 45°, 90°, 135°), lacking adaptive optimization mechanisms. Some technical literature has attempted to acquire single or multiple images under different polarization states, combining this with Stokes parameters for modeling and algorithm implementation to suppress scattering interference and improve target visibility. However, these methods are all ill-suited to various dynamic scenes and suffer from the following inherent limitations:

[0004] 1) Some applications use mechanically rotating polarizers for multiple sampling, but this method has poor stability and low efficiency;

[0005] 2) The above-mentioned method structure also does not fully incorporate mathematical constraints for modeling different scattering characteristics;

[0006] 3) Currently, enhancement algorithms based on multiple sampling are generally open-loop algorithms, which are designed for static analysis and processing static single-frame images, resulting in poor generalization.

[0007] 4) The implementation of the enhanced system and the algorithm processing are independent of each other, resulting in low overall efficiency and failing to achieve engineering implementation and system-level closed loop for dynamic real-time video.

[0008] In summary, existing methods typically employ fixed polarization sampling sequences for static targets. The lack of adaptive control over the sampling process leads to low sampling efficiency, excessive redundant information, and difficulty in achieving optimal reconstruction results in generalized complex dynamic environments.

[0009] Therefore, there is an urgent need for an image enhancement method and system that can combine polarization modulation and information-driven adaptive sampling mechanisms to reduce sampling redundancy and improve parameter estimation accuracy while ensuring imaging quality, so as to meet the needs of dynamic high-quality imaging in complex environments. Summary of the Invention

[0010] To address the shortcomings of the existing technologies, this invention utilizes the characteristic of liquid crystals to rapidly modulate polarized light (modulation speed of ordinary liquid crystals is 1 ms–20 ms, and that of fast liquid crystals is 10–100 μs). Combined with a noise convergence algorithm, and by constructing an integrated mathematical model of "polarization modulation—observation modeling—statistical estimation—closed-loop control", a dynamic image enhancement method based on liquid crystal polarization adaptive modulation is proposed. This method aims to achieve real-time, dynamic, and high-precision separation of target signals and noise components.

[0011] On the one hand, this invention proposes a dynamic image enhancement method based on liquid crystal polarization adaptive modulation, which includes the following process:

[0012] Collect target polarization observation images under illumination with different polarization states, and construct a polarization image sequence;

[0013] A mathematical model is established between the polarization observation image and the target signal and noise components, including: a liquid crystal polarization modulation model and a modulation intensity model;

[0014] The polarization image sequence acquired at the current moment is input into the mathematical model between the polarization observation image and the target signal and noise components, and the model parameters of the modulation intensity model are fitted by the parameter estimation method to obtain the estimated values ​​of the model parameters.

[0015] The obtained model parameter estimates are substituted into the mathematical model between the polarization observation image and the target signal and noise components, and the global error of the current model parameter estimates is evaluated using the sampling uncertainty evaluation function.

[0016] If the sampling uncertainty evaluation function converges, the current model parameter estimate will be used as the optimal model parameter for the modulation intensity model.

[0017] If the sampling uncertainty evaluation function does not converge, the polarization image sequence is updated by acquiring new polarization observation images of the target under the illumination light of the new polarization state, and then the model parameter estimates are obtained again and evaluated until the sampling uncertainty evaluation function converges.

[0018] Based on the optimal model parameters, an enhanced reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarization image sequence acquired at the current moment, thereby obtaining the enhanced target image and forming a dynamically enhanced video stream.

[0019] Furthermore, the specific method for acquiring target polarization observation images under illumination light with different polarization states and constructing a polarization image sequence is as follows:

[0020] By applying different driving voltages to the liquid crystal polarization modulator, illumination light with different polarization states is generated;

[0021] A polarizer with a known polarization angle is set at the image sampling receiver, and target polarization observation images under illumination light of each polarization state are acquired accordingly; the target polarization observation images contain light intensity distribution information.

[0022] A polarization image sequence is formed by collecting target polarization observation images under illumination of all polarization states.

[0023] Furthermore, the specific method for establishing the mathematical model between the polarization observation image and the target signal and noise components is as follows:

[0024] A liquid crystal polarization modulation model is established; the liquid crystal polarization modulation model characterizes the actual polarization modulation amount under the combined effect of the current driving voltage and the current ambient temperature as the superposition of the ideal voltage-polarization modulation amount mapping function of the liquid crystal polarization modulator and the nonlinear compensation term; the nonlinear compensation term is used to correct the influence of ambient temperature and the response hysteresis error of liquid crystal molecules.

[0025] Based on the liquid crystal polarization modulation model, a modulation intensity model is established under the condition that the polarization angle of the analyzer is known;

[0026] The modulation intensity model characterizes the image intensity of each polarization sample as the sum of the depolarized background noise component, the polarized target signal component containing the target signal modulation coefficient, and the random additive noise during the photoelectric detection process.

[0027] Furthermore, the parameters to be solved in the modulation intensity model include: the depolarized background noise component and the modulation coefficient of the target signal.

[0028] Furthermore, the obtained model parameter estimates are substituted into the mathematical model between the polarization observation image and the target signal and noise components, and evaluated using a sampling uncertainty evaluation function. The specific method is as follows:

[0029] Substitute the preliminary estimates of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components, and calculate the least squares estimated covariance matrix.

[0030] The sampling uncertainty evaluation function is constructed by estimating the trace of the covariance matrix using least squares, and the sampling certainty evaluation index at the current time is calculated.

[0031] If the sampling certainty evaluation index at the current moment and the sampling certainty evaluation index at the previous moment satisfy the convergence criterion, then the sampling uncertainty evaluation function is considered to be convergent; otherwise, the sampling uncertainty evaluation function is considered not to be convergent.

[0032] Furthermore, the specific method for substituting the obtained preliminary estimates of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components to calculate the least squares estimated covariance matrix is ​​as follows:

[0033] Using the preliminary estimates of the model parameters obtained, the vector of parameters to be determined is constructed in real time.

[0034] Based on the actual polarization modulation amount of each polarization sample, construct the polarization physical basis for polarization sampling.

[0035] Based on the parameter vector to be determined, the instantaneous noise variance is evaluated in real time using the forward fitting residual.

[0036] Based on the number of image frames in the polarization image sequence acquired at the current moment, the least squares estimation covariance matrix is ​​constructed in real time using the polarization physical basis and the instantaneous noise variance.

[0037] Furthermore, the specific method for evaluating the instantaneous noise variance in real time using the forward fitting residual based on the parameter vector to be determined is as follows:

[0038] Obtain the parameter vector to be determined, constructed based on the polarization image sequence acquired in the previous moment;

[0039] Substitute the parameter vector to be determined into the mathematical model between the polarization observation image and the target signal and noise components, and combine it with the actual polarization modulation amount of the current polarization sampling to calculate the theoretical pixel intensity value under the current modulation state.

[0040] Calculate the squared residual between the actual image intensity and the theoretical pixel intensity value of the current polarization sample pixel by pixel, and then calculate the average squared residual of all pixels in the target polarization observation image of the current polarization sample as the instantaneous noise variance.

[0041] Furthermore, the specific method for reconstructing the polarization image sequence acquired at the current moment using an enhancement reconstruction model based on spatiotemporal joint constraints according to the optimal model parameters to obtain the enhanced target image is as follows:

[0042] Obtain the maximum intensity value corresponding to each pixel position in the polarization image sequence acquired at the current moment;

[0043] Based on the maximum intensity value corresponding to each pixel position and the depolarized background noise component in the optimal model parameters, an enhancement reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarized image sequence acquired at the current moment to obtain the enhanced target image.

[0044] The enhanced reconstruction model based on spatiotemporal joint constraints is as follows: the enhanced intensity value of a pixel is represented as the difference between the maximum intensity value of a pixel in the polarization image sequence acquired at the current moment and the depolarization background noise component after background suppression parameter adjustment.

[0045] The enhanced target images are output in chronological order to form a dynamically enhanced video stream.

[0046] Furthermore, the background suppression parameter is determined using a temporal coherence penalty mechanism;

[0047] For the polarization image sequence acquired at the current moment, the optimization model for the background suppression parameters of the current frame is defined as:

[0048] Within a preset search interval, find the current frame background suppression parameter value that optimizes the target evaluation function;

[0049] The objective evaluation function is composed of a spatial information gain term and a temporal coherence smoothing term;

[0050] Among them, the spatial information gain term is obtained by weighted normalization of the information entropy and average gradient of the enhanced target image;

[0051] The temporal coherence smoothing term is determined by the background suppression parameters of the current frame and the previous frame, combined with the temporal coherence constraint weights, and is used to limit parameter abrupt changes between adjacent frames.

[0052] On the other hand, the present invention proposes a dynamic image enhancement system based on liquid crystal polarization adaptive modulation, the system comprising:

[0053] The polarization modulation module is used to generate illumination light with different polarization states, and when it receives the polarization modulation drive signal, it generates illumination light with a new polarization state according to the polarization state optimization criterion.

[0054] The imaging acquisition module is used to acquire target polarization observation images under illumination light with different polarization states and construct a polarization image sequence; and when an image acquisition trigger signal is received, it acquires target polarization observation images under illumination light with new polarization states and updates the polarization image sequence.

[0055] The model estimation module is used to input the polarization image sequence acquired at the current time into the mathematical model between the polarization observation image and the target signal and noise components, and to use the parameter estimation method to fit the model parameters of the modulation intensity model in the mathematical model to obtain the estimated values ​​of the model parameters;

[0056] An adaptive sampling control module is used to substitute the estimated model parameters into the mathematical model and evaluate it using a sampling uncertainty evaluation function: if the function converges, the current estimated model parameters are used as the optimal model parameters for the modulation intensity model; if the function does not converge, a control feedback signal is generated and output to the control module.

[0057] The control module is used to generate a polarization modulation drive signal and an image acquisition trigger signal respectively based on the received control feedback signal and a preset delay parameter, and send the polarization modulation drive signal to the polarization modulation module and the image acquisition trigger signal to the imaging acquisition module.

[0058] The image reconstruction module is used to reconstruct the polarization image sequence acquired at the current moment using an enhancement reconstruction model based on spatiotemporal joint constraints, according to the optimal model parameters, to obtain the enhanced target image and form a dynamically enhanced video stream.

[0059] The beneficial effects of adopting the above technical solution are as follows:

[0060] This invention integrates three major technologies: polarization imaging, optical detection, and computational imaging. It encompasses key techniques such as polarization state manipulation, polarization response modeling, and adaptive optimization, making it suitable for scenarios involving visual enhancement and eliminating the effects of medium scattering. Compared to existing technologies, this invention offers the following advantages:

[0061] (1) The method of the present invention utilizes liquid crystal polarization modulation technology to achieve fast and stable polarization modulation without mechanical structure, and is applied to dynamic video with stable system output and high frame rate.

[0062] (2) The method of this invention introduces a closed-loop control mechanism based on statistical convergence. During the sampling process, it no longer relies on a fixed number of samples, but instead uses the covariance matrix to evaluate the uncertainty of the current estimation result in real time, and uses its trace as a global error metric. When this metric converges to a stable state, the sampling process is automatically terminated. This method can adaptively adjust the number of sampling (depth) numbers according to environmental complexity, significantly reducing sampling redundancy while ensuring separation accuracy, improving system real-time performance and engineering adaptability, and demonstrating significant technological advancement and engineering application value.

[0063] (3) The method of the present invention combines scattering physics constraints to perform differentiated modeling of target signal and noise components, thereby effectively suppressing interference components introduced by scattering medium and significantly improving the separation accuracy of target signal and noise components.

[0064] (4) The present invention implements a system hardware and algorithm co-design, that is, the liquid crystal polarization modulation device and the control algorithm are deeply coupled, giving full play to the complementary advantages of hardware modulation speed and algorithm processing efficiency, and greatly improving the overall system performance and robustness.

[0065] (5) The method and system of this invention can be extended to other fields for spectral multimodal fusion applications, and have wide applicability. Specifically, it can be applied to the following scenarios: underwater dynamic imaging or atmospheric scattering environments (such as haze and dust) to improve imaging quality; fusion with time-of-flight (ToF) ranging technology to improve phase measurement stability through polarization information; and application in scenarios such as medical imaging and industrial inspection that require suppression of scattering interference. Therefore, this invention is not limited to the environment, but also has the ability to be promoted across scenarios. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the dynamic image enhancement method based on liquid crystal polarization adaptive modulation in this embodiment.

[0067] Figure 2 This is a schematic diagram of liquid crystal polarization modulation of the liquid crystal polarization modulator in this embodiment;

[0068] Figure 3 This is a schematic diagram of the dynamic image enhancement system based on liquid crystal polarization adaptive modulation in this embodiment;

[0069] In the diagram: 1-Polarizing light source; 2-Liquid crystal; 3-Controller. Detailed Implementation

[0070] To facilitate understanding of this application, specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following embodiments are illustrative of the invention but are not intended to limit its scope. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0071] Example 1:

[0072] This embodiment presents a dynamic image enhancement method based on liquid crystal polarization adaptive modulation, such as... Figure 1 As shown, the method includes the following steps:

[0073] Collect target polarization observation images under illumination with different polarization states, and construct a polarization image sequence;

[0074] The specific method for acquiring target polarization observation images under illumination with different polarization states and constructing a polarization image sequence is as follows:

[0075] Applying different driving voltages to a liquid crystal polarization modulator produces illumination light with different polarization states.

[0076] The analyzer angle is set at the image sampling receiver. The analyzer is used to acquire target polarization observation images under illumination light of various polarization states; the target polarization observation images contain light intensity distribution information.

[0077] In this embodiment, as Figure 2 As shown, polarization source 1 emits illumination light with a fixed polarization angle or phase. The illumination light passes through liquid crystal 2, which, under the control of controller 3, modulates the polarization state of the illumination light. The liquid crystal polarization modulation process rotates the originally fixed polarized light angle to a specified angle (or a specified phase delay), changing the polarization state. In the initial stage of polarization sampling, a preset uniform polarization state distribution is used for rapid scanning to obtain the basic observation image (i.e., the target polarization observation image), while ensuring the identifiability of the parameter model.

[0078] Specifically, since the basic observation image is essentially a matrix equation containing noise and useful signals, sufficient independent observation variables are needed in the initial stage of an unknown environment to ensure that the equation is solvable; generally speaking, the more samples, the higher the solution accuracy. Therefore, in order to ensure that the parametric equation is solvable, solving the polarization parameters requires at least 3-4 independent observation variables, that is, simultaneously acquiring multiple target polarization observation images under illumination with different and known polarization states.

[0079] In this embodiment, if it is the liquid crystal modulation angle, the polarization modulation angle of the illumination end is generally preset to 0°, 45°, 90°, and 135°, a total of four angles, for initial sampling and initial stage solution. However, the standard signal and noise components obtained at this time have a large error compared to the true value, and it is necessary to continue to input a large number of independent observation variables for solution. (Analysis angle) Depending on the selection of the analyzer, in this embodiment, the analyzer angle is... Set to one or more of the following angles: 0°, 45°, 90°, and 135°.

[0080] However, too many samples can lead to significant redundancy. Therefore, this embodiment proposes a feedback mechanism: when the noise solution reaches the expected level, sample supply is stopped to achieve optimization. After obtaining the initial estimation results, the subsequent sampling process is adaptively adjusted based on the error evaluation function and parameter uncertainty analysis results. This adjustment includes, but is not limited to, the number of samples and the selection of polarization states, thereby avoiding redundant sampling and improving information utilization efficiency. This strategy realizes the transition from fixed uniform sampling to data-driven adaptive sampling.

[0081] A polarization image sequence is formed by collecting target polarization observation images under illumination of all polarization states.

[0082] It should be noted that in this embodiment, each polarization sampling corresponds to a specific polarization state of illumination light, and also corresponds to acquiring one frame of target polarization observation image. As the sampling process proceeds, different polarization states are switched sequentially, and one frame of target polarization observation image is acquired each time a switch is performed. This frame of target polarization observation image is then added to the polarization image sequence according to the acquisition order. In the adaptive sampling process, several preset polarization states (such as the aforementioned four different angles) are first used for initial sampling to obtain an initial polarization image sequence with the corresponding number of frames. Subsequently, the uncertainty of the current estimation result (model parameter estimate) is used to determine whether further sampling is needed. If convergence is not achieved, a new polarization state is added, and a new frame of target polarization observation image is acquired and added to the polarization image sequence. At this time, the number of samplings, the total number of polarization states, and the number of frames in the image sequence increase synchronously. This process is repeated until the convergence condition is met, at which point the optimal parameter estimation result is output.

[0083] A mathematical model is established between the polarization observation image and the target signal and noise components, including: a liquid crystal polarization modulation model and a modulation intensity model.

[0084] In practical systems, the modulation effect of a liquid crystal polarization modulator needs to consider actual environmental factors. Its output polarization state depends not only on the driving voltage but also on temperature changes and response hysteresis. If an ideal polarization model is directly used, the basis vector of subsequent parameter estimation will be distorted, thus affecting the separation accuracy of the target signal and noise components. Therefore, this embodiment first establishes a nonlinear mapping relationship from the driving voltage to the actual polarization response, i.e., a physical mapping and compensation model of liquid crystal driving-polarization response, and introduces a compensation term to correct this error, thereby ensuring the accuracy of the observation model.

[0085] The specific method for establishing the mathematical model between the polarization observation image and the target signal and noise components is as follows:

[0086] A liquid crystal polarization modulation model is established; the liquid crystal polarization modulation model characterizes the actual polarization modulation amount under the combined effect of the current driving voltage and the current ambient temperature as the superposition of the ideal voltage-polarization modulation amount mapping function of the liquid crystal polarization modulator and the nonlinear compensation term; the nonlinear compensation term is used to correct the influence of ambient temperature and the response hysteresis error of liquid crystal molecules, and is expressed as:

[0087] (1)

[0088] in, For the first The driving voltage applied to the liquid crystal polarization modulator during secondary polarization sampling; The current ambient temperature is used. Due to the temperature drift characteristics of liquid crystal molecules, this embodiment will... As a compensation variable; Indicates the driving voltage and current ambient temperature The actual polarization modulation amount under the combined effect; This is the mapping function of the ideal voltage-polarization modulation amount of the liquid crystal polarization modulator; This is a nonlinear compensation term used to correct for the effects of ambient temperature and the response hysteresis error of liquid crystal molecules.

[0089] Unlike the ideal polarization state assumption, this embodiment first establishes the relationship between the liquid crystal polarization modulator and the driving voltage. With actual polarization modulation amount The underlying physical mapping is described. Meanwhile, considering the influence of ambient temperature and response hysteresis on liquid crystal molecules, this embodiment introduces a nonlinear compensation term to correct for the effects of ambient temperature and the response hysteresis error of the liquid crystal molecules. The nonlinear compensation term is calibrated through actual measurement.

[0090] Based on the aforementioned liquid crystal polarization modulation model, under a known polarization detection angle... Under these conditions, a modulation intensity model is established; the modulation intensity model characterizes the image intensity of each polarization sample as the sum of the depolarized background noise component, the polarized target signal component containing the target signal modulation coefficient, and the random additive noise during the photoelectric detection process, expressed as:

[0091] (2)

[0092] in, Indicates the detection angle is At that time, the first The image intensity of the second polarization sample, i.e., the first polarization sample Modulation intensity of secondary polarization sampling; This represents the background noise component that is depolarized, which is the noise portion in the image that is not modulated by polarization. and These are all target signal modulation coefficients, used to reflect the intensity of the polarized light component carrying the target signal; This refers to random additive noise in the photoelectric detection process, such as thermal noise and shot noise.

[0093] In this embodiment, the liquid crystal polarization modulation model ensures the accuracy of the basis vector for subsequent signal separation by physically compensating the driving voltage. Based on image data acquired under different polarization states, a polarization response mathematical model conforming to formula (2), namely the modulation intensity model, is constructed. This model expresses the intensity of the image light received by polarization detection as a function of the polarization state modulation amount (such as a trigonometric function response), and describes the evolution law of the polarization state of light in the scattering medium through a polarization propagation model. The main purpose of the above polarization response modeling is to distinguish the response characteristics of the target signal of strongly polarized light and the noise component of weakly polarized light.

[0094] The model parameters of the modulation intensity model include: depolarized background noise components. and target signal modulation coefficients and .

[0095] In this embodiment, the modulation intensity model can be implemented using a Stokes model to characterize the polarization state and establish a functional relationship between image intensity and polarization state. This embodiment uses a Stokes parameter model to characterize the polarization state and combines least squares or equivalent optimization methods to fit the model parameters, thereby obtaining the response characteristics of the target signal and noise components. Specifically, by analyzing the differences in statistical distribution, variation trend, and response sensitivity of image intensity under multiple modulation states, modulation response consistency or difference constraints are introduced to achieve direct separation of the target signal and noise components. This method avoids the dependence on model accuracy and noise sensitivity in traditional parameter inversion processes and has stronger robustness and adaptability in complex scattering environments.

[0096] It should be noted that, in addition to the Stokes parameter-based modeling and solution method, this embodiment also supports modeling methods that do not rely on the inversion of complete polarization parameters, such as deep learning. In this type of method, the Stokes vector is not directly solved; instead, the response differences of the observed signal under different polarization modulation states are used to construct an equivalent observation space separation model. This modeling process combines polarization modulation with scattering physics, providing a physical constraint basis for separation and reconstruction.

[0097] The modulation intensity model and polarization response model established above provide mathematical tools and physical prerequisites for separating the target signal from the noise component in the target polarization observation image. In practice, the basis for parameter inversion is first established based on the above mathematical model; then, observation image data from multiple frames under different polarization states are substituted into the model. Since target reflection usually has strong polarization response characteristics (subject to modulation changes), while background scattered light has a low degree of polarization (exhibiting a constant state without polarization), based on this difference in physical response, the model parameters in the model are calculated through optimization solution methods (such as parameter estimation), thereby achieving effective decoupling and separation of the target signal from the noise component at both mathematical and physical levels. This process is essentially a physically constrained parameter inversion process.

[0098] It should be noted that this information separation process does not rely on a specific analytical model of polarization parameters or a fixed algorithm, but is based on the separability of polarization observation image data in statistical regularities or feature spaces. Specifically, the information separation process can employ methods including, but not limited to, the following:

[0099] (1) Statistical feature-based analysis methods, including analysis of variance, analysis of covariance, principal component analysis or other statistical modeling methods, can extract stable information by mining the differences in the statistical distribution of signals and noise under different modulation states.

[0100] (2) Data-driven learning methods, including supervised or unsupervised machine learning models, achieve separation of target signal and noise components by mapping and learning multi-modulation observation sequences.

[0101] (3) Signal separation method combined with physical constraints, that is, using physical laws to establish a physical imaging degradation model of underwater or scattering media, and clarifying through the degradation model that the received image response is composed of the superposition of the target signal and the backscattered (or aerosol scattering) noise component. In the absence of solving the Stokes polarization parameter, the difference in image response under different modulation states can also be introduced as a constraint condition to achieve selective enhancement of information.

[0102] The three types of methods mentioned above correspond to three technical paths: parameter inversion model, data-driven model, and physical constraint-driven model, respectively.

[0103] Furthermore, all the aforementioned methods are equivalent implementations within the technical framework of this invention. Although the principles and algorithms used differ, they can all utilize the multidimensional observation results formed by their respective modulation responses to achieve adaptive separation of the target signal and noise components. Therefore, this invention is not limited to a specific method for separating the target signal and noise components.

[0104] The polarization image sequence acquired at the current moment is input into the mathematical model between the polarization observation image and the target signal and noise components, and the model parameters of the modulation intensity model are fitted by the parameter estimation method to obtain the estimated values ​​of the model parameters.

[0105] In this embodiment, all the collected target polarization observation images are sequentially input into the mathematical model between the polarization observation image and the target signal and noise components. The model parameters of the modulation intensity model determined by formula (2) are solved by matrix to obtain the estimated values ​​of the model parameters.

[0106] The obtained model parameter estimates are substituted into the mathematical model between the polarization observation image and the target signal and noise components, and the sampling uncertainty evaluation function is used for evaluation:

[0107] If the sampling uncertainty evaluation function converges, the current model parameter estimate will be used as the optimal model parameter for the modulation intensity model.

[0108] If the sampling uncertainty evaluation function does not converge, the polarization image sequence is updated by acquiring new polarization observation images of the target polarization under new polarization illumination light, and then the model parameter estimates are obtained and evaluated again until the sampling uncertainty evaluation function converges.

[0109] Specifically, the method for substituting the obtained model parameter estimates into the mathematical model between the polarization observation image and the target signal and noise components, and then evaluating it using the sampling uncertainty evaluation function, is as follows:

[0110] Substitute the preliminary estimates of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components to calculate the least squares estimated covariance matrix.

[0111] The specific method for substituting the preliminary estimated values ​​of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components to calculate the least squares estimated covariance matrix is ​​as follows:

[0112] Using the preliminary estimates of the model parameters, construct the vector of parameters to be determined in real time. .

[0113] According to the Actual polarization modulation amount of secondary polarization sampling , construct the first Polarization physical basis during sub-polarization sampling .

[0114] Based on the parameter vector to be determined The instantaneous noise variance is evaluated in real time using forward fitting residuals. .

[0115] The parameter vector to be determined The instantaneous noise variance is evaluated in real time using forward fitting residuals. The specific method is as follows:

[0116] Obtain the parameter vector to be determined, constructed from the polarization image sequence acquired at the previous moment. .

[0117] The parameter vector to be determined Substituting the mathematical model between the polarization observation image and the target signal and noise components, and combining it with the actual polarization modulation amount of the current polarization sampling, Calculate the theoretical pixel intensity value under the current modulation state.

[0118] Calculate the squared residual between the actual image intensity and the theoretical pixel intensity value of the current polarization sample pixel by pixel, and then calculate the average squared residual of all pixels in the target polarization observation image of the current polarization sample as the instantaneous noise variance.

[0119] In this embodiment, to achieve model self-consistency, the instantaneous noise variance is... The forward fit residuals are used for real-time evaluation to measure the residual level of the model fit, and are expressed as follows:

[0120] (3)

[0121] in, and These are the resolution width and height of the target polarization observation image, respectively, used to represent the number of pixel rows and columns; Pixels representing the polarization observation image of the target; This indicates the number of image frames in the polarization image sequence that have been acquired at the current moment; Indicates the first In the frame target polarization observation image The actual image intensity at that location; This indicates that the parameter vector to be determined Substituting the mathematical model between the polarization observation image and the target signal and noise components determined by formula (2), and combining it with the actual polarization modulation amount of the current polarization sampling... The theoretical pixel intensity value under this modulation state is calculated. Then, by comparing the deviation between the actual observed value (i.e., the actual image intensity) and this predicted value, the instantaneous noise variance is assessed. Real-time evaluation. It should be noted that in formula (2), The sampling sequence number indicates the number of samples in a complete sampling sequence. The image acquired the next time. When running to the current image in the polarization image sequence... When observing the target polarization image, the data is accumulated from frame 1 to frame 2. The target polarization observation image of the frame is used to determine the parameter vector using the least squares method. An estimate is made. Accordingly, in formula (4) That is, based on the results of previous observations, the current... Predicted frame intensity.

[0122] Based on the number of image frames in the polarization image sequence acquired at the current moment. Utilizing the polarization physical basis and instantaneous noise variance Real-time construction of least squares estimation of covariance matrix .

[0123] In this embodiment, while performing continuous polarization modulation, the noise component of the polarization modulation, namely the background noise component of depolarization, is known. The intensity of the polarized light component carrying the target signal, i.e., the target signal modulation coefficient. and Then, the least squares estimation covariance matrix is ​​constructed in real time, expressed as:

[0124] (4)

[0125] in, For based on The parameter vector to be determined from the target polarization observation image of the frame The least squares estimate of the covariance matrix.

[0126] The sampling uncertainty evaluation function is constructed by estimating the trace of the covariance matrix using least squares, and the sampling certainty evaluation index at the current time is calculated.

[0127] In this embodiment, the sampling uncertainty evaluation function is used to evaluate the current sampling effect, that is, to quantitatively evaluate the overall uncertainty of the model parameter estimation for the currently acquired multiple frames of images, representing the sum of all parameter estimation errors, and is expressed as:

[0128] (5)

[0129] in, This represents the sampling certainty (Uncertainty / Certainty) evaluation metric at the current moment, and It is a scalar used to quantify the overall uncertainty of the current model parameter estimates; The trace operation represents the sum of the elements on the main diagonal of a matrix.

[0130] In this embodiment, the physical meaning of the sampling uncertainty evaluation function is: It is equivalent to the sum of the estimation errors of each parameter, when The smaller the value, the more accurate the system's understanding of the physical characteristics of the current scene, and the more reliable the estimation results.

[0131] If the sampling certainty evaluation index at the current moment and the sampling certainty evaluation index at the previous moment satisfy the convergence criterion, then the sampling uncertainty evaluation function is considered to be convergent; otherwise, the sampling uncertainty evaluation function is considered not to be convergent.

[0132] In this embodiment, the relative rate of change of the sampling certainty evaluation index between adjacent samples is used as an adaptive criterion to achieve dynamic closed-loop control of sampling depth under varying turbidity conditions. Specifically, the convergence criterion is:

[0133] (6)

[0134] in, This represents the evaluation index of sampling certainty at the previous moment; This represents the change in uncertainty between two adjacent samples; The relative rate of change (or the slope of decline) of the evaluation indicator. The preset convergence threshold is typically set to 0.05 with standard precision. In special applications requiring extremely high video frame rates, the threshold can be adjusted based on water quality. When the rate of decline of the indicator falls below this value, the sampling uncertainty evaluation function is considered to have converged, immediately triggering a system interrupt, stopping sampling, and outputting the decoupling results. This mechanism ensures optimal signal-to-noise separation with minimal time overhead in turbidity scenarios. Finally, the optimal parameter vector for the current moment is output. At the physical level, this result achieves mathematical decoupling of image information, that is, it clearly separates the background scattering component representing depolarization noise. (Noise component), and modulation coefficient component representing the target polarization response intensity. and (Target signal), thereby achieving a clearer reconstruction of the image.

[0135] If the sampling uncertainty evaluation function converges, the current model parameter estimate will be used as the optimal model parameter for the modulation intensity model.

[0136] If the sampling uncertainty evaluation function does not converge, the polarization image sequence is updated by acquiring new polarization observation images of the target polarization under new polarization illumination light, and then the model parameter estimates are obtained and evaluated again until the sampling uncertainty evaluation function converges.

[0137] In this embodiment, an adaptive optimization truncation criterion based on the covariance trace is used, that is, a sampling uncertainty evaluation function is constructed through the trace of the covariance matrix to quantify the overall uncertainty of the current parameter estimation. When the evaluation function converges, sampling is terminated; when the evaluation function does not converge, new polarization state images are collected and parameter estimates are updated until convergence. It should be noted that turbidity is dynamically changing in real-world environments. When the environment is relatively clear, a small number of samples are sufficient for signal separation; however, in high-scattering environments, more samples are required. Using a fixed number of samples can lead to two problems: slow system response due to excessive sampling; and poor separation accuracy due to insufficient sampling. Therefore, the core idea of ​​this invention is to determine whether to continue sampling by evaluating the uncertainty of the current estimation result in real time. After each sampling, a set of parameter estimates is obtained. At the same time, the corresponding covariance matrix can be obtained by least squares estimation, which reflects the uncertainty of the current estimation result. In order to transform the uncertainty of multidimensional parameters into a single evaluation index, this invention uses the trace of the covariance matrix as a measure of overall uncertainty. This index is equivalent to the sum of the variances of each parameter and can characterize the current overall estimation error level of the system.

[0138] In this embodiment, a polarization state optimization criterion is employed to select the polarization modulation state that is most beneficial for reducing uncertainty within the candidate range as the next polarization state. Specifically, the polarization state optimization criterion is based on maximizing information gain or minimizing equivalent uncertainty, used to improve the contribution of each sample to the accuracy of parameter estimation. If the liquid crystal angles of each sample are very different (high correlation), even if there are many equations constructed, their coefficient matrix will be approximately singular (ill-conditioned), and an accurate solution cannot be obtained. To maximize the information gain provided by the new polarization sampling, within the modulation range supported by the liquid crystal polarization modulator, the polarization modulation state with the lowest linear correlation (i.e., the strongest orthogonality) with the historical basis vector set is selected as the next polarization state. This selection criterion ensures that each newly added target polarization observation image can provide independent, non-redundant mathematical constraints for the decoupling of multidimensional parameters (noise components and target signal components).

[0139] Based on the optimal model parameters, an enhanced reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarization image sequence acquired at the current moment, thereby obtaining the enhanced target image and forming a dynamically enhanced video stream.

[0140] In this embodiment, based on the optimized model parameters, the model is reconstructed through spatiotemporal joint constraints to eliminate severe noise interference. Specifically, the optimal parameter vector output after satisfying the statistical convergence criterion is used. The background scattering component in the optimal parameter vector is decoupled from the target signal modulation coefficient with high precision. Finally, the optimized depolarized background noise component obtained by decoupling is imported into the background elimination and spatiotemporal reconstruction model to remove severe noise interference, thereby achieving a clear reconstruction of the image and obtaining the final dynamically enhanced video stream.

[0141] The specific method for reconstructing the polarization image sequence acquired at the current moment using an enhancement reconstruction model based on spatiotemporal joint constraints according to the optimal model parameters, to obtain the enhanced target image and form a dynamically enhanced video stream, is as follows:

[0142] Obtain the maximum intensity value corresponding to each pixel position in the polarization image sequence acquired at the current moment;

[0143] Based on the maximum intensity value corresponding to each pixel position and the depolarized background noise component in the optimal model parameters, an enhancement reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarized image sequence acquired at the current moment, thereby obtaining the enhanced target image.

[0144] In video scenarios, disturbances can cause noise components to fluctuate wildly between adjacent frames. Optimizing only a single frame can lead to temporal instability in image brightness and contrast. Therefore, after obtaining the target signal and noise components, a background removal and spatiotemporal reconstruction model is used to reconstruct the polarization image sequence acquired at the current moment. This model utilizes a subtraction criterion to enhance the signal. Furthermore, under different polarization modulation states, the target reflected light is modulated (manifesting as flickering brightness), while the background scattered light remains relatively stable. Therefore, the maximum intensity value corresponding to each pixel position is selected as the benchmark, which typically preserves the energy of the target signal to the greatest extent.

[0145] The enhancement and reconstruction model based on spatiotemporal joint constraints is as follows: the enhanced intensity value of a pixel is represented as the difference between the maximum intensity value of a pixel in the polarization image sequence acquired at the current moment and the depolarization background noise component adjusted by the background suppression parameter, expressed as:

[0146] (7)

[0147] in, For pixels Enhanced strength value; For the pixels in the polarization image sequence acquired at the current moment The maximum intensity value is used to retain the target energy reference; For the pixel points in the optimal model parameters The corresponding depolarized background noise component represents the noise. The background suppression parameter is used to dynamically adjust the background suppression parameter, which can achieve adaptive optimization suppression of scattering media with different concentrations, thereby obtaining a target image with enhanced contrast.

[0148] The enhanced target images are output in chronological order to form a dynamically enhanced video stream.

[0149] For dynamic video streams, a temporal coherence penalty mechanism is introduced to determine background suppression parameters in order to suppress screen flicker caused by instantaneous scattering abrupt changes.

[0150] Specifically, the background suppression parameter is calculated as follows:

[0151] For the polarization image sequence acquired at the current moment, let the time index of the target polarization observation image in the current frame be . The optimization model for the background suppression parameters of the current frame is then defined as:

[0152] (8)

[0153] in, This is the background suppression parameter for the current frame, used to control the intensity of scattering cancellation; and These are the information entropy and average gradient of the enhanced target image, respectively, used to measure image detail and contrast; The weighting coefficient for information entropy; These are the weighting coefficients for the average gradient; and Used to balance the proportions of information entropy and gradient in the optimization process; This indicates a normalization operation.

[0154] In this embodiment, information entropy This represents the overall information richness of an image. If the image is blurry and indistinct, it's necessary to prioritize increasing the overall contrast and information content. In this case, it's necessary to increase... Average gradient This represents the edge sharpness (clarity) of a local area of ​​the image. If the image is relatively clear or the subsequent machine vision task is object recognition / edge extraction, which requires very sharp image edges, then it needs to be increased. In actual engineering implementation, the weighting coefficient and All parameters are normalized and satisfy the constraints. To adapt to different underwater or scattering detection missions, both steady-state and adaptive weight allocation strategies can be provided.

[0155] This is a temporal smoothing term (penalty mechanism) that limits the background suppression parameters between adjacent frames. The sudden change in value eliminates video flicker; The temporal coherence constraint weight determines the degree of enforcement of image brightness smoothing. In mathematics, it is a regularization parameter (penalty coefficient) used to constrain the background suppression parameters between two consecutive frames. The range of change; In order to be in Temporary value for the current frame background suppression parameter during the search within the interval.

[0156] In this embodiment, for scenarios with high environmental disturbance frequencies (such as turbulent waters or dusty environments) and where a smooth visual experience is required, the [unclear text - possibly a specific component or element] needs to be increased. To enhance temporal coherence penalty; in scenarios where environmental parameters change relatively slowly or where the real-time enhancement accuracy of a single frame is extremely high, the penalty can be reduced. This is to improve the tracking performance of the gain factor. Specifically, in real-world environments (such as water flow disturbances or the instantaneous drift of suspended objects), if only the sharpest single frame is desired, the sharpness of adjacent frames is less important. The value may change drastically; for example, the background suppression parameter in frame 1 may be calculated as follows: The background suppression parameter in the second frame is calculated as follows due to interference from a bubble: This manifests in the video as flickering, with the image appearing bright and dark. Therefore, a time smoothing term is needed to penalize the large fluctuations between the background suppression parameters of the current frame and the previous frame.

[0157] In this embodiment, based on the spatiotemporal joint constraint-based enhanced reconstruction model described in formula (7), global optimization of the background suppression parameters is achieved by comprehensively evaluating the spatial detail information gain and the coherence loss in the temporal dimension. This model not only ensures the maximization of contrast and information content in a single frame image, but also effectively suppresses instantaneous interference caused by dynamic environments through the temporal smoothing term, fundamentally solving the screen flicker problem of traditional polarization enhancement algorithms in video applications, and providing mathematical guarantees for the real-time engineering application of the system.

[0158] Example 2:

[0159] This embodiment presents a dynamic image enhancement system based on liquid crystal polarization adaptive modulation, such as... Figure 3 As shown, the system includes:

[0160] The polarization modulation module is used to generate illumination light with different polarization states, and when it receives a polarization modulation drive signal, it generates illumination light with a new polarization state according to the polarization state optimization criterion.

[0161] In this embodiment, the polarization modulation module can be implemented using an electronically controlled liquid crystal polarization modulator (LCVR). By adjusting the driving voltage applied across the liquid crystal device, its phase delay or polarization angle is changed, thereby modulating the polarization state of the incident light. Specifically, this embodiment can use a high-speed LCVR liquid crystal polarization modulator, with a typical response time of 1–10 ms; in high-speed cases, a high-speed FLC liquid crystal can be used, with a response time of 10–100 μs. The polarization modulation module can have multiple discrete modulation states or achieve continuous polarization modulation under continuous voltage. It meets the following basic operating requirements: controllable polarization modulation capability; response time meeting dynamic closed-loop control requirements; and support for at least two polarization modulation states.

[0162] The imaging acquisition module is used to acquire target polarization observation images under illumination light with different polarization states and construct a polarization image sequence; and when an image acquisition trigger signal is received, it acquires target polarization observation images under illumination light with new polarization states and updates the polarization image sequence.

[0163] In this embodiment, the imaging acquisition module is implemented using a CMOS image sensor. After receiving a trigger signal from the control module, it performs exposure and readout operations to acquire image data under the corresponding polarization modulation state. The imaging acquisition module satisfies the following cooperative relationship: the acquisition time is triggered and controlled by the control module; the exposure process corresponds to the polarization modulation state; and the output image data is used by the image processing module.

[0164] The frame rate of the imaging acquisition module is preferably tens to hundreds of frames per second. In this embodiment, the frame rate should be no less than 1 times the polarization modulation frequency, and more preferably more than 2 times the polarization modulation frequency; the exposure time is less than or equal to the stabilization time of the polarization modulation module. The spatial resolution is selected according to application requirements, preferably at the megapixel level or above.

[0165] Specifically, the imaging acquisition module may employ, but is not limited to, the following models: the image sensor of the Daheng-Mercury series MER-132-43GM / C industrial camera; frame rate 43fps, exposure time 6μs~1s, resolution 1292(H)×964(V); the Daheng-Mercury series MER-031-860U3M / C high-speed camera, frame rate 860 fps, exposure time 20μs~1s, resolution 640(H) × 480(V), used for high-speed applications; and the Daheng-Mercury series polarization camera MER2-502-79U3M POL, equipped with 4 angle polarization photosensitive chips, frame rate 79.1fps, exposure time 20μs~1s, resolution 2448(H) × 2048(V), used for harsh environments.

[0166] The model estimation module is used to input the polarization image sequence acquired at the current time into the mathematical model between the polarization observation image and the target signal and noise components, and to use the parameter estimation method to fit the model parameters of the modulation intensity model in the mathematical model to obtain the estimated values ​​of the model parameters.

[0167] An adaptive sampling control module is used to substitute the estimated model parameters into the mathematical model and evaluate it using a sampling uncertainty evaluation function: if the function converges, the current estimated model parameters are used as the optimal model parameters for the modulation intensity model; if the function does not converge, a control feedback signal is generated and output to the control module.

[0168] The image reconstruction module is used to reconstruct the polarization image sequence acquired at the current moment using an enhancement reconstruction model based on spatiotemporal joint constraints, according to the optimal model parameters, to obtain the enhanced target image and form a dynamically enhanced video stream.

[0169] The model estimation module, adaptive sampling control module, and image reconstruction module are all integrated on the same processor. In this embodiment, the processor can be implemented using the following devices: a general-purpose central processing unit (CPU), such as the Core i5 or Ryzen 5 series; an embedded processor platform, such as the Jetson Nano series or RK3588 series; or a graphics processing unit (GPU) for parallel computing, such as the RTX 3060 series.

[0170] The control module is used to generate a polarization modulation drive signal and an image acquisition trigger signal respectively based on the received control feedback signal and a preset delay parameter, and send the polarization modulation drive signal to the polarization modulation module and the image acquisition trigger signal to the imaging acquisition module.

[0171] In this embodiment, the control module coordinates the timing of operation between the polarization modulation module and the imaging acquisition module, and adjusts the polarization modulation state according to the control feedback signal. It has at least the following functions: triggering image acquisition after the polarization modulation state stabilizes; adjusting modulation parameters or modulation sequence according to the feedback signal; and controlling the progress or termination of the acquisition process.

[0172] Specifically, the control module can be implemented using, but is not limited to, programmable logic devices. Its timing control accuracy is preferably in the nanosecond to microsecond range; the output trigger signal jitter is preferably less than 1 μs; it supports multiple synchronous outputs for separately controlling the polarization modulation module and the imaging acquisition module; and it can dynamically adjust the output timing or modulation sequence based on feedback signals. The programmable logic device includes a field-programmable gate array (FPGA) or a system-on-a-chip (SoC FPGA). The FPGA can be selected from one of the following models: Spartan-6 series (e.g., XC6SLX9); Artix-7 series (e.g., XC7A35T); Cyclone IV series (e.g., EP4CE10); or Cyclone V series devices. The SoC device can be a Zynq-7000 series device, etc.

[0173] The control module also includes a synchronization and delay control unit for coordinating the timing of the polarization modulation module and the imaging acquisition module. Specifically, after outputting the polarization modulation drive signal, the control module introduces delay control based on the response characteristics of the polarization modulation module to compensate for its response time, ensuring the polarization state reaches a stable state before triggering the imaging acquisition module to perform image acquisition. The delay control can be implemented based on preset parameters or dynamically adjusted according to the control feedback signal output by the image processing module, thereby achieving adaptive optimization compensation for the polarization modulation stabilization time. Through delay control, the imaging acquisition time corresponds one-to-one with the current polarization modulation state, ensuring timing consistency between images acquired under different polarization states.

[0174] During the closed-loop control process, the control feedback signal output by the image processing module is used not only to adjust the polarization modulation parameters, but also to update the delay parameters of the delay control unit, so that the synchronization mechanism is dynamically adjusted according to the system state, thereby realizing closed-loop synchronous control between polarization modulation and image acquisition.

[0175] It should be noted that the above device models are merely examples, and those skilled in the art can choose other devices with the same or similar functions to replace them as needed.

[0176] In this embodiment, the control module and the image processing module can be integrated into the same processing unit, with the same hardware platform performing image processing and control signal output functions. The control module can be implemented based on a processor, programmable logic device, or system-on-a-chip, completing image processing operations and control timing generation through internal resource scheduling. Alternatively, the control module and the image processing module can be implemented separately, connected via a communication interface, including but not limited to serial interfaces, parallel interfaces, or high-speed data interfaces, to transmit image data and control feedback signals. The control module and the image processing module can be implemented using an integrated package or a modular combination, depending on system requirements. Integrated packaging reduces system size and latency, while modular combination improves system flexibility and scalability.

[0177] The goal of the dynamic image enhancement system proposed in this embodiment is to accurately separate the target signal from the noise component with as few samplings as possible. To achieve this goal, this invention constructs a closed-loop process: first, modeling; then, estimating; then, determining whether it is "good enough," and finally deciding whether to continue sampling. Its core lies in achieving high-precision separation of the signal and noise components by establishing physical constraints for liquid crystal modulation, dynamic statistical convergence criteria, and spatiotemporal continuity optimization.

[0178] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0179] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0180] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this disclosure and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A dynamic image enhancement method based on liquid crystal polarization adaptive modulation, characterized in that, This method includes the following steps: Collect target polarization observation images under illumination with different polarization states, and construct a polarization image sequence; A mathematical model is established between the polarization observation image and the target signal and noise components, including: a liquid crystal polarization modulation model and a modulation intensity model; The polarization image sequence acquired at the current moment is input into the mathematical model between the polarization observation image and the target signal and noise components, and the model parameters of the modulation intensity model are fitted by the parameter estimation method to obtain the estimated values ​​of the model parameters. The obtained model parameter estimates are substituted into the mathematical model between the polarization observation image and the target signal and noise components, and the global error of the current model parameter estimates is evaluated using the sampling uncertainty evaluation function. If the sampling uncertainty evaluation function converges, the current model parameter estimate will be used as the optimal model parameter for the modulation intensity model. If the sampling uncertainty evaluation function does not converge, the polarization image sequence is updated by acquiring new polarization observation images of the target under the illumination light of the new polarization state, and then the model parameter estimates are obtained again and evaluated until the sampling uncertainty evaluation function converges. Based on the optimal model parameters, an enhanced reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarization image sequence acquired at the current moment, thereby obtaining the enhanced target image and forming a dynamically enhanced video stream.

2. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 1, characterized in that, The specific method for acquiring target polarization observation images under illumination with different polarization states and constructing a polarization image sequence is as follows: By applying different driving voltages to the liquid crystal polarization modulator, illumination light with different polarization states is generated; A polarizer with a known polarization angle is set at the image sampling receiver, and target polarization observation images under illumination light of each polarization state are acquired accordingly; the target polarization observation images contain light intensity distribution information. A polarization image sequence is formed by collecting target polarization observation images under illumination of all polarization states.

3. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 2, characterized in that, The specific method for establishing the mathematical model between the polarization observation image and the target signal and noise components is as follows: A liquid crystal polarization modulation model is established; the liquid crystal polarization modulation model characterizes the actual polarization modulation amount under the combined effect of the current driving voltage and the current ambient temperature as the superposition of the ideal voltage-polarization modulation amount mapping function of the liquid crystal polarization modulator and the nonlinear compensation term; the nonlinear compensation term is used to correct the influence of ambient temperature and the response hysteresis error of liquid crystal molecules. Based on the liquid crystal polarization modulation model, a modulation intensity model is established under the condition that the polarization angle of the analyzer is known; The modulation intensity model characterizes the image intensity of each polarization sample as the sum of the depolarized background noise component, the polarized target signal component containing the target signal modulation coefficient, and the random additive noise during the photoelectric detection process.

4. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 3, characterized in that, The parameters to be solved in the modulation intensity model include: the depolarized background noise component and the target signal modulation coefficient.

5. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 4, characterized in that, The method involves substituting the obtained model parameter estimates into the mathematical model between the polarization observation image and the target signal and noise components, and then evaluating the model using a sampling uncertainty evaluation function. Substitute the preliminary estimates of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components, and calculate the least squares estimated covariance matrix. The sampling uncertainty evaluation function is constructed by estimating the trace of the covariance matrix using least squares, and the sampling certainty evaluation index at the current time is calculated. If the sampling certainty evaluation index at the current moment and the sampling certainty evaluation index at the previous moment satisfy the convergence criterion, then the sampling uncertainty evaluation function is considered to be convergent; otherwise, the sampling uncertainty evaluation function is considered to be non-convergent.

6. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 5, characterized in that, The specific method for substituting the preliminary estimated values ​​of the model parameters into the mathematical model between the polarization observation image and the target signal and noise components to calculate the least squares estimated covariance matrix is ​​as follows: Using the preliminary estimates of the model parameters obtained, the vector of parameters to be determined is constructed in real time. Based on the actual polarization modulation amount of each polarization sample, construct the polarization physical basis for polarization sampling. Based on the parameter vector to be determined, the instantaneous noise variance is evaluated in real time using the forward fitting residual. Based on the number of image frames in the polarization image sequence acquired at the current moment, the least squares estimation covariance matrix is ​​constructed in real time using the polarization physical basis and the instantaneous noise variance.

7. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 6, characterized in that, The specific method for real-time evaluation of instantaneous noise variance based on the forward fitting residual, using the parameter vector to be determined, is as follows: Obtain the parameter vector to be determined, constructed based on the polarization image sequence acquired in the previous moment; Substitute the parameter vector to be determined into the mathematical model between the polarization observation image and the target signal and noise components, and combine it with the actual polarization modulation amount of the current polarization sampling to calculate the theoretical pixel intensity value under the current modulation state. Calculate the squared residual between the actual image intensity and the theoretical pixel intensity value of the current polarization sample pixel by pixel, and then calculate the average squared residual of all pixels in the target polarization observation image of the current polarization sample as the instantaneous noise variance.

8. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 7, characterized in that, The specific method for reconstructing the polarization image sequence acquired at the current moment using an enhanced reconstruction model based on spatiotemporal joint constraints according to the optimal model parameters to obtain the enhanced target image is as follows: Obtain the maximum intensity value corresponding to each pixel position in the polarization image sequence acquired at the current moment; Based on the maximum intensity value corresponding to each pixel position and the depolarized background noise component in the optimal model parameters, an enhancement reconstruction model based on spatiotemporal joint constraints is used to reconstruct the polarized image sequence acquired at the current moment to obtain the enhanced target image. The enhanced reconstruction model based on spatiotemporal joint constraints is as follows: the enhanced intensity value of a pixel is represented as the difference between the maximum intensity value of a pixel in the polarization image sequence acquired at the current moment and the depolarization background noise component after background suppression parameter adjustment. The enhanced target images are output in chronological order to form a dynamically enhanced video stream.

9. The dynamic image enhancement method based on liquid crystal polarization adaptive modulation according to claim 8, characterized in that, The background suppression parameters are determined using a temporal coherence penalty mechanism; For the polarization image sequence acquired at the current moment, the optimization model for the background suppression parameters of the current frame is defined as: Within a preset search interval, find the current frame background suppression parameter value that optimizes the target evaluation function; The objective evaluation function is composed of a spatial information gain term and a temporal coherence smoothing term; Among them, the spatial information gain term is obtained by weighted normalization of the information entropy and average gradient of the enhanced target image; The temporal coherence smoothing term is determined by the background suppression parameters of the current frame and the previous frame, combined with the temporal coherence constraint weights, and is used to limit parameter abrupt changes between adjacent frames.

10. A dynamic image enhancement system based on liquid crystal polarization adaptive modulation, used to implement the dynamic image enhancement method based on liquid crystal polarization adaptive modulation as described in any one of claims 1-9, characterized in that, The system includes: The polarization modulation module is used to generate illumination light with different polarization states, and when it receives the polarization modulation drive signal, it generates illumination light with a new polarization state according to the polarization state optimization criterion. The imaging acquisition module is used to acquire target polarization observation images under illumination light with different polarization states and construct a polarization image sequence; and when an image acquisition trigger signal is received, it acquires target polarization observation images under illumination light with new polarization states and updates the polarization image sequence. The model estimation module is used to input the polarization image sequence acquired at the current time into the mathematical model between the polarization observation image and the target signal and noise components, and to use the parameter estimation method to fit the model parameters of the modulation intensity model in the mathematical model to obtain the estimated values ​​of the model parameters; An adaptive sampling control module is used to substitute the estimated model parameters into the mathematical model and evaluate it using a sampling uncertainty evaluation function: if the function converges, the current estimated model parameters are used as the optimal model parameters for the modulation intensity model; if the function does not converge, a control feedback signal is generated and output to the control module. The control module is used to generate a polarization modulation drive signal and an image acquisition trigger signal respectively based on the received control feedback signal and a preset delay parameter, and send the polarization modulation drive signal to the polarization modulation module and the image acquisition trigger signal to the imaging acquisition module. The image reconstruction module is used to reconstruct the polarization image sequence acquired at the current moment using an enhancement reconstruction model based on spatiotemporal joint constraints, according to the optimal model parameters, to obtain the enhanced target image and form a dynamically enhanced video stream.

Images