A multi-scattering polarization feature extraction and equivalent imaging method

Abstract

The application discloses a kind of multiple scattering polarization feature extraction and equivalent imaging method, comprising: obtaining multiple linear polarization direction intensity images of multiple preset incident positions in object plane;Stokes vector of each incident position is calculated and normalized partial stokes parameter, then filtered by annular band-pass filter, so that the change rate of mutual energy of filtered normalized parameter with the increase of scattering layer is lower than the preset threshold;The covariance of the filtered parameter is calculated in local window, and the covariance map of each incident position is obtained;The large field of view covariance map is constructed from the covariance map of all incident positions, and it is used as physical constraint information to reconstruct scattering image by scattering imaging model.The application extracts polarization covariance feature which is not sensitive to scattering times, breaks through the field of view limit of memory effect, realizes large field of view stable imaging, and provides physical prior to guide scattering imaging model, significantly improves the image reconstruction quality and robustness in complex scattering scene.

Description

Technical Field

[0001] This invention belongs to the field of scattering imaging, specifically relating to a method for extracting polarization features of multiple scattering and equivalent imaging. Background Technology

[0002] Imaging through scattering media such as smoke, turbid water, and biological tissue has significant practical applications in remote sensing, public safety, and biomedicine. Under weak scattering conditions, photons undergo only a single scattering event with the target surface and propagate in a ballistic manner. In this case, the light field carrying target information maintains a definite phase and polarization structure, allowing the imaging system to efficiently reconstruct the target's morphology. However, when the target is hidden behind a strong scattering medium, photons undergo multiple random scattering events. Their propagation direction, amplitude, phase, and polarization state all undergo complex evolutions that are difficult to predict. The light field information received by the detector is an incoherent superposition of a large number of photons with different paths and scattering times, and the target information is deeply submerged in the scattering background and noise.

[0003] Current scattering imaging methods can reconstruct target information using scattered photons, but they are ineffective in complex scattering scenarios such as real-world smoke and biological tissue. This is because current scattering transmission models cannot accurately describe the non-uniform scattering transmission process, and the multiple scattering effects of the scattering medium make it difficult to extract and reconstruct information from weak targets. Existing scattering imaging techniques have severely limited reconstruction capabilities when passing through non-uniform, super-optical memory effects, and complex scattering media, and traditional single-scattering theory is difficult to apply to complex multiple-scattering imaging.

[0004] Therefore, there is an urgent need for a scattering imaging method that can be applied to complex scattering scenarios, realize the mapping and equivalence of multiple scattering to single scattering, and provide physical priors for subsequent neural network learning. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for multiple scattering polarization feature extraction and equivalent imaging. In a first aspect, the present invention provides a method for multiple scattering polarization feature extraction and equivalent imaging, comprising: Acquire a set of polarization images of multiple preset incident positions in the object plane, the set of polarization images including intensity images of the scattering scene under multiple linear polarization directions after multiple scatterings; For each incident position, a Stokes vector is calculated based on the polarization image set corresponding to the incident position, and multiple Stokes parameters of the Stokes vector are normalized to obtain the corresponding first normalization parameter and second normalization parameter. Based on a preset ring bandpass filter, the first normalized parameter and the second normalized parameter corresponding to the incident position are bandpass filtered. The rate of change of the mutual energy between the filtered first normalized parameter and the second normalized parameter with the increase of the number of scattering layers is lower than a preset threshold. Within a preset local window, the covariance between the filtered first normalized parameter and the second normalized parameter corresponding to the incident position is calculated to obtain the corresponding covariance map. A large field-of-view covariance map is constructed based on the covariance maps corresponding to all incident positions. The large field-of-view covariance map is used as physical constraint information, and the scattering scene is reconstructed using a preset scattering imaging model to obtain a scattering image.

[0006] In one embodiment of the present invention, the polarization image set includes intensity images of the scattering scene in four linear polarization directions: 0°, 45°, 90° and 135°.

[0007] In one embodiment of the present invention, the Stokes parameters of the Stokes vector include a first light intensity, a second light intensity, and a third light intensity. The first light intensity is the result of adding the first intensity image and the second intensity image pixel by pixel. The second light intensity is the result of subtracting the first intensity image and the second intensity image pixel by pixel. The third light intensity is the result of subtracting the third intensity image and the fourth intensity image pixel by pixel. The first intensity image, the second intensity image, the third intensity image, and the fourth intensity image are the intensity images of the scattering scene in four linear polarization directions of 0°, 90°, 45°, and 135°, respectively. The normalization of the multiple Stokes parameters of the Stokes vector includes: Calculate the ratio of the second light intensity to the first light intensity to obtain the first normalized parameter; The ratio of the third light intensity to the first light intensity is calculated to obtain the second normalized parameter.

[0008] In one embodiment of the present invention, constructing the ring bandpass filter includes: At a preset first incident position, multiple first polarization image sets are acquired. Each first polarization image set includes an intensity image of a preset scattering scene in multiple linear polarization directions. The scattering medium structure of the preset scattering scene includes a single-layer scattering medium, a double-layer closely attached scattering medium, and a sandwich structure of two scattering media separated by air. For the first polarization image set corresponding to the preset scattering scenario for each scattering medium structure, calculate the first normalization parameter and the second normalization parameter corresponding to the first polarization image set, and calculate the cross power spectrum of the first normalization parameter and the second normalization parameter to obtain the cross power spectrum corresponding to the first polarization image set. The frequency range in which the rate of change of energy with increasing number of scattering layers is lower than the first preset threshold is defined as the stable frequency range in each cross power spectrum corresponding to different scattering medium structures. The passband of the ring bandpass filter is determined based on the lower and upper limits of the stable frequency range, and the ring bandpass filter is constructed.

[0009] In one embodiment of the present invention, the scale of the local window is determined based on the speckle coherence scale generated by the scattering scene.

[0010] In one embodiment of the present invention, the step of using the large field-of-view covariance map as physical constraint information and reconstructing the scattering scene using a preset scattering imaging model includes: The large field-of-view covariance map is used as a spatial weight mask and embedded into the loss function of the scattering imaging model to weight the error between the scattering image output by the scattering imaging model and the preset target reference image. In the loss function, when weighting the reconstruction error of each pixel of the scattering image, if the covariance value of the pixel in the large field-of-view covariance map is higher, the weight of the pixel is greater.

[0011] In one embodiment of the present invention, after constructing a large field-of-view covariance map based on the covariance maps corresponding to all incident positions, the method further includes: The large field-of-view covariance map and the polarization feature map calculated from the polarization image set are stitched together in the channel dimension, and the resulting multi-channel feature map is used as the input to the scattering imaging model.

[0012] Secondly, the present invention also provides a device for multiple scattering polarization feature extraction and equivalent imaging, comprising: The polarization information acquisition module is used to acquire a set of polarization images of multiple preset incident positions in the object plane. The set of polarization images includes intensity images of the scattering scene after multiple scatterings in multiple linear polarization directions. The normalization module is used to calculate the Stokes vector based on the polarization image set corresponding to each incident position, and normalize multiple Stokes parameters of the Stokes vector to obtain the corresponding first normalization parameter and second normalization parameter. The bandpass filtering module is used to perform bandpass filtering on the first normalized parameter and the second normalized parameter corresponding to the incident position based on a preset ring bandpass filter. The rate of change of the mutual energy between the first normalized parameter and the second normalized parameter after filtering with the increase of the number of scattering layers is lower than a preset threshold. The covariance plot calculation module is used to calculate the covariance between the first normalized parameter and the second normalized parameter after filtering at the incident position within a preset local window, and obtain the corresponding covariance plot. The imaging module is used to construct a large field-of-view covariance map based on the covariance map corresponding to all incident positions, and use the large field-of-view covariance map as physical constraint information to reconstruct the scattering scene through a preset scattering imaging model to obtain a scattering image.

[0013] Thirdly, the present invention also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of any of the above-described multiple scattering polarization feature extraction and equivalent imaging methods.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described multiple scattering polarization feature extraction and equivalent imaging methods.

[0015] The beneficial effects of this invention are: This invention acquires polarization image sets at multiple preset incident positions within the object plane, and sequentially performs Stokes parameter normalization, annular bandpass filtering, and local window covariance calculation at each incident position to obtain a covariance map for each incident position, thereby constructing a large field-of-view covariance map covering the target's field of view. The rate of change of the cross-energy between the filtered, normalized Stokes parameters with increasing scattering layer number is lower than a preset threshold, making the extracted covariance map essentially insensitive to changes in the number of scattering events. Therefore, it can stably reflect the intrinsic polarization properties of the medium surface even under multiple scattering conditions, achieving an equivalent mapping from multiple scattering to single scattering. This solves the problem of existing technologies struggling to extract weak target information due to inaccurate scattering transmission models. Meanwhile, by acquiring covariance values ​​point by point within a large field of view and stitching them together, the method no longer relies on the translation invariance required by the optical memory effect, effectively overcoming the rigid limitation of the memory effect on the imaging field of view. This allows the method to be applied to non-uniform and complex scattering scenarios beyond the memory effect range (such as smoke, turbid water, and biological tissue). Furthermore, the obtained large field of view covariance map is directly used as physical constraint information input into a preset scattering imaging model for image reconstruction. This covariance map provides spatial priors independent of the number of scattering events, guiding the imaging model to prioritize the recovery of the target structure in physically reliable areas. This significantly improves the reconstruction quality and stability under strong scattering environments, overcoming the shortcomings of traditional methods that are severely limited in reconstruction capabilities in complex scattering media.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for extracting polarization features and achieving equivalent imaging through multiple scattering provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a multiple scattering polarization feature extraction and equivalent imaging device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware entity of an electronic device to which embodiments of the present invention are applied. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0019] Example 1 It should be noted that while various studies have explored target information recovery using multiple scattering, they all have significant limitations when dealing with real-world, complex scattering environments. For example, wavefront shaping techniques compensate the incident wavefront using spatial light modulators to enhance light focusing within or behind the scattering medium. However, this method relies on point-by-point iterative optimization and static scattering assumptions. When faced with time-varying scattering environments such as living biological tissue, flowing turbid water, or atmospheric turbulence, it cannot meet real-time control requirements, leading to a sharp degradation in focusing and imaging performance. Speckle correlation imaging utilizes the translation invariance of the point spread function within the optical memory effect range to calculate the autocorrelation or Fourier amplitude of the target from the speckle image. However, the spatial span of the memory effect is drastically compressed in multiple scattering media, limiting this method to simple targets within a very small field of view and failing to meet the practical imaging requirements of a wide spectral band and large field of view. Scattering imaging methods based on optical transfer matrices establish a deterministic mapping between input and output by pre-calibrating the full-field transfer matrix of the medium. This process requires not only extremely time-consuming multi-channel interferometry and very high system stability, but also the transfer matrix itself is highly sensitive to external disturbances such as ambient temperature and structural deformation. Any slight drift in the medium state can cause the previously calibrated matrix to fail, making it difficult to operate reliably in uncontrolled environments. Furthermore, while coherence gates, polarization gates, and time gates based on ballistic or serpentine photon screening can effectively suppress diffuse background in thin scattering media, the ratio of ballistic to serpentine light decreases exponentially with increasing scattering frequency, and the effective detection depth of the detector is rigidly limited by physical limits.

[0020] In summary, current classical speckle imaging techniques all attempt to extract ballistic light information from the scattered light field while ignoring or stripping away multiple scattered light. They fail to recognize the structural polarization information carried by the multiple scattered photons themselves, which can be recovered through statistical laws. Their reconstruction capability is severely limited when passing through non-uniform, super-optical memory effects, or complex scattering media.

[0021] Therefore, in order to solve the above-mentioned technical problems, this embodiment breaks away from the conventional paradigm and does not treat multiple scattering as a pure noise source to be suppressed or gated. Instead, it analyzes the statistical stability of polarization information during the depolarization process of multiple scattering from a physical perspective, extracts stable polarization features that are not sensitive to changes in the number of scatterings, and thus maps multiple scattering to single scattering, providing a physical prior that does not require manual annotation for intelligent reconstruction methods such as neural networks.

[0022] like Figure 1 As shown, this embodiment provides a method for multiple scattering polarization feature extraction and equivalent imaging, including: S1. Obtain a set of polarization images of multiple preset incident positions in the object plane, wherein the set of polarization images includes intensity images of the scattering scene under multiple linear polarization directions after multiple scatterings. Understandably, for single-layer or weakly scattering media, the optical memory effect allows the point spread function of the imaging system to approximately satisfy translation invariance within a local range. Therefore, image reconstruction can be achieved using single-point calibration or local convolution kernels. However, when the medium extends to multiple layers, inter-layer coupling is significantly enhanced, the number of scattering paths increases dramatically, the system response no longer satisfies spatial translation invariance, and the statistical distribution of the response changes significantly between different spatial locations, different scattering layers, and different polarization states. Under these conditions, attempting to decouple complex multiple scattering according to a single scattering model often leads to problems such as strong ill-conditioning and irreversibility, resulting in a lack of stability and generalization ability of the method. Therefore, this embodiment aims to extract a sub-region with polarization statistical stability from the chaotic light field of multiple scattering and physically map it to the single scattering field, thereby providing a low-level physical prior for subsequent neural networks that transcends the limitations of scattering order.

[0023] In a multiple scattering field, when a light wave penetrates a strongly scattering medium, the outgoing light field... From ballistic light Global scattered light A mixed composition can be represented as: ; Deeply scattered photons, having undergone numerous random walks, have their spatial phase and polarization state completely disrupted, exhibiting isotropic random speckle noise. However, some photons that have deviated from their original paths due to multiple scattering still retain polarization memory highly similar to that of single scattering, constrained by the anisotropy factor of the medium. Therefore, by effectively separating and extracting the single scattering component and the stable contribution term in multiple scattering from the scattered light field, the mapping and equivalence of multiple scattering effects to the single scattering process can be achieved at the physical characterization level.

[0024] Therefore, in this embodiment, intensity images of multiple linear polarization directions are first acquired at multiple preset incident positions within the object plane. This breaks the rigid limitation of the field of view imposed by the traditional memory effect, allowing the acquisition of raw data across the entire field of view through point-by-point scanning even if the medium exceeds the range of the memory effect. Secondly, Stokes parameter normalization and annular bandpass filtering are performed sequentially at each incident position, ensuring that the rate of change of the cross-energy of the normalized Stokes parameter with increasing scattering layer is below a preset threshold. Then, local window covariance calculation is performed to extract polarization statistical features insensitive to changes in the number of scattering events. These features directly originate from the intrinsic polarization joint fluctuations in the multiple scattering field, rather than relying on ballistic light or single scattering approximations. Therefore, they can exist stably under strong scattering conditions where the polarization degree approaches zero, physically equating complex multiple scattering to single scattering. Finally, a large field-of-view covariance map is constructed from the covariance maps of all incident positions and used as physical constraint information input into the scattering imaging model for reconstruction. The large field-of-view covariance map provides spatial priors independent of the number of scattering events, guiding the model to prioritize the recovery of the target structure in physically reliable regions, thereby significantly improving the reconstruction quality and stability under strong scattering environments. Thus, this scheme does not treat multiple scattering events as pure noise, but actively utilizes the stable polarization information they carry, overcoming the limitations of existing methods in reconstructing capabilities severely restricted in non-uniform, super-memory, and complex scattering media.

[0025] Furthermore, in step S1, the polarization image set includes intensity images of the scattering scene in four linear polarization directions: 0°, 45°, 90°, and 135°.

[0026] Understandably, when the propagation process extends to a multi-layered scattering medium, each layer can be considered as a cascade of a random polarization scattering operator and a free-space propagation operator. Thus, the total outgoing light field becomes a composite output of multiple random operators, and its statistical structure becomes complex, but not entirely disordered. The superposition of multiple random layers will form certain stable statistical regularities, which are often reflected in polarization correlation, normalized polarization, and the covariance structure of certain frequency bands. Therefore, this embodiment uses polarization as the core information dimension, directly participating in the physical degrees of freedom for stable information extraction.

[0027] Polarization, as a fundamental physical property of light, can provide additional information related to the surface characteristics and scattering mechanisms of a target. It exhibits unique advantages in scattering environments by suppressing scattering noise and enhancing information transmission. This information is used to facilitate the analysis and utilization of polarization information.

[0028] In this embodiment, polarization images are acquired at multiple preset incident positions within the object plane. At each incident position, an intensity image of the scattering scene at four linear polarization angles (0°, 45°, 90°, and 135°) is obtained from a fixed viewing angle using a polarizer placed in front of the detector. , , and This allows for the acquisition of a set of polarization images at each incident position. Subsequently, basic polarization parameters are calculated based on the data obtained from these measurements.

[0029] S2. For each incident position, calculate the Stokes vector based on the polarization image set corresponding to the incident position, and normalize multiple Stokes parameters of the Stokes vector to obtain the corresponding first normalization parameter and second normalization parameter. Furthermore, in step S2, the Stokes parameters of the Stokes vector include a first light intensity, a second light intensity, and a third light intensity. The first light intensity is the result of adding the first intensity image and the second intensity image pixel by pixel. The second light intensity is the result of subtracting the first intensity image and the second intensity image pixel by pixel. The third light intensity is the result of subtracting the third intensity image and the fourth intensity image pixel by pixel. The first intensity image, the second intensity image, the third intensity image, and the fourth intensity image are the intensity images of the scattering scene in four linear polarization directions: 0°, 90°, 45°, and 135°, respectively. The normalization of the multiple Stokes parameters of the Stokes vector includes: Calculate the ratio of the second light intensity to the first light intensity to obtain the first normalized parameter; The ratio of the third light intensity to the first light intensity is calculated to obtain the second normalized parameter.

[0030] In this embodiment, the polarization image set is used to calculate the basic polarization parameters, which can be represented by Stokes vectors, as follows: ; In Stokes vector middle, This represents the total light intensity, also known as the first light intensity, and is the first intensity image. With the second intensity image The result of adding pixels one by one; The second intensity is the result of pixel-by-pixel subtraction between the horizontal and vertical components, i.e., the first intensity image. With the second intensity image The result of pixel-by-pixel subtraction; The third intensity is the result of pixel-by-pixel subtraction between the 45° and 135° components, i.e., the third intensity image. With the fourth intensity image The result of pixel-by-pixel subtraction; It is a right-handed circularly polarized component. and left-handed circular polarization component The result is obtained by subtracting pixels one by one. Due to the limitations of current detectors, directly obtaining circular polarization information is not feasible. Therefore, the analysis in this embodiment mainly focuses on linear polarization.

[0031] The degree of polarization and the polarization angle, as important parameters characterizing polarization information, respectively measure the ratio of the linearly polarized component to the total light intensity and the orientation of the polarized photoelectric field vector vibration direction in space. To eliminate the absolute energy attenuation caused by beam divergence and absorption during subsequent propagation, a normalized Stokes parameter is defined, expressed by the formula: ; in, The first normalization parameter, For the second normalization parameter, , and These represent the first light intensity, the second light intensity, and the third light intensity, respectively.

[0032] It is understandable that by separately increasing the second light intensity and third light intensity Divide by the total light intensity, that is, the first light intensity This removes the absolute energy attenuation caused by beam divergence and medium absorption during propagation, resulting in the first normalized parameter. Second normalized parameter The polarization degree (DoP) no longer decreases overall with increasing scattering layers. This lays the foundation for extracting effective features even under multiple scattering conditions. Traditional DoP approaches zero after multiple scattering, while the normalized parameter... and The local structure still carries useful information.

[0033] S3. Based on a preset ring bandpass filter, the first normalized parameter and the second normalized parameter corresponding to the incident position are bandpass filtered. The rate of change of the mutual energy between the filtered first normalized parameter and the second normalized parameter with the increase of the number of scattering layers is lower than a preset threshold. Furthermore, constructing the ring bandpass filter includes: At a preset first incident position, multiple first polarization image sets are acquired. Each first polarization image set includes an intensity image of a preset scattering scene in multiple linear polarization directions. The scattering medium structure of the preset scattering scene includes a single-layer scattering medium, a double-layer closely attached scattering medium, and a sandwich structure of two scattering media separated by air. For the first polarization image set corresponding to the preset scattering scenario for each scattering medium structure, calculate the first normalization parameter and the second normalization parameter corresponding to the first polarization image set, and calculate the cross power spectrum of the first normalization parameter and the second normalization parameter to obtain the cross power spectrum corresponding to the first polarization image set. The frequency range in which the rate of change of energy with increasing number of scattering layers is lower than the first preset threshold is defined as the stable frequency range in each cross power spectrum corresponding to different scattering medium structures. The passband of the ring bandpass filter is determined based on the lower and upper limits of the stable frequency range, and the ring bandpass filter is constructed.

[0034] In this embodiment, the passband of the ring bandpass filter is determined in advance through experiments, as follows: At the same incident position, polarization image sets are acquired under three scattering conditions: single-layer, closely adjacent double-layer, and sandwich-spaced structures. These are known as the first polarization image sets. Following the normalization method in step S2, normalized polarization parameters corresponding to each first polarization image set are obtained, namely the first normalized parameter and the second normalized parameter. Cross-power spectrum estimation based on Fourier transform is then performed to calculate the first normalized parameter of the first polarization image set corresponding to each scattering medium structure. With the second normalized parameter Cross power spectrum The formula is as follows: ; in, For cross-power spectrum, This indicates the first normalization parameter Perform a Fourier transform. This indicates the second normalization parameter The conjugate of the Fourier transform result is calculated. The cross-power spectrum corresponding to each scattering medium structure is calculated, and the frequency response consistency of the cross-power spectrum under different scattering conditions is compared. The continuous frequency range where the cross-power spectrum fluctuates little with the number of scattering layers is defined as the stable frequency band; the lower limit frequency corresponding to this frequency range is defined as the low-frequency threshold. The upper limit frequency is defined as the high-frequency threshold. This is used to construct a stable frequency band, which serves as the passband of the ring bandpass filter. It should be noted that the stable frequency band value will vary under different circumstances. Therefore, the preset threshold used for selecting the stable frequency band is set based on actual conditions and the experience of those skilled in the art. Its purpose is to ensure that within this frequency band, the first normalized parameter... With the second normalized parameter The mutual energy hardly changes with the number of scattering layers.

[0035] It is understood that this embodiment determines the frequency range with small changes and a high energy ratio by comparing the cross power spectrum of the normalized polarization parameter, based on the actual application and purpose. This determines the high and low thresholds of the passband of the ring bandpass filter. Then, the covariance is calculated within this passband. In this embodiment, the covariance at this time is defined as the polarization statistical stability region.

[0036] Furthermore, after constructing a ring bandpass filter, the first normalized parameter is calculated at each incident position corresponding to the scattering scene to be imaged using the ring bandpass filter. Second normalized parameter Apply filters separately, within the passband of the ring bandpass filter, the first normalized parameter With the second normalized parameter The cross-energy of the bandpass is almost unchanged with the increase of the number of scattering layers, making it a naturally "scattering-number-insensitive passband". Extracting the first and second normalized parameters after filtering is equivalent to constructing a local covariance estimator using the physically driven bandpass cross spectrum, which avoids the scale bias of the fixed window and filters out low-frequency debiasing drift. This step does not require transfer matrix measurement and is not limited by the field of view of optical memory effect.

[0037] S4. Within a preset local window, calculate the covariance between the first normalized parameter and the second normalized parameter after filtering corresponding to the incident position, and obtain the corresponding covariance map. Furthermore, the scale of the local window is determined based on the speckle coherence scale generated by the scattering scene.

[0038] Specifically, in this embodiment, a speckle intensity image of the scattering medium's exit surface is acquired, its autocorrelation function is calculated, and the full width at half maximum (FWHM) of the autocorrelation function is taken as the speckle coherence scale. Local window side length Pick ,in The value of is in the range of 3 to 5, so as to ensure that the window contains a sufficient number of pixels for stable estimation of the local covariance of the normalized polarization parameter, so that the covariance estimation has statistical significance.

[0039] Furthermore, regarding the window side length Within a local window, calculate the first normalized parameter after filtering. Second normalized parameter The covariance is expressed by the formula: ; in, For the scale of a local window, and These are the first and second normalized parameters after filtering, respectively. The first normalized parameter after filtering Second normalized parameter covariance, It represents the mathematical expectation.

[0040] It is understandable that in this embodiment, the first normalized parameters are extracted by constructing a ring bandpass filter. Second normalized parameter The high-frequency components are used to obtain the first normalized parameter after filtering. Second normalized parameter The first normalized parameter after filtering is calculated within a neighborhood matching the speckle size. Second normalized parameter covariance This allows us to obtain a clean local polarization covariance feature map. This process is equivalent to constructing a physically driven local covariance estimator using the bandpass cross spectrum, which avoids the scale bias of a fixed window size, filters out intensity attenuation and low-frequency depolarization drift, and extracts polarization common-mode modes consistent with single scattering statistics.

[0041] Furthermore, it is understandable that during the depolarization process of multiple scattering, the light intensity continuously attenuates due to spatial divergence, and the first-order expectation of the normalized Stokes parameter eventually tends towards zero due to the disordered superposition of random phases. However, the local spatial covariance exhibits different variation characteristics. The physical essence of this phenomenon is that the local covariance measures the joint fluctuation characteristics of the filtered first and second normalized parameters at a spatial microscale. According to the definition of the normalized Stokes parameter, i.e., the first and second normalized parameters, the external envelope generated by the light intensity degradation is eliminated. At this point, the local spatial correlation between the filtered first and second normalized parameters is determined solely by the intrinsic optical properties of the medium surface in a single scattering event.

[0042] It is worth noting that in this embodiment, the first normalized parameter after filtering is calculated within a local window that matches the speckle size. Second normalized parameter covariance The first-order expectation tends to zero after multiple scatterings due to the random phase superposition, while the first normalized parameter after filtering... Second normalized parameter The local spatial joint fluctuation characteristics, i.e., the covariance, are determined solely by the intrinsic optical properties of the medium surface during a single scattering and exhibit a relatively stable trend. This quantity is the core statistic of this embodiment and a direct measure of the polarization statistical stability region.

[0043] S5. Construct a large field-of-view covariance map based on the covariance maps corresponding to all incident positions, and use the large field-of-view covariance map as physical constraint information to reconstruct the scattering scene using a preset scattering imaging model to obtain a scattering image.

[0044] Furthermore, in step S5, the large field-of-view covariance map is used as physical constraint information, and the scattering scene is reconstructed using a preset scattering imaging model, including: The large field-of-view covariance map is used as a spatial weight mask and embedded into the loss function of the scattering imaging model to weight the error between the scattering image output by the scattering imaging model and the preset target reference image. In the loss function, when weighting the reconstruction error of each pixel of the scattering image, if the covariance value of the pixel in the large field-of-view covariance map is higher, the weight of the pixel is greater.

[0045] Specifically, in this embodiment, steps S2 to S4 are repeated for all preset incident positions in the object plane to obtain the covariance map of each position. The maps are then stitched together in spatial coordinate order to form a large field-of-view covariance map that covers the target's field of view.

[0046] A large field-of-view covariance map is used as physical constraint information and input into a deep learning-based descattering network. During model training or inference, the large field-of-view covariance map can be used as a spatial weight mask embedded in the loss function to weight the error between the reconstructed image and the target image: the higher the covariance value, the greater the weight of the pixel position, guiding the model to prioritize the recovery of the target structure in physically stable regions. The final output is a high-quality equivalent single-scattering image.

[0047] Furthermore, after constructing the large field-of-view covariance map based on the covariance maps of all incident positions, the method further includes: The large field-of-view covariance map and the polarization feature map calculated from the polarization image set are stitched together in the channel dimension, and the resulting multi-channel feature map is used as the input to the scattering imaging model.

[0048] Understandably, this embodiment concatenates the large field-of-view covariance map and polarization feature map along the channel dimension as input to the scattering imaging model. This approach enables the network to simultaneously perceive the original polarization information and the derived stable covariance features during the feature extraction stage, preventing the physical prior from being diluted during deep network propagation. Furthermore, channel concatenation, as a standard operation in deep learning, does not introduce additional parameters and requires no modification to the network backbone structure, achieving effective injection of physical prior with zero computational cost.

[0049] It is worth noting that this embodiment also experimentally collected the polarization point spread function at the same point under three scattering conditions: single-layer, closely adjacent double-layer, and sandwich-spaced structure. Under the premise that the incident position, detection conditions, and polarization state were consistent, the polarization angle variance, polarization degree variance, Stokes mean, and normalized covariance were compared. This verifies that the polarization stability characteristic is not a local random phenomenon, but a statistically invariant with spatial repeatability under the condition of super-optical memory effect, thus providing experimental evidence for the promotion of large field of view. Furthermore, in the experiment, the polarization point spread function at each spatial location was also compared. Define interlayer stability index The formula is expressed as: ; in, and Representing different scattering medium structures, The polarization characteristics after stable frequency band filtering, interlayer stability index The smaller the value, the more consistent the polarization characteristics at that location remain with increasing scattering layers. (Based on a threshold value) Selecting stable propagation subspaces This allows for the spatial location of the "physically reliable region." Experimental results show that a large number of spatial locations satisfy this condition, confirming the objective existence of the polarization statistical stability region.

[0050] Understandably, the large field-of-view covariance map itself is a stable feature map that is insensitive to the number of scatterings and can be directly used as the polarization statistical stability region. In this embodiment, the polarization statistical stability region is embedded in the network in two ways. First, it serves as a physical constraint anchor point in the loss function, providing stronger supervision signals in stable regions rich in polarization information. For example, the stability region can be used as a spatial weight mask, assigning higher weights to pixels within the stability region in the loss function, making the network focus more on physically reliable regions during training. Second, it relies more on data-driven compensation capabilities in regions with severe depolarization. For example, the stability region (or large field-of-view covariance map) can be used as an additional input channel of the network, concatenated with the original polarization image or the polarization feature map calculated from the polarization image, and input into the network together, guiding the network to learn the mapping relationship between stable regions and reconstruction results. This fundamentally alleviates the dependence of data-driven methods on paired training samples and gives the network generalization robustness across scattering medium parameters.

[0051] The final result is a target reconstruction image with a super optical memory effect range and a large field of view, which can be applied to complex dynamic scattering scenes such as smoke, turbid water, and biological tissue.

[0052] It is understood that this embodiment acquires intensity images of multiple linear polarization directions at multiple preset incident positions within the object plane, and performs normalization of Stokes vector parameters, annular bandpass filtering, and local window covariance calculation on each incident position sequentially to obtain the covariance value of each incident position. These images are then stitched together to form a large-field-of-view covariance map covering the target's field of view. The polarization statistical characteristics represented by this large-field-of-view covariance map are insensitive to changes in the number of scattering events and can stably exist under strong scattering conditions where the polarization degree approaches zero. Physically, it equates complex multiple scattering to single scattering, solving the fundamental problem of existing technologies' difficulty in extracting weak target information due to inaccurate scattering transmission models. Simultaneously, by acquiring covariance values ​​point-by-point within a large field of view and stitching them together, it no longer relies on the translation invariance required by traditional optical memory effects, effectively... This method overcomes the rigid limitation of the imaging field of view imposed by the memory effect, enabling it to be applied to non-uniform and complex scattering scenarios (such as smoke, turbid water, and biological tissue) within the range of super-optical memory effects, achieving stable imaging with a large field of view. Furthermore, the obtained large field of view covariance map is embedded into the scattering imaging model as physical constraint information. For example, it can be used as a spatial weight mask to weight the reconstruction error in the loss function, or as an additional input channel to guide the network to learn the mapping relationship between stable regions and target structures. This combination of physical prior and data-driven approach alleviates the dependence on paired training samples, endows the network with cross-parameter generalization ability for changes in scattering medium parameters, significantly improves the image reconstruction quality and robustness under strong scattering environments, and overcomes the shortcomings of existing methods in terms of severely limited reconstruction capabilities in non-uniform, super-memory effect, and complex scattering media.

[0053] Example 2 Unlike Example 1, as Figure 2 As shown, this embodiment provides a device for multiple scattering polarization feature extraction and equivalent imaging, including a polarization information acquisition module 201, a normalization module 202, a bandpass filtering module 203, a covariance map calculation module 204, and an imaging module 205. Wherein: The polarization information acquisition module 201 is used to acquire a set of polarization images of multiple preset incident positions in the object plane, the set of polarization images including intensity images of the scattering scene after multiple scattering in multiple linear polarization directions. The normalization module 202 is used to calculate the Stokes vector based on the polarization image set corresponding to each incident position, and normalize multiple Stokes parameters of the Stokes vector to obtain the corresponding first normalization parameter and second normalization parameter. The bandpass filter module 203 is used to perform bandpass filtering on the first normalized parameter and the second normalized parameter corresponding to the incident position based on a preset ring bandpass filter. The rate of change of the mutual energy between the filtered first normalized parameter and the second normalized parameter with the increase of the number of scattering layers is lower than a preset threshold. The covariance plot calculation module 204 is used to calculate the covariance between the filtered first normalized parameter and the second normalized parameter corresponding to the incident position within a preset local window, and obtain the corresponding covariance plot. The imaging module 205 is used to construct a large field-of-view covariance map based on the covariance map corresponding to all incident positions, and use the large field-of-view covariance map as physical constraint information to reconstruct the scattering scene through a preset scattering imaging model to obtain a scattering image.

[0054] Example 3 Unlike Embodiments 1 and 2, this embodiment provides an electronic device, such as... Figure 3 As shown, it includes a processor 301, a communication interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other through the communication bus 304. Memory 303 is used to store computer programs; When the processor 301 executes the program stored in the memory 303, it implements the steps of the multiple scattering polarization feature extraction and equivalent imaging method in Embodiment 1.

[0055] The communication bus 304 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 304 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0056] Communication interface 302 is used for communication between the above-mentioned electronic device and other devices.

[0057] The memory 303 may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory 303 may also be at least one storage device located remotely from the aforementioned processor.

[0058] The processor 301 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0059] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0060] Example 4 Unlike the above embodiments, this embodiment also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the multiple scattering polarization feature extraction and equivalent imaging method provided in Embodiment 1.

[0061] For the embodiments of the device / electronic device / storage medium, since they are basically similar to the method embodiments, the description is relatively simple. For specific details and beneficial effects, please refer to the description of the method embodiments.

[0062] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0063] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0064] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus (devices), or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.

[0065] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for extracting polarization features and equivalent imaging through multiple scattering, characterized in that, include: Acquire a set of polarization images at multiple preset incident positions within the object plane. These polarization images include intensity images of the scattered scene after multiple scattering events at four linear polarization directions: 0°, 45°, 90°, and 135°. For each incident position, a Stokes vector is calculated based on the polarization image set corresponding to that incident position, and multiple Stokes parameters of the Stokes vector are normalized to obtain a first normalized parameter and a second normalized parameter. The Stokes parameters of the Stokes vector include a first light intensity, a second light intensity, and a third light intensity. The first light intensity is the result of pixel-by-pixel addition of a first intensity image and a second intensity image; the second light intensity is the result of pixel-by-pixel subtraction of a first intensity image and a second intensity image; and the third light intensity is the result of pixel-by-pixel subtraction of a third intensity image and a fourth intensity image. The first intensity image, the second intensity image, the third intensity image, and the fourth intensity image are the intensity images of the scattering scene at four linear polarization directions: 0°, 90°, 45°, and 135°, respectively. The normalization of the multiple Stokes parameters of the Stokes vector includes: Calculate the ratio of the second light intensity to the first light intensity to obtain the first normalized parameter; Calculate the ratio of the third light intensity to the first light intensity to obtain the second normalized parameter; Constructing a ring bandpass filter includes: At a preset first incident position, multiple first polarization image sets are acquired. Each first polarization image set includes an intensity image of a preset scattering scene in multiple linear polarization directions. The scattering medium structure of the preset scattering scene includes a single-layer scattering medium, a double-layer closely attached scattering medium, and a sandwich structure of two scattering media separated by air. For the first polarization image set corresponding to the preset scattering scenario for each scattering medium structure, calculate the first normalization parameter and the second normalization parameter corresponding to the first polarization image set, and calculate the cross power spectrum of the first normalization parameter and the second normalization parameter to obtain the cross power spectrum corresponding to the first polarization image set. The frequency range in which the rate of change of energy with increasing number of scattering layers is lower than the first preset threshold is defined as the stable frequency range in each cross power spectrum corresponding to different scattering medium structures. The passband of the ring bandpass filter is determined based on the lower and upper limits of the stable frequency range, and the ring bandpass filter is constructed. The first normalized parameter and the second normalized parameter corresponding to the incident position are bandpass filtered based on the ring bandpass filter. Within a preset local window, the covariance between the filtered first normalized parameter and the second normalized parameter corresponding to the incident position is calculated to obtain the corresponding covariance map. A large field-of-view covariance map is constructed based on the covariance maps corresponding to all incident positions. The large field-of-view covariance map is used as physical constraint information, and the scattering scene is reconstructed using a preset scattering imaging model to obtain a scattering image.

2. The method for extracting polarization features and equivalent imaging through multiple scattering according to claim 1, characterized in that, The scale of the local window is determined based on the speckle coherence scale generated by the scattering scene.

3. The method for extracting polarization features and equivalent imaging through multiple scattering according to claim 1, characterized in that, The step of using the large field-of-view covariance map as physical constraint information to reconstruct the image of the scattering scene through a preset scattering imaging model includes: The large field-of-view covariance map is used as a spatial weight mask and embedded into the loss function of the scattering imaging model to weight the error between the scattering image output by the scattering imaging model and the preset target reference image. In the loss function, when weighting the reconstruction error of each pixel of the scattering image, if the covariance value of the pixel in the large field-of-view covariance map is higher, the weight of the pixel is greater.

4. The method for extracting polarization features and equivalent imaging through multiple scattering according to claim 1, characterized in that, After constructing the large field-of-view covariance map based on the covariance maps corresponding to all incident positions, the method further includes: The large field-of-view covariance map and the polarization feature map calculated from the polarization image set are stitched together in the channel dimension, and the resulting multi-channel feature map is used as the input to the scattering imaging model.

5. A device for extracting polarization features of multiple scattering and equivalent imaging, characterized in that, The method for multiple scattering polarization feature extraction and equivalent imaging according to any one of claims 1-4 includes: The polarization information acquisition module is used to acquire a set of polarization images of multiple preset incident positions in the object plane. The set of polarization images includes intensity images of the scattering scene after multiple scatterings in multiple linear polarization directions. The normalization module is used to calculate the Stokes vector based on the polarization image set corresponding to each incident position, and normalize multiple Stokes parameters of the Stokes vector to obtain the corresponding first normalization parameter and second normalization parameter. The bandpass filtering module is used to perform bandpass filtering on the first normalized parameter and the second normalized parameter corresponding to the incident position based on a preset ring bandpass filter. The rate of change of the mutual energy between the first normalized parameter and the second normalized parameter after filtering with the increase of the number of scattering layers is lower than a preset threshold. The covariance plot calculation module is used to calculate the covariance between the first normalized parameter and the second normalized parameter after filtering at the incident position within a preset local window, and obtain the corresponding covariance plot. The imaging module is used to construct a large field-of-view covariance map based on the covariance map corresponding to all incident positions, and use the large field-of-view covariance map as physical constraint information to reconstruct the scattering scene through a preset scattering imaging model to obtain a scattering image.

6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the multiple scattering polarization feature extraction and equivalent imaging method according to any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multiple scattering polarization feature extraction and equivalent imaging method according to any one of claims 1-4.

Images