Night vision imaging method and system with low-light and infrared image fusion

Abstract

The invention relates to the technical field of image processing, in particular to a low-light and infrared image fused night vision imaging method and system. According to the method, multi-source data collaboration is realized by constructing a dual-band polarization feature cube, a degradation rule of fog scattering on low-light texture is accurately quantified, and a scattering invariance enhanced image is generated; based on fog concentration gradient dynamic partitioning, low-light details and infrared contours are fused in a low-concentration area, polarization state penetrating dense fog is recombined in a high-concentration area, finally, a night vision image with high definition, high scattering resistance and scene adaptability is output, and the identification capacity and monitoring reliability of fog night road targets are improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a night vision imaging method and system for fusing low-light-level and infrared images. Background Art

[0002] In foggy night road environments, traditional night vision imaging technology is limited by problems such as fog scattering, low illumination, and single band information, making it difficult to achieve clear and reliable scene perception. Although low-light imaging can enhance visible light information, it is easily affected by Mie scattering, resulting in texture degradation; although infrared imaging has the ability to penetrate fog, it lacks detailed texture and polarization characteristics, making it difficult to meet monitoring needs in complex environments. Existing fusion methods are mostly based on single light intensity or radiation characteristics, which do not fully utilize the scattering invariance of polarization information and are not adaptable to dynamic changes in fog concentration, resulting in problems such as loss of details, blurred edges, or thermal radiation distortion in the fused image. In addition, traditional fusion algorithms usually ignore the physical relationship between polarization state and optical path difference, making it difficult to effectively correct scattering degradation, especially in dense fog areas, where the complementarity of infrared and low-light information has not been fully explored. In view of this, the present application proposes a night vision imaging method and system that fuses low-light and infrared images to achieve optimized fusion of multimodal information under fog interference, thereby improving imaging clarity and scene understanding capabilities. Summary of the Invention

[0003] The present invention overcomes the deficiencies of the prior art and provides a night vision imaging method and system for fusing low-light-level and infrared images.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: The first aspect of the present invention discloses a night vision imaging method for fusing low-light level and infrared images, comprising the following steps: S102: In a foggy night road environment, synchronously collect low-light polarization images and long-wave infrared polarization images to construct a spatiotemporally aligned dual-band polarization feature cube; S104: Extracting the fog concentration gradient tensor of the monitoring scene and separating the degraded texture feature map modulated by Mie scattering based on the dual-band polarization feature cube, establishing an optical path difference mapping relationship between the low-light degraded texture and the fog concentration field, and generating an optical path difference mapping relationship matrix; S106: generating a joint compensation operator based on the optical path difference mapping relationship matrix, performing polarization channel transmittance correction on the low-light degradation texture in combination with the compensation operator, and outputting an enhanced low-light feature map with fog scattering invariance; S108: Analyze the fog concentration gradient tensor and dynamically divide the fog concentration area into a low concentration area and a high concentration area according to the Laplace norm response value of the fog concentration gradient tensor; S110: In low-concentration areas, the enhanced low-light feature map is fused with the infrared thermal radiation intensity using a gradient tensor with fidelity. In high-concentration areas, the infrared polarization vector direction angle and the Stokes component of the enhanced low-light feature map are recombined using a polarization state tensor to generate a night vision image.

[0005] Preferably, the step 102 is specifically: Synchronously collect light intensity data in four polarization directions in the low-light band and radiation intensity data in two orthogonal polarization directions in the long-wave infrared band, generating a low-light polarization intensity image group and a long-wave infrared polarization radiation intensity image group respectively; Perform Stokes vector calculation on the low-light polarization image group to obtain the low-light intensity component and polarization characteristic component; perform thermal radiation polarization calculation on the long-wave infrared polarization image group to obtain the infrared radiation intensity component and polarization degree component; Based on the spatial gradient characteristics of the low-light intensity component and the thermal radiation profile characteristics of the infrared radiation intensity component, a spatial registration matrix is ​​generated through affine transformation to achieve sub-pixel alignment of the two band images. The polarization correlation between the low-light polarization characteristic component and the infrared polarization degree component is used to determine the polarization correlation between the bands, and a fusion weight matrix is ​​generated based on the polarization correlation. Apply the spatial registration matrix to perform geometric correction on the infrared radiation intensity component to obtain the corrected infrared radiation intensity; perform polarization direction compensation on the infrared polarization component to eliminate the polarization angle offset caused by parallax and output the corrected infrared polarization degree; The corrected infrared radiation intensity and the low-light intensity component are fused according to the fusion weight matrix to generate a fused intensity feature; the corrected infrared polarization degree and the low-light polarization characteristic component are recombined in polarization state to generate a fused polarization feature; The fused light intensity features, corrected infrared radiation intensity and fused polarization features are superimposed according to the spatial dimension to construct a six-channel dual-band polarization feature cube containing light intensity, radiation and polarization information.

[0006] Preferably, the step 104 is specifically: Perform pixel-by-pixel difference calculation on the low-light intensity component and infrared radiation intensity component in the dual-band polarization characteristic cube to obtain the dual-band differential intensity parameter. Use the Stokes parameters of the low-light polarization characteristic component to calculate the polarization difference ratio and obtain the polarization modulation difference parameter. An initial fog concentration distribution map is generated based on the polarization modulation difference parameter and the dual-band differential intensity parameter; an anisotropic diffusion filter is performed on the initial fog concentration distribution map, and the thermal radiation polarization constraint condition of the infrared polarization component is combined to suppress noise and retain the fog edge structure, and an optimized fog concentration gradient tensor is output; By using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation, a mapping relationship between polarization modulation and light intensity degradation is established. Through this mapping relationship, a degraded texture feature map modulated by Mie scattering is decoupled from the low-light intensity component. A three-dimensional distribution model of the fog concentration field is constructed based on the optimized fog concentration gradient tensor. At the same time, the optical path accumulation of the low-light degraded texture in the fog concentration field is determined according to the local contrast attenuation rate of the degraded texture feature map. The mapping relationship between the grayscale attenuation gradient of the degraded texture feature map and the optical path accumulation in the fog concentration field is established through the nonlinear regression method, and a calibrated optical path difference mapping relationship matrix is ​​generated. Each element of the matrix represents the degree of scattering degradation of the low-light texture under a specific fog concentration.

[0007] Preferably, a mapping relationship between polarization modulation and light intensity degradation is established by using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation. The degraded texture feature map modulated by Mie scattering is decoupled from the low-light intensity component through the mapping relationship, specifically: Extract the Stokes vector of the low-light polarization characteristic component in the dual-band polarization characteristic cube, obtain the phase angle distribution diagram of the Stokes vector, and simultaneously obtain the local contrast attenuation rate matrix of the low-light intensity component; Normalizing the phase angle distribution diagram to generate a polarization modulation phase characteristic diagram, and establishing a polarization phase-attenuation statistical coupling relationship by Pearson correlation analysis based on a local contrast attenuation rate matrix; Based on the polarization phase-attenuation statistical coupling relationship, the nonlinear least squares method is used to fit the mapping curve between the polarization modulation phase characteristic map and the local contrast attenuation rate matrix, and the polarization phase-attenuation coupling coefficient matrix is ​​obtained by fitting. The polarization phase-attenuation coupling coefficient matrix is ​​applied to the low-light intensity component, and the global attenuation base layer dominated by Mie scattering is separated through inverse mapping operation to obtain the residual texture component. Anisotropic guided filtering is performed on the residual texture components. The polarization degree component of the Stokes vector is used as the edge constraint weight to suppress noise and preserve the topological structure of the scattering-degraded texture. The degraded texture feature map modulated by Mie scattering is output. The degraded texture feature map is differentially calculated with the global attenuation base layer to verify the linear independence of the light intensity attenuation gradient and the polarization modulation phase angle, and finally the scattering modulation domain of the degraded texture feature map is calibrated.

[0008] Preferably, the step 106 is specifically: According to the optical path accumulation of each pixel in the optical path difference mapping relationship matrix, the transmittance attenuation coefficient of the polarization channel corresponding to the low-light degradation texture feature map is determined, and a transmittance attenuation coefficient distribution map is generated; Combined with the polarization phase angle distribution of the Stokes vector, a nonlinear coupling relationship between the transmittance attenuation coefficient and the polarization phase angle is established to generate the polarization-transmittance coupling tensor. The infrared radiation intensity component in the dual-band polarization characteristic cube is used to extract the penetration characteristic curve of thermal radiation in foggy medium. By calibrating the inverse proportional relationship between infrared thermal radiation intensity and fog concentration, a thermal radiation penetration compensation coefficient matrix is ​​constructed. Performing a tensor dot multiplication operation on the polarization-transmittance coupling tensor and the thermal radiation penetration compensation coefficient matrix to obtain a joint compensation operator that simultaneously integrates the polarization modulation characteristics and the thermal radiation penetration characteristics; The joint compensation operator is applied to the low-light degradation texture feature map. The transmittance attenuation coefficient is corrected pixel by pixel to eliminate the polarization channel energy attenuation caused by Mie scattering, and the intermediate low-light texture after transmittance correction is output. The polarization component of the Stokes vector is used to enhance the edge of the intermediate low-light texture, and the local contrast lost due to scattering is restored by polarization weighting to generate an enhanced low-light feature map with fog scattering invariance.

[0009] Preferably, the step 108 is specifically: Perform Laplace operator convolution on the fog concentration gradient tensor, calculate the second-order spatial derivative at each pixel position, and generate a Laplace response map of the fog concentration; Based on the local extreme value distribution characteristics of the Laplace response graph, the mutation boundary of the fog concentration change is extracted to form the concentration partition boundary; Performing morphological closing operation smoothing on the concentration partition boundary to eliminate holes caused by noise, and combining the spatial consistency constraint of the infrared polarization component to correct the mis-segmented area caused by thermal radiation interference, and generate a dynamic partition mask; According to the high and low concentration areas marked in the dynamic partition mask, the pixels of the monitoring scene are classified into low concentration areas and high concentration areas; the low concentration area is the area with a mask value of 0, and the high concentration area is the area with a mask value of 1.

[0010] Preferably, the step 110 is specifically: In low-concentration areas, the fog concentration gradient tensor is extracted to enhance the low-light feature map, and the thermal radiation profile gradient field is parsed from the infrared thermal radiation intensity. The fog concentration gradient tensor and the thermal radiation profile gradient field are subjected to pixel-by-pixel gradient structure similarity measurement to generate a gradient fidelity weight map. Based on the gradient fidelity weight map, the low-frequency component of the enhanced low-light feature map and the high-frequency component of the infrared thermal radiation intensity are adaptively weighted and fused to obtain a primary fusion feature; The local anisotropy coefficient of the fog concentration gradient tensor is used to rebalance the gradient field of the primary fusion features, eliminate the radiation distortion at the fusion boundary, and output the fusion result of the low concentration area.

[0011] Also includes: In high-concentration areas, the spatial distribution matrix of the infrared polarization vector direction angle is calculated, and the polarization phase angle characteristics in the Stokes component of the enhanced low-light feature map are extracted. The spatial distribution matrix and the polarization phase angle characteristics are subjected to polarization state covariance analysis to establish the polarization direction-phase coupling tensor. Performing Stokes space projection transformation on the infrared polarization direction angle by using the coupling tensor to generate a recombined polarization basis; The Laplace norm response value of the fog concentration gradient tensor is used to dynamically adjust the weight ratio of the reconstructed polarization basis and the enhanced low-light feature map, and the tensor product operation of the polarization channel is performed to obtain the polarization reconstructed feature in the high-concentration area. Finally, the fusion results of the low-concentration area and the polarization reconstruction features of the high-concentration area are spatially spliced ​​through a dynamic partition mask, and the final night vision imaging image is generated after a smooth transition using bilinear interpolation.

[0012] A second aspect of the present invention discloses a night vision imaging system for fusing low-light level and infrared images. The night vision imaging system includes a memory and a processor. The memory stores a night vision imaging method program for fusing low-light level and infrared images. When the night vision imaging method program for fusing low-light level and infrared images is executed by the processor, any one of the steps of the night vision imaging method for fusing low-light level and infrared images is implemented.

[0013] The present invention solves the technical defects existing in the background technology, and has the following beneficial effects: by constructing a dual-band polarization feature cube to realize multi-source data collaboration, the degradation law of fog scattering on low-light texture is accurately quantified, and a scattering invariance enhanced image is generated; based on the dynamic partitioning of the fog concentration gradient, the low-light details and infrared contours are integrated in the low-concentration area, and the polarization state is recombined in the high-concentration area to penetrate the thick fog, and finally the night vision image with high definition, strong anti-scattering and scene adaptability is output, thereby improving the recognition ability and monitoring reliability of road targets in foggy nights. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0015] Figure 1 This is the overall method flow chart of the night vision imaging method; Figure 2 This is a partial flow chart of the night vision imaging method; Figure 3 This is the system block diagram of this night vision imaging system. DETAILED DESCRIPTION

[0016] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0018] like Figure 1 As shown, the first aspect of the present invention discloses a night vision imaging method for fusing low-light level and infrared images, comprising the following steps: S102: In a foggy night road environment, synchronously collect low-light polarization images and long-wave infrared polarization images to construct a spatiotemporally aligned dual-band polarization feature cube; S104: Extracting the fog concentration gradient tensor of the monitoring scene and separating the degraded texture feature map modulated by Mie scattering based on the dual-band polarization feature cube, establishing an optical path difference mapping relationship between the low-light degraded texture and the fog concentration field, and generating an optical path difference mapping relationship matrix; S106: generating a joint compensation operator based on the optical path difference mapping relationship matrix, performing polarization channel transmittance correction on the low-light degradation texture in combination with the compensation operator, and outputting an enhanced low-light feature map with fog scattering invariance; S108: Analyze the fog concentration gradient tensor and dynamically divide the fog concentration area into a low concentration area and a high concentration area according to the Laplace norm response value of the fog concentration gradient tensor; S110: In low-concentration areas, the enhanced low-light feature map is fused with the infrared thermal radiation intensity using a gradient tensor with fidelity. In high-concentration areas, the infrared polarization vector direction angle and the Stokes component of the enhanced low-light feature map are recombined using a polarization state tensor to generate a night vision image.

[0019] It should be noted that the present invention solves the problems of low-light image texture degradation, infrared image detail loss, and poor adaptability of dual-band fusion caused by fog scattering in traditional night vision imaging in foggy night road environments, overcoming the limitations of single-modal information and imaging distortion defects under dynamic fog interference. By constructing a dual-band polarization feature cube to achieve multi-source data collaboration, the degradation law of low-light texture caused by fog scattering is accurately quantified, and a scattering-invariant enhanced image is generated; based on the dynamic partitioning of fog concentration gradient, low-light details and infrared contours are fused in low-concentration areas, and polarization states are recombined in high-concentration areas to penetrate dense fog. The final output is a night vision image with high definition, strong anti-scattering and scene adaptability, which improves the recognition ability and monitoring reliability of foggy night road targets.

[0020] Preferably, the step 102 is specifically: Synchronously collect light intensity data in four polarization directions (0°, 45°, 90°, and 135°) in the low-light band (400-700nm), and radiation intensity data in two orthogonal polarization directions (horizontal and vertical) in the long-wave infrared band (8-14μm), generating a low-light polarization intensity image group and a long-wave infrared polarization radiation intensity image group respectively; Perform Stokes vector calculation on the low-light polarization image group to obtain the low-light intensity component and polarization characteristic component; perform thermal radiation polarization calculation on the long-wave infrared polarization image group to obtain the infrared radiation intensity component and polarization degree component; It should be noted that Stokes vector solution is performed on the low-light polarization image group (in four directions: 0°, 45°, 90°, and 135°). By statistically analyzing the linear combination relationship of the light intensity in each polarization direction, the low-light intensity component representing the total light intensity and the polarization degree and polarization angle reflecting the polarization characteristics are obtained. At the same time, thermal radiation polarization solution is performed on the long-wave infrared polarization image group (horizontal and vertical directions). The infrared radiation intensity component is calculated using the intensity difference and sum of the two orthogonal polarization components, and the infrared polarization degree is derived in combination with Malus's law, thereby separating the radiation and polarization characteristic information of the target.

[0021] Based on the spatial gradient characteristics of the low-light intensity component and the thermal radiation profile characteristics of the infrared radiation intensity component, a spatial registration matrix is ​​generated through affine transformation to achieve sub-pixel alignment of the two band images. The polarization correlation between the low-light polarization characteristic component and the infrared polarization degree component is used to determine the polarization correlation between the bands, and a fusion weight matrix is ​​generated based on the polarization correlation. It should be noted that by extracting the edge gradient features of the low-light intensity component (such as road markings and vehicle contours) and the thermal distribution features of the infrared radiation intensity (such as pedestrians and vehicle heat sources), the affine transformation parameters are calculated through feature point matching to generate a spatial registration matrix, so that the low-light and infrared images are aligned to sub-pixel accuracy; at the same time, the spatial distribution consistency of the low-light polarization angle and the infrared polarization degree is analyzed, the covariance matrix of the two is calculated, and the polarization correlation between the bands is determined (such as the correlation between the polarized reflection of the road surface and the polarization of thermal radiation), and an adaptive fusion weight matrix is ​​generated based on this, so that the fusion process retains the effective information of each band.

[0022] Apply the spatial registration matrix to perform geometric correction on the infrared radiation intensity component to obtain the corrected infrared radiation intensity; perform polarization direction compensation on the infrared polarization component to eliminate the polarization angle offset caused by parallax and output the corrected infrared polarization degree; It should be noted that a spatial registration matrix is ​​used to geometrically transform the infrared radiation intensity components (including translation, rotation, and scaling) to ensure strict alignment between the infrared image and the low-light-level image, outputting position-corrected infrared radiation intensity data. Simultaneously, based on the registered coordinate offset, the polarization direction angle of the infrared polarization component is compensated and calibrated to eliminate polarization angle deviations caused by dual-camera parallax (such as phase correction in the horizontal and vertical polarization directions). Ultimately, the output is infrared polarization information with both geometric and polarization direction corrections.

[0023] The corrected infrared radiation intensity and the low-light intensity component are fused according to the fusion weight matrix to generate a fused intensity feature; the corrected infrared polarization degree and the low-light polarization characteristic component are recombined in polarization state to generate a fused polarization feature; The fused light intensity features, corrected infrared radiation intensity and fused polarization features are superimposed according to the spatial dimension to construct a six-channel dual-band polarization feature cube containing light intensity, radiation and polarization information.

[0024] In a specific embodiment of the present invention, for monitoring dense fog at night on highways, a split-focus plane polarization low-light-level camera (with a response band of 400-700 nm) simultaneously captures low-light-level images at four polarization directions: 0°, 45°, 90°, and 135°. A long-wave infrared polarization camera (with a response band of 8-14 μm) simultaneously acquires radiation images in the horizontal and vertical polarization directions. The linear polarization degree and polarization angle of the low-light-level images are extracted using Stokes decomposition, and then combined with infrared polarization decomposition to obtain thermal radiation intensity and polarization degree. Affine transformation registration (with a registration error of <0.5 pixel) is performed based on road guardrail edge feature points (spatial gradient threshold >15 dB) and vehicle thermal radiation profiles (signal-to-noise ratio ≥8 dB). Fusion weights are generated using polarization correlation. Finally, a six-channel feature cube (data dimensions 1280×1024×6) is constructed, containing low-light intensity, infrared radiation, and polarization phase / degree. This effectively overcomes the detail loss problem of traditional single-band imaging in dense fog (visibility <50 m).

[0025] It should be noted that the present invention realizes high-precision registration and polarization information fusion of dual-band images, effectively improving the spatial alignment accuracy of images and the integrity of polarization information.

[0026] Preferably, if Figure 2 As shown, the 104 is specifically: S202: performing pixel-by-pixel differential calculation on the low-light intensity component and the infrared radiation intensity component in the dual-band polarization characteristic cube to obtain a dual-band differential intensity parameter, and calculating a polarization difference ratio using the Stokes parameters of the low-light polarization characteristic component to obtain a polarization modulation difference parameter; It should be noted that the Stokes parameters are extracted from the polarization characteristic components of the low-light, and the difference ratio of the polarization light intensities in orthogonal directions (such as 0° and 90°) is calculated (that is, the light intensity in the 0-degree polarization direction is subtracted from the light intensity in the 90-degree polarization direction, and then divided by the sum of the light intensities in these two directions). This obtains the difference parameters reflecting the polarization modulation characteristics of the target surface. At the same time, the phase relationship between different polarization directions is combined to eliminate the interference of ambient stray light, and finally the modulation difference parameters representing the polarization characteristics of the scene material are output.

[0027] S204: Generate an initial fog concentration distribution map based on the polarization modulation difference parameter and the dual-band differential intensity parameter; perform anisotropic diffusion filtering on the initial fog concentration distribution map, combine the thermal radiation polarization constraint condition of the infrared polarization component, suppress noise and retain the fog edge structure, and output the optimized fog concentration gradient tensor; wherein, the thermal radiation polarization constraint condition refers to the process of using the polarization characteristics generated by the thermal radiation of the object itself in long-wave infrared polarization imaging (such as the differential polarization response of metal / non-metal surfaces) as a physical constraint to correct and optimize fog concentration detection.

[0028] It should be noted that a weighted fusion is performed on the polarization modulation difference parameter (reflecting the scattering characteristics of light polarized in different directions) and the dual-band differential intensity parameter (the intensity difference between the low-light and infrared images). The polarization parameter primarily indicates the degree of fog modulation of polarized light, while the dual-band differential parameter reflects the attenuation characteristics of light intensity due to fog. The fused characteristic values ​​are then converted into an initial fog concentration estimate for each pixel using a preset fog concentration calibration curve. Finally, this calculation is performed for all pixels in the entire image, generating a two-dimensional matrix diagram reflecting the fog concentration distribution in each area of ​​the scene, where higher values ​​indicate greater fog concentration at that location.

[0029] S206: establishing a mapping relationship between polarization modulation and light intensity degradation by using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation, and decoupling the degraded texture feature map modulated by Mie scattering from the low-light intensity component through the mapping relationship; S208: constructing a three-dimensional distribution model of the fog concentration field based on the optimized fog concentration gradient tensor, and determining the optical path accumulation of the low-light degraded texture in the fog concentration field according to the local contrast attenuation rate of the degraded texture feature map; It should be noted that the optimized two-dimensional fog concentration gradient tensor (including the horizontal and vertical concentration change rates) is interpolated and expanded in three-dimensional space. Combined with scene depth information (such as distance data obtained through binocular vision or lidar), the fog concentration gradient at each pixel is converted into a concentration distribution vector in three-dimensional space. These vector fields are then integrated using the Poisson reconstruction method to generate a continuous three-dimensional fog concentration field model, where the X / Y axes correspond to the image plane coordinates and the Z axis represents the fog concentration value (ranging from 0 to 1). The final output is a three-dimensional concentration field model that reflects the distribution and variation of fog in three-dimensional space.

[0030] It should be noted that a sliding window (e.g., 5×5 pixels) is used on the degraded texture feature map to calculate the local contrast decay rate (the contrast ratio between the original and degraded images) for each region. This is combined with the spatial concentration values ​​at corresponding locations in the 3D fog concentration field model to establish a physical relationship between light intensity attenuation and fog concentration. The cumulative effect of light passing through fog layers of varying concentrations is then integrated along the line of sight, ultimately outputting the optical path cumulate (a dimensionless parameter) corresponding to each pixel. This parameter directly reflects the overall scattering attenuation experienced by the low-light texture during propagation.

[0031] S210: A mapping relationship between the grayscale attenuation gradient of the degraded texture feature map and the optical path accumulation amount of the fog concentration field is established through a nonlinear regression method, and a calibrated optical path difference mapping relationship matrix is ​​generated. Each element of the matrix represents the degree of scattering degradation of the low-light texture under a specific fog concentration.

[0032] It should be noted that low-light image samples were collected at different fog concentrations, and the grayscale attenuation gradients (such as the local contrast drop rate) of their degraded texture feature maps were extracted along with the optical path accumulation data at the corresponding locations to form a training set. A nonlinear fitting method, such as Gaussian process regression, was then used to establish a mapping function from optical path accumulation to grayscale attenuation gradients. This function was then applied to the entire scene, generating an optical path difference mapping value for each pixel location that reflects the degree of scattering attenuation at that specific fog concentration. This formed a calibration matrix of the same size as the original image, in which each element represents the correction factor for the light intensity attenuation caused by fog at that location.

[0033] In a specific embodiment of the present invention, in a highway fog monitoring scenario, pixel-by-pixel differentiation is performed between the registered low-light intensity component and infrared radiation intensity to generate a dual-band differential intensity map (dynamic range 0-255). This is combined with the polarization difference ratio (0°-90° direction ratio 1.2-1.8) calculated from the low-light Stokes parameters to generate an initial fog concentration distribution map. Anisotropic diffusion filtering and infrared polarization degree constraints (threshold 0.15-0.3) are used to output an optimized fog gradient tensor. The Pearson correlation between the Stokes phase angle (0-180°) and light intensity attenuation is used to isolate degraded texture dominated by Mie scattering. Finally, an optical path difference mapping matrix (256×256 quantization levels) is constructed to accurately reflect the scattering attenuation of lane markings (15 cm width) at varying fog concentrations (0.1-0.5 g / m³).

[0034] It should be noted that this method can effectively distinguish the fog interference in the scene from the texture of real objects, retain the details of the fog edge while suppressing noise. The optical path difference mapping relationship matrix finally constructed can accurately reflect the scattering effect of different fog concentrations on low-light imaging, providing a reliable degradation feature basis for subsequent image restoration, thereby improving the accuracy of target recognition in foggy environments.

[0035] Preferably, a mapping relationship between polarization modulation and light intensity degradation is established by using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation. The degraded texture feature map modulated by Mie scattering is decoupled from the low-light intensity component through the mapping relationship, specifically: Extract the Stokes vector of the low-light polarization characteristic component in the dual-band polarization characteristic cube, obtain the phase angle distribution diagram of the Stokes vector, and simultaneously obtain the local contrast attenuation rate matrix of the low-light intensity component; Normalizing the phase angle distribution diagram to generate a polarization modulation phase characteristic diagram, and establishing a polarization phase-attenuation statistical coupling relationship by Pearson correlation analysis based on a local contrast attenuation rate matrix; It should be noted that the Stokes vector phase angle distribution map is normalized, converting the phase angle values ​​to a uniform range of 0-1 to generate a standardized polarization modulation phase signature map. This phase signature map is then pixel-wise matched with the local contrast decay matrix of the low-light-level image at the same spatial location, and the linear correlation between the two is calculated using the Pearson correlation coefficient. Finally, a quantitative statistical relationship between polarization phase and contrast decay is established based on the correlation coefficient, forming a mapping model that reflects the coupling strength between the two.

[0036] Based on the polarization phase-attenuation statistical coupling relationship, the nonlinear least squares method is used to fit the mapping curve between the polarization modulation phase characteristic map and the local contrast attenuation rate matrix, and the polarization phase-attenuation coupling coefficient matrix is ​​obtained by fitting. It should be noted that, based on the established statistical coupling relationship between polarization phase and attenuation, a nonlinear least-squares method is used to curve fit the polarization modulation phase eigenvalues ​​and the local contrast decay rate at the corresponding locations. Through iterative optimization to minimize the prediction error, characteristic parameters of the fitting curve (such as polynomial coefficients or exponential term weights) are extracted, and a coupling coefficient matrix is ​​constructed that reflects the nonlinear relationship between polarization phase and light intensity attenuation. Finally, this matrix is ​​dot-producted with the original phase characteristic map to generate a polarization phase-attenuation coupling coefficient matrix that accurately quantifies the influence of scattering.

[0037] The polarization phase-attenuation coupling coefficient matrix is ​​applied to the low-light intensity component, and the global attenuation base layer dominated by Mie scattering is separated through inverse mapping operation to obtain the residual texture component. It should be noted that the coupling coefficient matrix is ​​multiplied pixel by pixel by the low-light intensity component to obtain a predicted scattering attenuation distribution map. The predicted attenuation value is then subtracted from the original light intensity through an inverse operation to isolate the global attenuation base layer caused primarily by Mie scattering. Finally, this base layer is subtracted from the original image to obtain a residual texture component containing real scene details. This residual component preserves the features of objects unaffected by scattering while reducing the contrast degradation caused by fog.

[0038] Anisotropic guided filtering is performed on the residual texture components. The polarization degree component of the Stokes vector is used as the edge constraint weight to suppress noise and preserve the topological structure of the scattering-degraded texture. The degraded texture feature map modulated by Mie scattering is output. The degraded texture feature map is differentially calculated with the global attenuation base layer to verify the linear independence of the light intensity attenuation gradient and the polarization modulation phase angle, and finally the scattering modulation domain of the degraded texture feature map is calibrated.

[0039] It should be noted that this method effectively eliminates the confusion between environmental noise and true attenuation, utilizes the physical correlation between polarization characteristics and light attenuation to extract pure scattering degradation features, and at the same time establishes a quantifiable scattering modulation domain for subsequent image restoration, thereby improving the reliability of low-light-level image feature extraction in dense fog environments.

[0040] Preferably, the step 106 is specifically: According to the optical path accumulation of each pixel in the optical path difference mapping relationship matrix, the transmittance attenuation coefficient of the polarization channel corresponding to the low-light degradation texture feature map is determined, and a transmittance attenuation coefficient distribution map is generated; It should be noted that the optical path cumulative value of each pixel in the optical path difference mapping relationship matrix is ​​read and converted into the corresponding polarization channel transmittance attenuation coefficient based on a pre-calibrated optical path-to-transmittance conversion curve (such as the exponential attenuation model of the Lambert-Beer law). The attenuation coefficients of all pixels are then arranged according to their spatial position in the image to generate a transmittance attenuation coefficient distribution map with the same size as the original image, where each pixel value represents the degree of light intensity attenuation due to fog scattering at that location.

[0041] Combined with the polarization phase angle distribution of the Stokes vector, a nonlinear coupling relationship between the transmittance attenuation coefficient and the polarization phase angle is established to generate the polarization-transmittance coupling tensor. It is important to note that the transmittance attenuation coefficient distribution map is spatially aligned with the polarization phase angle distribution map of the Stokes vector to ensure a one-to-one correspondence between the attenuation coefficient and phase angle at each pixel. A quantitative relationship model is established between the two through nonlinear regression analysis (such as polynomial fitting or neural network modeling). Finally, the established phase-attenuation relationship is parameterized to generate a three-dimensional tensor structure containing the coupling coefficient at each pixel. This structure also contains the nonlinear response characteristics of the attenuation coefficient as the polarization phase changes.

[0042] The infrared radiation intensity component in the dual-band polarization characteristic cube is used to extract the penetration characteristic curve of thermal radiation in foggy medium. By calibrating the inverse proportional relationship between infrared thermal radiation intensity and fog concentration, a thermal radiation penetration compensation coefficient matrix is ​​constructed. It should be noted that infrared radiation intensity components are extracted from the dual-band polarization feature cube, and the corresponding thermal radiation intensity values ​​for different fog concentration areas (e.g., low concentration areas of 0.1-0.3 g / m³ and high concentration areas of 0.3-0.5 g / m³) are analyzed. A model is then established that establishes an inverse relationship between infrared radiation intensity and fog concentration (radiation intensity = baseline value / (1 + k × concentration)), where k is the dielectric characteristic coefficient. Finally, based on this model, the fog concentration value at each pixel is converted into a corresponding thermal radiation penetration compensation coefficient (ranging from 0.4 to 1.0). This compensation coefficient matrix, the same size as the image, is generated for subsequent joint correction processing.

[0043] Performing a tensor dot multiplication operation on the polarization-transmittance coupling tensor and the thermal radiation penetration compensation coefficient matrix to obtain a joint compensation operator that simultaneously integrates the polarization modulation characteristics and the thermal radiation penetration characteristics; The joint compensation operator is applied to the low-light degradation texture feature map. The transmittance attenuation coefficient is corrected pixel by pixel to eliminate the polarization channel energy attenuation caused by Mie scattering, and the intermediate low-light texture after transmittance correction is output. It should be noted that the joint compensation operator is multiplied pixel by pixel with the low-light degradation texture feature map. Using the transmittance correction coefficient stored in the compensation operator, the attenuation of each pixel affected by Mie scattering is specifically compensated. Local contrast adjustment is then performed on the compensated image to restore detail lost due to scattering. Finally, a transmittance-corrected intermediate low-light texture image is output.

[0044] The polarization component of the Stokes vector is used to enhance the edge of the intermediate low-light texture, and the local contrast lost due to scattering is restored by polarization weighting to generate an enhanced low-light feature map with fog scattering invariance.

[0045] In one specific embodiment of the present invention, a transmittance attenuation coefficient (ranging from 0.3 to 0.9) is determined based on the optical path accumulation at each pixel (0.1-1.2 optical path units) for a calibrated optical path difference mapping matrix (256×256 quantization levels), generating an attenuation coefficient distribution map. A polarization-transmittance coupling tensor is constructed using the Stokes vector phase angle (0-180°), while a thermal radiation compensation matrix (with a transmittance of 0.4-0.8) is constructed using infrared radiation intensity (8-14μm band). A joint compensation operator is generated by multiplying these two tensors, applying pixel-by-pixel correction to the 1280×1024 pixel low-light degraded texture, thereby enhancing the contrast of key features such as lane markings. Finally, edge enhancement is performed using the polarization degree component, resulting in a scattering-invariant enhanced image. This addresses the detail loss inherent in traditional methods in dense fog (fog concentration > 0.3g / m³).

[0046] It should be noted that this method can improve the texture clarity and feature fidelity of low-light images in dense fog environments, so that the enhanced images have stable scattering invariant characteristics.

[0047] Preferably, the step 108 is specifically: Perform Laplace operator convolution on the fog concentration gradient tensor, calculate the second-order spatial derivative at each pixel position, and generate a Laplace response map of the fog concentration; It should be noted that a two-dimensional convolution operation is performed on the fog concentration gradient tensor using a 3×3 Laplace kernel, calculating the second-order spatial derivative of each pixel within its eight-neighborhood. The convolution result is normalized to generate a Laplace response map that reflects the characteristics of sudden changes in fog concentration. Positive values ​​indicate the fog edge where concentration increases rapidly, negative values ​​correspond to the transition zone where concentration decreases, and zero values ​​represent areas of uniform concentration distribution. Finally, significant extreme points (Laplace response values ​​≥ +0.15) in the response map are extracted to provide boundary features for subsequent concentration zoning.

[0048] Based on the local extreme value distribution characteristics of the Laplace response graph, the mutation boundary of the fog concentration change is extracted to form the concentration partition boundary; Performing morphological closing operation smoothing on the concentration partition boundary to eliminate holes caused by noise, and combining the spatial consistency constraint of the infrared polarization component to correct the mis-segmented area caused by thermal radiation interference, and generate a dynamic partition mask; According to the high and low concentration areas marked in the dynamic partition mask, the pixels of the monitoring scene are classified into low concentration areas and high concentration areas; the low concentration area is the area with a mask value of 0, and the high concentration area is the area with a mask value of 1.

[0049] In one specific embodiment of the present invention, in a highway fog monitoring scenario (e.g., visibility classification: low concentration >100m, high concentration <50m), an optimized fog concentration gradient tensor (resolution 1280×1024) is Laplacian convolved (3×3 kernel) to generate a response map (extreme threshold ±0.15). Local maxima / minima (spacing >30 pixels) are detected to extract concentration abrupt change boundaries. After morphological closing (circular structuring element radius 5 pixels) and infrared polarization constraint (threshold 0.25-0.35), a dynamic partitioning mask is generated. The scene is ultimately divided into low concentration areas (mask value 0, fog concentration <0.2g / m³) and high concentration areas (mask value 1, concentration ≥0.2g / m³), achieving precise segmentation of lanes (low concentration) and fog clusters (high concentration).

[0050] Preferably, the step 110 is specifically: In low-concentration areas, the fog concentration gradient tensor is extracted to enhance the low-light feature map, and the thermal radiation profile gradient field is parsed from the infrared thermal radiation intensity. The fog concentration gradient tensor and the thermal radiation profile gradient field are subjected to pixel-by-pixel gradient structure similarity measurement to generate a gradient fidelity weight map. Based on the gradient fidelity weight map, the low-frequency component (0-0.5πrad / m spatial frequency) of the enhanced low-light feature map and the high-frequency component (0.5π-2πrad / m) of the infrared thermal radiation intensity are adaptively weighted fused to obtain a primary fusion feature; wherein the weight range is 0.3 (focusing on infrared details) to 0.7 (focusing on the low-light base); The local anisotropy coefficient of the fog concentration gradient tensor is used to rebalance the gradient field of the primary fusion features, eliminate the radiation distortion at the fusion boundary, and output the fusion result of the low concentration area.

[0051] In high-concentration areas, the spatial distribution matrix of the infrared polarization vector direction angle is calculated, and the polarization phase angle characteristics of the Stokes component of the enhanced low-light feature map are extracted. Polarization state covariance analysis is performed on the spatial distribution matrix and the polarization phase angle characteristics to establish a polarization direction-phase coupling tensor. The Stokes component of the enhanced low-light feature map refers to the set of characteristic parameters including polarization intensity, linear polarization degree, and polarization angle obtained by performing Stokes vector calculation and scattering correction on the original low-light polarization image. These components collectively represent the polarization characteristic distribution of the target scene after fog scattering compensation. Performing Stokes space projection transformation on the infrared polarization direction angle by using the coupling tensor to generate a recombined polarization basis; It should be noted that the infrared polarization angle data is input into the coupling tensor (which stores the mapping relationship between the infrared and low-light polarization states). Tensor multiplication then projects the infrared polarization angle into Stokes vector space (three-dimensional coordinates). The linear polarization component is then re-determined in Stokes space based on the projection result, generating a reconstructed polarization basis (containing the new polarization angle and polarization degree parameters) compatible with dual-band polarization characteristics. The final output is reconstructed polarization basis data that simultaneously expresses the fusion characteristics of the infrared polarization direction and the low-light polarization phase.

[0052] The Laplace norm response value of the fog concentration gradient tensor is used to dynamically adjust the weight ratio of the reconstructed polarization basis and the enhanced low-light feature map, and the tensor product operation of the polarization channel is performed to obtain the polarization reconstructed feature in the high-concentration area. It is important to note that the Laplace norm response of the fog concentration gradient tensor (reflecting the intensity of concentration mutations) is obtained and normalized to a dynamic weight coefficient ranging from 0 to 1 (larger values ​​indicate denser fog). This coefficient is used as the fusion weight for the reconstructed polarization basis (weight range 0.2-0.8). Simultaneously, the weight for enhancing the low-light signature map is set to a complementary value (1-weight coefficient). Finally, a channel-level tensor product operation (i.e., weighted product fusion of each polarization channel) is performed on the reconstructed polarization basis and the low-light signature map to generate a polarization reconstructed signature for high-concentration areas that simultaneously preserves infrared polarization direction information and low-light texture details. The dense fog core (weight > 0.6) prioritizes polarization penetration, while the edge transition zone (weight < 0.4) enhances low-light detail.

[0053] Finally, the fusion results of the low-concentration area and the polarization reconstruction features of the high-concentration area are spatially spliced ​​through a dynamic partition mask, and the final night vision imaging image is generated after a smooth transition using bilinear interpolation.

[0054] Among them, the "infrared thermal radiation intensity" used in the fusion of low-concentration areas corresponds to the corrected infrared radiation intensity (the polarization information has been stripped away and the thermal radiation scalar value has been retained); the "infrared polarization vector direction angle" used in the high-concentration areas is the vector information calculated from the original infrared polarization data (reorganized with the Stokes component).

[0055] In a specific embodiment of the present invention, in a foggy night monitoring scenario on a highway, for low-concentration areas (fog concentration <0.2g / m³) marked by a dynamic partitioning mask, an adaptive weight map is generated by calculating the structural similarity between the low-light feature map gradient tensor (gradient amplitude 15-30dB) and the infrared thermal radiation gradient field (temperature resolution 0.05K), thereby achieving faithful fusion of lane markings (low-light low frequency) and vehicle heat sources (high-infrared frequency). For high-concentration areas (≥0.2g / m³), a coupling tensor is constructed using the infrared polarization direction angle and the low-light Stokes phase angle (0-180°), and the polarization basis is dynamically weighted and recombined using the Laplace norm (response value 0.1-0.3) to enhance obstacle recognition in fog. Finally, after bilinear interpolation and splicing, a 1280×1024 pixel fused image is output.

[0056] It should be noted that through the partition fusion strategy, the gradient structure similarity measurement is combined in the low-concentration area to achieve the complementary advantages of low-light and infrared features, retaining the detailed texture and thermal radiation profile; in the high-concentration area, polarization state reorganization is used to fully explore the penetration ability of polarization information on scattering, and improve the imaging quality in dense fog areas; finally, through dynamic weight fusion and smooth stitching, a night vision image with high definition, complete thermal radiation characteristics and polarization enhancement effect is generated, which effectively solves the fusion distortion problem of traditional methods in areas with sudden changes in fog concentration.

[0057] During actual operation, the night vision imaging method further includes the following steps: Deploy orthogonal three-axis fluxgate sensors at the imaging equipment installation point. Based on the low-light image acquisition frame rate, they synchronously capture the geomagnetic field vector intensity and deflection time series data to generate a dynamic geomagnetic disturbance baseline. It is important to note that an orthogonal three-axis fluxgate sensor is fixed next to the imaging device, strictly synchronized with the low-light camera's acquisition frame rate (e.g., 30 fps), recording real-time X / Y / Z geomagnetic field intensity and declination data. Timestamp alignment is used to match the geomagnetic data with the image frame sequence. A sliding window average (1-second window width) is used to eliminate transient noise, generating a dynamic geomagnetic disturbance baseline dataset with a temporal resolution of 33 ms. This dataset includes magnetic field vector intensity curves and declination fluctuation time series.

[0058] The real-time phase angle matrix of the Stokes vector in the dual-band polarization feature cube is extracted and spatially and temporally aligned with the deflection component of the dynamic geomagnetic disturbance baseline. A phase-deflection coupling response function is established using a long short-term memory network. Applying the phase-bias coupling response function to the Stokes phase angle matrix, the polarization state drift component caused by electromagnetic interference is calculated and the polarization drift vector field is constructed. It should be noted that the pre-trained phase-deviation coupled response function (LSTM network model) is used to calculate the Stokes phase angle matrix frame by frame, predicting the intrinsic phase angle of each pixel in the absence of magnetic field interference. The predicted value is then subtracted from the actual measured phase angle to obtain the polarization state drift caused by electromagnetic interference. The drift values ​​of all pixels are arranged according to their spatial position in the image to construct a two-dimensional polarization drift vector field (containing magnitude and direction information).

[0059] The intrinsic polarization characteristics of the fog medium are extracted from the spatial distribution of the infrared polarization degree component. Combined with the spectrum characteristics of the dynamic geomagnetic disturbance baseline, the polarization eigenaxis reference system is separated through Wiener filtering. Based on the polarization eigenaxis reference system, the polarization drift vector field is decomposed into an antisymmetric tensor, the non-conservative component that is strongly correlated with the geomagnetic declination is extracted, and the geomagnetic drift compensation field is output; The geomagnetic drift compensation field is subtracted from the original Stokes vector, and the orthogonal projection normalization is performed using the polarization eigenaxis reference system to reconstruct the diamagnetic biased Stokes vector to eliminate electromagnetic interference.

[0060] For example, in foggy night monitoring scenarios near high-voltage transmission lines on highways, a three-axis fluxgate sensor and a low-light-level camera are used to synchronize data acquisition, recording real-time fluctuations in the geomagnetic field declination. A pre-trained LSTM network analyzes the correlation between the Stokes phase angle (1280×1024 pixels) and the magnetic declination, resolving the polarization drift field caused by electromagnetic interference. The intrinsic polarization axis of the fog is extracted by combining the infrared polarization degree. After Wiener filtering and antisymmetric tensor decomposition, the geomagnetic compensation field is generated, ultimately outputting the antimagnetic deflected Stokes vector, improving the reliability of polarization imaging in electromagnetic interference environments.

[0061] It should be noted that this embodiment solves the problem of polarization imaging drift caused by geomagnetic disturbances in complex electromagnetic environments (such as near high-voltage lines on highways) through geomagnetic disturbance monitoring and polarization state drift modeling, so that the imaging results can still maintain the physical authenticity of the polarization information under strong electromagnetic interference, thereby improving the reliability and environmental adaptability of target polarization feature recognition in foggy night road monitoring.

[0062] like Figure 3 As shown, the second aspect of the present invention discloses a night vision imaging system 8 for fusing low-light level and infrared images, the night vision imaging system comprising a memory 60 and a processor 80, the memory 60 storing a night vision imaging method program for fusing low-light level and infrared images, and when the night vision imaging method program for fusing low-light level and infrared images is executed by the processor 80, any one of the steps of the night vision imaging method for fusing low-light level and infrared images is implemented.

[0063] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0064] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0065] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0066] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0067] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods of the various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0068] The above are only specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A night vision imaging method for fusing low-light-level and infrared images, characterized in that: The following steps are involved: S102: In a foggy night road environment, synchronously collect low-light polarization images and long-wave infrared polarization images to construct a spatiotemporally aligned dual-band polarization feature cube; S104: Extracting the fog concentration gradient tensor of the monitoring scene and separating the degraded texture feature map modulated by Mie scattering based on the dual-band polarization feature cube, establishing an optical path difference mapping relationship between the low-light degraded texture and the fog concentration field, and generating an optical path difference mapping relationship matrix; S106: generating a joint compensation operator based on the optical path difference mapping relationship matrix, performing polarization channel transmittance correction on the low-light degradation texture in combination with the compensation operator, and outputting an enhanced low-light feature map with fog scattering invariance; S108: Analyze the fog concentration gradient tensor and dynamically divide the fog concentration area into a low concentration area and a high concentration area according to the Laplace norm response value of the fog concentration gradient tensor; S110: In low-concentration areas, the enhanced low-light feature map is fused with the infrared thermal radiation intensity using a gradient tensor with fidelity. In high-concentration areas, the infrared polarization vector direction angle and the Stokes component of the enhanced low-light feature map are recombined using a polarization state tensor to generate a night vision image.

2. The night vision imaging method for fusing low-light level and infrared images according to claim 1, characterized in that: The 102 is specifically: Synchronously collect light intensity data in four polarization directions in the low-light band and radiation intensity data in two orthogonal polarization directions in the long-wave infrared band, generating a low-light polarization intensity image group and a long-wave infrared polarization radiation intensity image group respectively; Perform Stokes vector calculation on the low-light polarization image group to obtain the low-light intensity component and polarization characteristic component; perform thermal radiation polarization calculation on the long-wave infrared polarization image group to obtain the infrared radiation intensity component and polarization degree component; Based on the spatial gradient characteristics of the low-light intensity component and the thermal radiation profile characteristics of the infrared radiation intensity component, a spatial registration matrix is ​​generated through affine transformation to achieve sub-pixel alignment of the two band images. The polarization correlation between the low-light polarization characteristic component and the infrared polarization degree component is used to determine the polarization correlation between the bands, and a fusion weight matrix is ​​generated based on the polarization correlation. Applying the spatial registration matrix to geometrically correct the infrared radiation intensity component to obtain the corrected infrared radiation intensity; Perform polarization direction compensation on the infrared polarization component to eliminate the polarization angle deviation caused by parallax and output the corrected infrared polarization degree; The corrected infrared radiation intensity and the low-light intensity component are fused according to the fusion weight matrix to generate a fused intensity feature; the corrected infrared polarization degree and the low-light polarization characteristic component are recombined in polarization state to generate a fused polarization feature; The fused light intensity features, corrected infrared radiation intensity and fused polarization features are superimposed according to the spatial dimension to construct a six-channel dual-band polarization feature cube containing light intensity, radiation and polarization information.

3. The night vision imaging method for fusing low-light level and infrared images according to claim 1, characterized in that: The 104 is specifically: Perform pixel-by-pixel difference calculation on the low-light intensity component and infrared radiation intensity component in the dual-band polarization characteristic cube to obtain the dual-band differential intensity parameter. Use the Stokes parameters of the low-light polarization characteristic component to calculate the polarization difference ratio and obtain the polarization modulation difference parameter. An initial fog concentration distribution map is generated based on the polarization modulation difference parameter and the dual-band differential intensity parameter; an anisotropic diffusion filter is performed on the initial fog concentration distribution map, and the thermal radiation polarization constraint condition of the infrared polarization component is combined to suppress noise and retain the fog edge structure, and an optimized fog concentration gradient tensor is output; By using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation, a mapping relationship between polarization modulation and light intensity degradation is established. Through this mapping relationship, a degraded texture feature map modulated by Mie scattering is decoupled from the low-light intensity component. A three-dimensional distribution model of the fog concentration field is constructed based on the optimized fog concentration gradient tensor. At the same time, the optical path accumulation of the low-light degraded texture in the fog concentration field is determined according to the local contrast attenuation rate of the degraded texture feature map. The mapping relationship between the grayscale attenuation gradient of the degraded texture feature map and the optical path accumulation in the fog concentration field is established through the nonlinear regression method, and a calibrated optical path difference mapping relationship matrix is ​​generated. Each element of the matrix represents the degree of scattering degradation of the low-light texture under a specific fog concentration.

4. The night vision imaging method for fusing low-light level and infrared images according to claim 3, characterized in that: By using the statistical correlation between the Stokes vector phase angle of the low-light polarization characteristic component and the light intensity attenuation, a mapping relationship between polarization modulation and light intensity degradation is established. Through this mapping relationship, the degraded texture feature map modulated by Mie scattering is decoupled from the low-light intensity component. Specifically, Extract the Stokes vector of the low-light polarization characteristic component in the dual-band polarization characteristic cube, obtain the phase angle distribution diagram of the Stokes vector, and simultaneously obtain the local contrast attenuation rate matrix of the low-light intensity component; Normalizing the phase angle distribution diagram to generate a polarization modulation phase characteristic diagram, and establishing a polarization phase-attenuation statistical coupling relationship by Pearson correlation analysis based on a local contrast attenuation rate matrix; Based on the polarization phase-attenuation statistical coupling relationship, the nonlinear least squares method is used to fit the mapping curve between the polarization modulation phase characteristic map and the local contrast attenuation rate matrix, and the polarization phase-attenuation coupling coefficient matrix is ​​obtained by fitting. The polarization phase-attenuation coupling coefficient matrix is ​​applied to the low-light intensity component, and the global attenuation base layer dominated by Mie scattering is separated through inverse mapping operation to obtain the residual texture component. Anisotropic guided filtering is performed on the residual texture components. The polarization degree component of the Stokes vector is used as the edge constraint weight to suppress noise and preserve the topological structure of the scattering-degraded texture. The degraded texture feature map modulated by Mie scattering is output. The degraded texture feature map is differentially calculated with the global attenuation base layer to verify the linear independence of the light intensity attenuation gradient and the polarization modulation phase angle, and finally the scattering modulation domain of the degraded texture feature map is calibrated.

5. The night vision imaging method for fusing low-light level and infrared images according to claim 1, characterized in that: The 106 is specifically: According to the optical path accumulation of each pixel in the optical path difference mapping relationship matrix, the transmittance attenuation coefficient of the polarization channel corresponding to the low-light degradation texture feature map is determined, and a transmittance attenuation coefficient distribution map is generated; Combined with the polarization phase angle distribution of the Stokes vector, a nonlinear coupling relationship between the transmittance attenuation coefficient and the polarization phase angle is established to generate the polarization-transmittance coupling tensor. The infrared radiation intensity component in the dual-band polarization characteristic cube is used to extract the penetration characteristic curve of thermal radiation in foggy medium. By calibrating the inverse proportional relationship between infrared thermal radiation intensity and fog concentration, a thermal radiation penetration compensation coefficient matrix is ​​constructed. Performing a tensor dot multiplication operation on the polarization-transmittance coupling tensor and the thermal radiation penetration compensation coefficient matrix to obtain a joint compensation operator that simultaneously integrates the polarization modulation characteristics and the thermal radiation penetration characteristics; The joint compensation operator is applied to the low-light degradation texture feature map. The transmittance attenuation coefficient is corrected pixel by pixel to eliminate the polarization channel energy attenuation caused by Mie scattering, and the intermediate low-light texture after transmittance correction is output. The polarization component of the Stokes vector is used to enhance the edge of the intermediate low-light texture, and the local contrast lost due to scattering is restored by polarization weighting to generate an enhanced low-light feature map with fog scattering invariance.

6. The night vision imaging method for fusing low-light level and infrared images according to claim 1, characterized in that: The 108 is specifically: Perform Laplace operator convolution on the fog concentration gradient tensor, calculate the second-order spatial derivative at each pixel position, and generate a Laplace response map of the fog concentration; Based on the local extreme value distribution characteristics of the Laplace response graph, the mutation boundary of the fog concentration change is extracted to form the concentration partition boundary; Performing morphological closing operation smoothing on the concentration partition boundary to eliminate holes caused by noise, and combining the spatial consistency constraint of the infrared polarization component to correct the mis-segmented area caused by thermal radiation interference, and generate a dynamic partition mask; According to the high and low concentration areas marked in the dynamic partition mask, the pixels of the monitoring scene are classified into low concentration areas and high concentration areas; the low concentration area is the area with a mask value of 0, and the high concentration area is the area with a mask value of 1.

7. The night vision imaging method for fusing low-light level and infrared images according to claim 1, characterized in that: The 110 is specifically: In low-concentration areas, the fog concentration gradient tensor is extracted to enhance the low-light feature map, and the thermal radiation profile gradient field is parsed from the infrared thermal radiation intensity. The fog concentration gradient tensor and the thermal radiation profile gradient field are subjected to pixel-by-pixel gradient structure similarity measurement to generate a gradient fidelity weight map. Based on the gradient fidelity weight map, the low-frequency component of the enhanced low-light feature map and the high-frequency component of the infrared thermal radiation intensity are adaptively weighted and fused to obtain a primary fusion feature; The local anisotropy coefficient of the fog concentration gradient tensor is used to rebalance the gradient field of the primary fusion features, eliminate the radiation distortion at the fusion boundary, and output the fusion result of the low concentration area.

8. The night vision imaging method for fusing low-light level and infrared images according to claim 7, characterized in that: The 110 further includes: In high-concentration areas, the spatial distribution matrix of the infrared polarization vector direction angle is calculated, and the polarization phase angle characteristics in the Stokes component of the enhanced low-light feature map are extracted. The spatial distribution matrix and the polarization phase angle characteristics are subjected to polarization state covariance analysis to establish the polarization direction-phase coupling tensor. Performing Stokes space projection transformation on the infrared polarization direction angle by using the coupling tensor to generate a recombined polarization basis; The Laplace norm response value of the fog concentration gradient tensor is used to dynamically adjust the weight ratio of the reconstructed polarization basis and the enhanced low-light feature map, and the tensor product operation of the polarization channel is performed to obtain the polarization reconstructed feature in the high-concentration area. Finally, the fusion results of the low-concentration area and the polarization reconstruction features of the high-concentration area are spatially spliced ​​through a dynamic partition mask, and the final night vision imaging image is generated after a smooth transition using bilinear interpolation.

9. A night vision imaging system that fuses low-light and infrared images, characterized in that: The night vision imaging system includes a memory and a processor, wherein the memory stores a night vision imaging method program for fusing low-light level and infrared images. When the night vision imaging method program for fusing low-light level and infrared images is executed by the processor, the steps of the night vision imaging method for fusing low-light level and infrared images as described in any one of claims 1 to 8 are implemented.

Images