A method for testing multi-spectral polarized light transmission characteristics based on passive imaging

Abstract

The present invention discloses a method for testing the transmission characteristics of multi-spectral polarized light based on passive imaging, which relates to the field of optical engineering technology. The method comprises: utilizing a multi-dimensional image fusion method to perform spatiotemporal alignment and weighted fusion of a multi-spectral polarized image dataset to generate a polarization coupling matrix; inputting the polarization coupling matrix into a non-uniform polarized light transmission model; extracting depolarization features using a multi-scale convolution layer; analyzing polarization correlations between spectral bands using an attention mechanism layer; outputting a multi-spectral polarized light transmission characteristic test scheme; inputting the multi-spectral polarized light transmission characteristic test scheme into an optical simulation engine, and driving a LabVIEW unit to perform polarized scattered light imaging simulation in a smoky environment. The present invention utilizes advanced multi-dimensional image fusion technology and a non-uniform polarized light transmission model to improve the accuracy of image data fusion and imaging simulation, effectively resolving the problem of transmission characteristic inversion errors caused by registration deviations, and significantly improving the reliability and accuracy of transmission characteristic testing in complex environments.

Description

Technical Field

[0001] The present invention relates to the field of optical engineering technology, and in particular to a method for testing multi-spectral polarized light transmission characteristics based on passive imaging. Background Art

[0002] With the development of optical technology, multi-spectral polarization imaging has been widely used in fields such as environmental monitoring, remote sensing, and biomedicine. By emitting polarized light of specific wavelengths and receiving its reflected or scattered signals, multi-spectral polarization imaging can not only analyze the polarization characteristics of target objects for accurate identification and classification, but also accurately characterize and identify the polarization transmission characteristics of target objects through quantitative analysis of transmission characteristic parameters such as polarization degree and depolarization rate.

[0003] Existing technologies have two major shortcomings: First, the collaborative processing capabilities of multi-spectral polarization data are insufficient, and existing fusion algorithms cannot simultaneously ensure the accuracy of spatiotemporal alignment and the integrity of feature preservation, resulting in large errors in the inversion of transmission characteristics tests; second, the adaptability of existing polarization transmission models is limited, and they cannot accurately characterize the depolarization gradient changes in inhomogeneous media such as smoke. In addition, there is a lack of cross-spectral correlation modeling mechanisms, which leads to deviations in the prediction of multi-spectral transmission characteristics, seriously restricting the test accuracy and application effect of polarization transmission characteristics in complex environments. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a multi-spectral polarized light transmission characteristic testing method based on passive imaging to solve the problems of large transmission characteristic inversion error and insufficient testing accuracy of existing transmission characteristics.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for testing the transmission characteristics of multi-spectral polarized light based on passive imaging, which comprises: using a multi-dimensional image fusion method to perform spatiotemporal alignment and weighted fusion on a multi-spectral polarized image data set to generate a polarization coupling matrix;

[0008] The polarization coupling matrix is ​​input into the non-uniform polarized light transmission model. The multi-scale convolution layer extracts depolarization features. The attention mechanism layer analyzes the polarization correlation between spectrum bands and outputs a multi-spectral polarized light transmission characteristic test solution.

[0009] The multi-spectral polarized light transmission characteristic test scheme is input into the optical simulation engine, driving the LabVIEW unit to perform polarized scattered light imaging simulation in the smoke environment and obtain the simulated polarization transmission characteristic distribution cloud map;

[0010] The simulated polarization transmission characteristic distribution cloud map is compared with the multi-spectral polarization image dataset, and the residuals between the actual polarization values ​​and the simulated polarization values ​​are analyzed to obtain the adjustment calibration parameters and optimize the non-uniform polarized light transmission model.

[0011] As a preferred solution of the multi-spectral polarization light transmission characteristics testing method based on passive imaging described in the present invention, the multi-spectral polarization image dataset includes multi-spectral polarization image data, environmental parameters and device metadata.

[0012] As a preferred solution of the multi-spectral polarization light transmission characteristic testing method based on passive imaging of the present invention, the generating of the polarization coupling matrix specifically includes the following steps:

[0013] Wavelet decomposition is used to suppress noise in multi-spectral polarization image datasets, and SIFT algorithm and linear interpolation are applied to perform cross-spectral spatial registration and temporal synchronization to obtain multi-spectral polarization intensity tensors.

[0014] A multi-dimensional image fusion method is used to assign dynamic weights to the multi-spectral polarization intensity tensor, which is coupled with environmental parameters. The polarization coupling matrix is ​​output through tensor reorganization.

[0015] As a preferred solution of the multi-spectral polarized light transmission characteristic testing method based on passive imaging of the present invention, the specific construction process of the non-uniform polarized light transmission model is as follows:

[0016] Residual connections and gating mechanisms are used to perform hierarchical parameter fusion of multi-scale convolutional layers and attention mechanism layers to construct a non-uniform polarization light transmission model.

[0017] As a preferred solution of the multi-spectral polarized light transmission characteristic testing method based on passive imaging of the present invention, the output multi-spectral polarized light transmission characteristic testing solution specifically includes the following steps:

[0018] The multi-scale convolution layer uses multi-branch convolution kernels to perform cross-scale debiased feature extraction and residual fusion to generate debiased feature tensors;

[0019] The attention mechanism layer dynamically weights the polarization correlation through a gating mechanism to obtain the spectral segment attention weight matrix;

[0020] The depolarization feature tensor and the spectral attention weight matrix are spliced ​​in the feature channel to form a polarization coupling feature map;

[0021] ReLU activation and batch normalization are used to perform feature channel compression and task adaptation mapping on the polarization coupling feature map to generate a spectral polarization imaging test solution.

[0022] As a preferred solution of the multi-spectral polarization transmission characteristic testing method based on passive imaging of the present invention, the step of obtaining a simulated polarization transmission characteristic distribution cloud map specifically includes the following steps:

[0023] The Stokes parameters and polarization parameters in the multi-spectral polarization light transmission characteristics test scheme are extracted using the amplitude-division polarization modulation method and input into the optical simulation engine through the Matlab-Stokes interface;

[0024] The optical simulation engine drives the PID controller of the LabVIEW unit to adjust the smoke concentration gradient and simulate polarized scattered light imaging. It also uses the four-quadrant difference algorithm to calculate the depolarization rate and obtain the polarization distribution parameters across the spectral band.

[0025] Collect real-time light intensity distribution data, perform three-dimensional spatial sampling and fusion on the real-time light intensity distribution data and cross-spectral polarization distribution parameters, and perform polarization characteristic mapping to form a simulated polarization transmission characteristic distribution cloud map.

[0026] As a preferred solution of the multi-spectral polarized light transmission characteristic testing method based on passive imaging of the present invention, the step of obtaining and adjusting the calibration parameters specifically includes the following steps:

[0027] The three-dimensional photon path is traced on the simulated polarization transmission characteristic distribution cloud map using the three-dimensional ray tracing method to obtain the simulated polarization value;

[0028] Perform four-channel light intensity interpolation and frequency domain demodulation on the multi-spectral polarization image dataset to obtain the actual polarization value;

[0029] The actual polarization value and the simulated polarization value are differentially compared and calculated to obtain the residual distribution parameters; the momentum gradient descent algorithm is used to perform gradient iterative correction on the residual distribution parameters to generate adjustment calibration parameters.

[0030] As a preferred solution of the multi-spectral polarized light transmission characteristic testing method based on passive imaging of the present invention, the optimization of the non-uniform polarized light transmission model specifically includes the following steps:

[0031] The adjusted calibration parameters are input into the inhomogeneous polarization light transmission model, and the Markov chain Monte Carlo method is used to perform parameter inversion to obtain the posterior confidence interval of the parameters.

[0032] According to the posterior confidence interval of the parameters, the weight parameters of the non-uniform polarization light transmission model are incrementally updated, and the optimized non-uniform polarization light transmission model is output.

[0033] The present invention has the following beneficial effects: It utilizes advanced multi-dimensional image fusion technology to efficiently handle polarization correlations and noise interference between different spectral bands, improving the accuracy of image data and significantly reducing depolarization rate measurement errors in smoky environments. This effectively addresses the transmission characteristic inversion errors caused by registration bias in traditional methods. Furthermore, the use of a non-uniform polarized light transmission model and an optical simulation engine for precise simulation more realistically reflects the complex factors in the actual environment, significantly improving the reliability and accuracy of transmission characteristic testing in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 The flowchart of the multi-spectral polarization light transmission characteristics testing method based on passive imaging.

[0036] Figure 2 Flowchart of the polarization coupling matrix generation process.

[0037] Figure 3 Flowchart of the process for building a model for the transmission of inhomogeneously polarized light.

[0038] Figure 4 Flowchart of the optical simulation engine working process. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0040] 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. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0042] Reference Figure 1, is an embodiment of the present invention, which provides a method for testing multi-spectral polarized light transmission characteristics based on passive imaging, comprising the following steps:

[0043] S1, uses the multi-dimensional image fusion method to perform spatiotemporal alignment and weighted fusion of multi-spectral polarization image datasets to generate a polarization coupling matrix.

[0044] See also Figure 2 , the specific steps are as follows:

[0045] S1.1, collect a multi-spectral polarization image dataset, which includes multi-spectral polarization image data, environmental parameters, and device metadata;

[0046] Aim the multispectral polarization camera at the target object or scene, then set the camera parameters to adapt to different lighting conditions and target characteristics. Then start the multispectral polarization camera to continuously capture multi-spectral polarization images, and integrate them through image processing software (such as ENVI) to obtain multi-spectral polarization image data.

[0047] Place temperature and humidity sensors, barometers, anemometers, and smoke concentration sensors in the measurement environment. Then connect the environmental sensor group to the data acquisition and control center, set the sampling frequency and storage format, and finally start the environmental sensor group to monitor and record environmental parameters such as temperature, humidity, air pressure, wind speed, and smoke concentration in real time.

[0048] Connect the multispectral polarization camera and environmental sensor group to the device management software. Then, configure the basic information of each device in the device management software, such as model, serial number, firmware version, and calibration date. Then, enable the synchronous recording function in the device management software to record the device metadata during operation, including device status, operation logs, calibration data, etc.

[0049] S1.2, preprocessing the multi-spectral polarization image dataset. In the specific operation, the Daubechies wavelet basis is first used to perform multi-scale decomposition of the multi-spectral polarization image dataset to obtain high-frequency detail components; the high-frequency detail components are noise filtered to generate denoised polarization image data; the SIFT algorithm is used to perform block feature extraction on the denoised polarization image data to obtain the feature points of the polarization image of each spectral band, and the feature points of the polarization image of each spectral band are spatially aligned across spectral bands through geometric transformation to form a registered polarization image; the registered polarization image is sub-pixel resampled using the bilinear interpolation method, and dynamic time alignment is performed using cubic spline interpolation to obtain spatiotemporal continuous polarization intensity data; the spatiotemporal continuous polarization intensity data is fused with the timestamp information to form a multi-spectral polarization intensity tensor.

[0050] S1.3. Use a multi-dimensional image fusion method to perform weighted fusion on the multi-spectral polarization intensity tensors to generate a polarization coupling matrix. Specifically, the multi-spectral polarization intensity tensor is divided into multiple windows. Within each window, the noise level is quantified using the multi-dimensional image fusion method to obtain the signal-to-noise ratio of the multi-spectral polarization intensity tensor. Based on the signal-to-noise ratio of the multi-spectral polarization intensity tensor, a weighted algorithm is used to dynamically assign weights to the multi-spectral polarization intensity tensors, and a nonlinear activation function is used to perform normalization processing to form a fusion weight coefficient.

[0051] Next, the fusion weight coefficients are optimized and smoothed using Tikhonov regularization to ensure the smoothness of the weight distribution and avoid fusion artifacts caused by local parameter mutations. Based on the optimized fusion weight coefficients, element-by-element weighted multiplication is applied to fuse the multi-spectral data to ensure the effective preservation of the polarization characteristics of each spectral band and obtain weighted multi-spectral polarization intensity data.

[0052] The weighted multi-spectral polarization intensity data and environmental parameters are converted into a unified space through affine transformation, and multi-physical quantity coupling is performed through tensor recombination method. Principal component analysis is simultaneously applied to perform feature dimensionality reduction to generate a polarization coupling matrix.

[0053] In S2, the polarization coupling matrix is ​​input into the non-uniform polarized light transmission model. The multi-scale convolution layer extracts depolarization features. The attention mechanism layer analyzes the polarization correlation between spectral bands and outputs a multi-spectral polarized light transmission characteristic test solution.

[0054] The specific steps are as follows:

[0055] S2.1, see Figure 3 , build an inhomogeneous polarization light transmission model and train it. Specifically, in the PyTorch framework, call and initialize the multi-scale convolution layer and attention mechanism layer through nn.ModuleList: set the convolution kernel size to [1,3,5], the number of input channels to 3, the number of output channels to 64, and the stride to 1; batch normalization and LeakyReLU activation function are applied after the multi-scale convolution layer to enhance the image feature expression capability; then, residual connection is applied to call 1×1 convolution for channel matching to achieve dimensional alignment of image features; use the gating mechanism to dynamically assign weights to the multi-scale convolution layer and attention mechanism layer, use the fully connected layer to perform feature channel transformation, and simultaneously apply average pooling for hierarchical parameter fusion to complete the construction of the inhomogeneous polarization light transmission model;

[0056] Next, the non-uniform polarization light transmission model is trained. First, the polarization coupling matrix is ​​divided into a sample set, a training set, and a validation set in a ratio of 6:2:2. The sample set is normalized using Z-score standardization, and data augmentation is performed through random sampling to form enhanced training samples. On the training set, the Adam optimizer uses the loss function to backpropagate the augmented training samples and dynamically adjusts the learning rate to update the non-uniform polarization light transmission model parameters, generating optimized non-uniform polarization light transmission model parameters. On the validation set, when the optimized non-uniform polarization light transmission model parameters reach the convergence threshold, training is terminated, and the trained non-uniform polarization light transmission model is output synchronously using the torch.save function.

[0057] It should be noted that the convergence threshold is defined based on the rate of decrease of the loss function and its value range is [0.001-0.0001].

[0058] S2.2, generate polarization coupling feature map through non-uniform polarization light transmission model. In the specific operation, the polarization coupling matrix is ​​input into the non-uniform polarization light transmission model, and the multi-scale convolution layer calls three independent convolution branches in parallel through nn.ModuleList, and configures three different sizes of convolution kernels of 1×1, 3×3 and 5×5 for the convolution branches respectively; in the feature extraction stage, the input polarization coupling matrix is ​​sent to the three convolution branches at the same time, where 1×1 convolution is used to extract global polarization characteristics, 3×3 convolution captures local depolarization characteristics, and 5×5 convolution focuses on scattering characteristics; the outputs of the three branches are spliced ​​in the channel dimension through the torch.ca function to form cross-scale fusion features; then, using the residual connection mechanism, the number of channels of the original polarization coupling matrix is ​​adjusted to 192 using 1×1 convolution, and then added element-by-element with the cross-scale fusion features, which not only retains the original polarization information but also enhances the feature reuse capability, and finally generates the depolarization feature tensor;

[0059] The attention mechanism layer performs global average pooling on the polarization coupling matrix, compresses the spatial dimension, and generates a channel-level statistical feature vector; based on the channel-level statistical feature vector, a fully connected layer projection and Tanh activation transformation are performed through a gating mechanism to generate channel attention features; a 1×1 convolution kernel is used to perform cross-channel feature interaction on the channel attention features, and a batch normalization layer is used to normalize the feature distribution of the channel attention features to obtain cross-channel interaction features; a residual connection is applied to stabilize the gradient of the cross-channel interaction features to obtain polarization correlation representation; spatial convolution is used to dynamically weight the polarization correlation representation to generate spatial enhancement features; and the ReLU activation function is used to perform nonlinear transformation on the spatial enhancement features to obtain nonlinear enhancement features; the nonlinear enhancement features are simultaneously normalized through the Sigmoid function to generate a spectral attention weight matrix;

[0060] The depolarization feature tensor and the spectral band attention weight matrix are processed to make the spatial dimensions consistent through bilinear interpolation to ensure that the depolarization feature tensor and the spectral band attention weight matrix are fully matched; on the unified channel dimension, a 1×1 convolution kernel is used to unify the number of channels of the depolarization feature tensor and the spectral band attention weight matrix respectively to eliminate the dimension mismatch problem; then, a depthwise separable convolution is called to perform cross-modal interactive splicing of the depolarization feature tensor and the spectral band attention weight matrix, and a group normalization layer is used to stabilize the feature distribution and output the cross-modal fusion feature; subsequently, the cross-modal fusion feature is shuffled through the channels to enhance the feature fusion effect by disrupting the feature order, and a batch normalization layer and a LeakyReLU activation function are applied for nonlinear enhancement, and a 3×3 dilated convolution is used to expand the receptive field, and finally a polarization coupling feature map is output.

[0061] S2.3, perform feature channel compression and task adaptation mapping on the polarization coupling feature map to generate a spectral polarization imaging test plan. In the specific operation, the ReLU activation function is used to perform nonlinear feature enhancement on the polarization coupling feature map, and the feature distribution is standardized through batch normalization. The 1×1 convolution kernel is simultaneously applied to perform channel dimension compression to obtain the feature expression after channel compression; then, global average pooling is applied to perform spatial dimension aggregation on the feature expression after channel compression, and 3×3 deformable convolution is applied to perform geometric deformation to form a task-adaptive feature map; global average pooling is used to flatten the spatial features of the task-adaptive feature map to generate a task-specific vector representation, and nonlinear projection is performed on the task-specific vector representation. The gating unit is simultaneously applied to perform task adaptation mapping and output the classification probability distribution; the classification probability distribution is parameter-decoded through the parameter regression head to obtain the optimal polarization parameters and exposure time parameters, and structured packaging is performed to generate a spectral polarization imaging test plan.

[0062] S3, input the multi-spectral polarization light transmission characteristic test plan into the optical simulation engine, drive the LabVIEW unit to perform polarization scattered light imaging simulation in the smoke environment, and obtain the simulated polarization transmission characteristic distribution cloud map.

[0063] See also Figure 4 , the specific steps are as follows:

[0064] S3.1. Extract the Stokes parameters and polarization parameters from the multi-spectral polarization transmission characteristics test solution using the amplitude-split polarization modulation method, and input them into the optical simulation engine through the Matlab-Stokes interface. Specifically, perform multi-angle polarization sampling on the multi-spectral polarization transmission characteristics test solution using the amplitude-split polarization modulation method to obtain initial polarization intensity data. Perform intensity normalization on the initial polarization intensity data using a sliding window, and simultaneously extract the polarization parameters using bilinear interpolation.

[0065] At the same time, the initial polarization intensity data is converted into instructions through Matlab's serial communication toolbox (Serial Communication Toolbox) to obtain the wave plate control instructions; according to the wave plate control instructions, the stepper motor is driven to control the wave plate to perform four-step phase shift polarization modulation: the stepper motor controls the wave plate to rotate at four azimuth angles of 0°, 45°, 90° and 135°, and a high-speed CCD camera is used to perform multi-spectral image acquisition to obtain the original intensity image; the original intensity image is corrected for non-uniformity and bad pixels are repaired using Matlab's image processing tool (Image Processing Toolbox) to generate a calibrated intensity image; the Stokes parameter of the calibrated intensity image is solved through the inverse operation of the Mueller matrix to obtain the Stokes parameter of each pixel point. The specific mathematical formula is as follows,

[0066] ;

[0067] in represents the Stokes parameter, Indicates the linear polarization intensity of 0° direction after calibration; Indicates the intensity of linearly polarized light at 45° after calibration; Indicates the intensity of linearly polarized light at 90° after calibration; Indicates the linear polarization intensity of 135° direction after calibration; represents the first component of the Stokes parameter; represents the second component of the Stokes parameter; represents the third component of the Stokes parameter; represents the fourth component of the Stokes parameter;

[0068] The Matlab-Stokes interface is called through the Matlab script control unit to input the Stokes parameters and polarization parameters into the optical simulation engine.

[0069] S3.2: The optical simulation engine drives the PID controller to adjust the smoke concentration gradient and simulate polarized scattered light imaging. Specifically, the optical simulation engine first uses the smoke concentration emitter to set the initial smoke particle parameters (e.g., particle size distribution 0.1-10 μm, refractive index 1.33-1.55) and the initial concentration gradient value (e.g., initial concentration C0 = 100 mg / m³).

[0070] The initial parameters of smoke particles and the initial values ​​of the concentration gradient are standardized to obtain the smoke concentration parameters. The proportional coefficient of the PID controller amplifies the real-time error of the smoke concentration parameter, the integral coefficient eliminates the accumulated error of the historical smoke concentration parameter, and the differential coefficient predicts the rate of change of the smoke concentration parameter based on the change parameter, outputting an optimized control signal. Based on the optimized control signal, the smoke concentration gradient is dynamically adjusted through the regulating valve of the smoke concentration transmitter to form a stable concentration gradient distribution. The Polarization RayTrace component in the optical simulation engine is called to perform four-step phase-shift polarization imaging, and the scattered light field distribution is simulated through a virtual Mie scattering medium (such as a scattering volume) to obtain simulated polarized light intensity distribution data.

[0071] S3.3. Calculate the depolarization rate based on the simulated polarized light intensity distribution data using a four-quadrant difference algorithm to obtain cross-spectral polarization distribution parameters. Specifically, first import the simulated polarized light intensity distribution data into Matlab and use an image segmentation algorithm to divide the simulated polarized light intensity distribution data into four spectral bands (e.g., 450 nm, 550 nm, 650 nm, and 750 nm).

[0072] The depolarization rate is calculated by applying the four-quadrant difference algorithm in each spectral band. Specifically, the pixel traversal algorithm is applied to scan the spectral band pixel by pixel to obtain the target pixel. A 5×5 processing window is established with the target pixel as the center. The processing window is divided into four quadrants, and the depolarization rate formula is used to calculate the depolarization rate of each pixel in the quadrant. The specific mathematical formula is as follows;

[0073] ;

[0074] in, represents the depolarization rate; Indicates the vertical polarization difference value; represents the diagonal polarization difference value; Indicates total light intensity;

[0075] The weighted fusion method is used to fuse the depolarization rates in multiple spectral bands, and three-dimensional spatial interpolation is applied to align the cross-band data to generate the cross-spectral polarization distribution parameters.

[0076] S3.4 collects real-time light intensity distribution data, performs three-dimensional spatial sampling and fusion on the real-time light intensity distribution data and cross-spectral polarization distribution parameters, and performs polarization characteristic mapping to form a simulated polarization transmission characteristic distribution cloud map. In the specific operation, first use a high-speed polarization camera to collect real-time light intensity distribution data; then use the SIFT spatial registration algorithm to unify the real-time light intensity distribution data and cross-spectral polarization distribution parameters into the same coordinate system to obtain the registered spatiotemporal polarization dataset;

[0077] Next, three-dimensional voxelization processing is used to perform three-dimensional spatial sampling on the registered spatiotemporal polarization dataset, and the real-time light intensity distribution data and cross-spectral polarization distribution parameters are fused in three-dimensional space through a weighted fusion algorithm to generate fused polarization feature volume data; the discrete coordinate method is used to discretize the fused polarization feature volume data in eight directions to obtain discrete polarization radiation data; based on the discretized polarization radiation data, polarized light path simulation is performed, and noise suppression is simultaneously performed using a 3×3×3 Gaussian kernel to obtain polarized light transmission field data; the polarized light transmission field data is subjected to multi-channel shading using the OpenGL rendering engine, and polarization characteristics are mapped using the HSV color space converter to generate a simulated polarization transmission characteristic distribution cloud map.

[0078] S4, compare the simulated polarization transmission characteristic distribution cloud map with the multi-spectral polarization image dataset, analyze the residual between the actual polarization value and the simulated polarization value, obtain the adjustment calibration parameters, and optimize the non-uniform polarization light transmission model.

[0079] The specific steps are as follows:

[0080] S4.1. Use the three-dimensional ray tracing method to trace the three-dimensional photon paths of the simulated polarization transmission characteristic distribution cloud map to obtain the simulated polarization value. Specifically, the simulated polarization transmission characteristic distribution cloud map is first divided into voxel grids, and the initial light source parameters are set. For example, the wavelength is set to 532nm, the polarization state is set to horizontal linear polarization (S1=1), and the divergence angle is set to 5mrad, completing the construction of the three-dimensional voxel space.

[0081] A photon beam is emitted in a three-dimensional voxel space through a GPU parallel computing engine, and the three-dimensional photon path is traced using a three-dimensional ray tracing method. During the tracing process, the voxel position of each photon beam in the current three-dimensional voxel space is recorded, and the polarization state is converted using spatial coordinate mapping. The probability sampling method is simultaneously applied to randomly sample the scattering direction to generate a photon path dataset. The polarization parameters of the photon path dataset are accumulated, and the path is smoothed using quaternion interpolation to generate a three-dimensional path integral result containing complete polarization characteristics. Based on the three-dimensional path integral result, a polarization value generator is used to perform data extraction and formatting processing to generate simulated polarization values.

[0082] S4.2, perform four-channel intensity interpolation and frequency domain demodulation on the multi-spectral polarization image dataset to generate actual polarization values. Specifically, the multi-spectral polarization image dataset is first unified to the same resolution using the bicubic interpolation algorithm, eliminating pixel offsets between channels. Four-channel intensity interpolation is then performed synchronously to obtain a four-channel intensity image. Next, the four-channel intensity image is subjected to frequency domain demodulation using Fourier transform: a discrete Fourier transform is performed on the four-step phase-shifted intensity sequence of each pixel to extract the fundamental frequency component. An inverse tangent operation is simultaneously performed on the fundamental frequency component to obtain the polarization angle, and the amplitude component is used for normalization calculation to obtain the degree of polarization. Phase continuity fitting is performed on the polarization angle and degree of polarization to generate a preliminary polarization parameter map.

[0083] It should be noted that the four-step phase-shifted light intensity sequence is obtained by synchronously collecting the real-time light intensity distribution data in four directions of polarization; the amplitude component is obtained by performing amplitude analysis on the fundamental frequency component;

[0084] To eliminate spectrum aliasing, a Hanning window filter is applied to the preliminary polarization parameter map to generate an actual polarization value matrix. The actual polarization value matrix is ​​normalized and format converted to output the actual polarization value.

[0085] S4.3: Perform differential comparison calculations on the actual polarization values ​​and the simulated polarization values ​​to obtain residual distribution parameters. The residual distribution parameters are then iteratively corrected using the momentum gradient descent algorithm to generate adjustment calibration parameters. Specifically, the actual polarization values ​​and the simulated polarization values ​​are first aligned pixel by pixel and differentially calculated to generate residual values. The residual values ​​are then normalized and a sliding window filter is simultaneously applied to enhance local consistency to obtain residual distribution parameters. Next, weights are assigned to the residual distribution parameters and nonlinear superposition is performed to construct a loss function.

[0086] Based on the loss function, the momentum gradient descent algorithm is used to iteratively optimize the residual distribution parameters: in each iteration, the partial derivatives of the loss function are calculated, the historical gradient weights are applied to the partial derivatives, and the momentum attenuation factor is adjusted to generate the calibration parameter correction value. The specific mathematical formula is as follows:

[0087] ;

[0088] in, Indicates the calibration parameter correction value, represents the projection operator, represents the residual parameter vector of the current iteration step, represents the hyperparameter, represents the momentum decay factor, Represents the momentum accumulation vector of the previous iteration step; Represents the gradient vector of the loss function at the current iteration step;

[0089] It should be noted that the projection operator is based on the feasible domain of the residual distribution parameters, and is calibrated through polarization state transmission characteristics measurement and nonlinear fitting in optical experiments, with a value range of [1.3, 1.6]. The hyperparameter is defined based on the convergence stability of the loss function using grid search experiments, and has a value range of (0, 0.1]. The momentum decay factor is defined based on the memory strength of the historical gradient using the momentum decay rule, and has a value range of [0.8, 0.99].

[0090] When the correction value of the calibration parameter is lower than the iteration threshold, the iterative optimization is terminated, and the correction value of the calibration parameter is weighted averaged to output the adjusted calibration parameter;

[0091] It should be noted that the iteration threshold is defined based on the statistical stability of the residual distribution parameters and its value range is [0.0001, 0.01].

[0092] S4.4, input the adjusted calibration parameters into the non-uniform polarized light transmission model, apply the Markov chain Monte Carlo method to perform parameter inversion, and obtain the parameter posterior confidence interval. Specifically, the adjusted calibration parameters are input into the non-uniform polarized light transmission model as the prior distribution, and feature fusion is performed using a multi-scale convolutional layer. Parameter space mapping is performed through a variational autoencoder to generate a posterior probability distribution; the Markov chain Monte Carlo method is used to randomly sample the posterior probability distribution to obtain an initial parameter sampling sequence; based on the initial parameter sampling sequence, parameter inversion is performed through a ray tracing engine to form a candidate parameter probability distribution; and the Gaussian suggestion distribution is used to perform parameter perturbation on the candidate parameter probability distribution to generate a candidate parameter set;

[0093] The Metropolis-Hastings sampler is used to perform probability density sampling on the candidate parameter set and perform matching conversion to obtain the acceptance probability value. Based on the acceptance probability value, the parameter acceptance of the candidate parameter set is judged by a random threshold: if the acceptance probability value is greater than the random threshold, the new parameter is judged to be accepted; if the acceptance probability value is less than the random threshold, the new parameter is judged to be rejected.

[0094] It should be noted that the random threshold is defined based on the uniform distribution probability benchmark and has a value range of (0,1);

[0095] After completing the parameter acceptance judgment, the warm-up period filter is used to perform initial sample elimination and sample thinning on the candidate parameter set to generate stationary posterior distribution samples; the quantile extractor is applied to intercept the sample interval of the stationary posterior distribution samples to obtain the parameter posterior confidence interval.

[0096] S4.5, incrementally update the weight parameters of the non-uniform polarized light transmission model based on the posterior confidence interval of the parameters, and output the optimized non-uniform polarized light transmission model. Specifically, first, use the interval midpoint calculator to perform arithmetic averaging on the posterior confidence interval of the parameters to extract the center value of the interval; use the center value of the interval as the reference point, calculate the upper and lower bound differences of the posterior confidence interval of the parameters, and apply the step size scaling factor to scale the safety factor to obtain the parameter update step size;

[0097] It should be noted that the step size scaling factor refers to the proportional coefficient between the parameter update step size and the width of the parameter posterior confidence interval, and its value range is (0,0.2];

[0098] The automatic differentiation engine of PyTorch is used to perform gradient backpropagation on the non-uniform polarization light transmission model to obtain the gradient values ​​of the parameters of each layer. The L2 norm of the gradient values ​​of the parameters of each layer is calculated and normalized to generate the parameter importance score. The parameter importance scores are sorted by importance to obtain the weight parameters of the non-uniform polarization light transmission model. According to the parameter update step size, the weight parameters of the non-uniform polarization light transmission model are incrementally updated through the AdamW optimizer, and the optimized non-uniform polarization light transmission model is synchronously output.

[0099] This embodiment also provides a computer device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the multi-spectral polarization light transmission characteristic testing method based on passive imaging as proposed in the above embodiment.

[0100] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0101] This embodiment further provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for testing the multi-spectral polarization light transmission characteristics based on passive imaging proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0102] In summary, this invention utilizes advanced multidimensional image fusion technology to efficiently handle polarization correlations and noise interference between different spectral bands, improving the accuracy of image data and significantly reducing depolarization rate measurement errors in smoky environments. This effectively addresses the transmission characteristic inversion errors caused by registration bias in traditional methods. Precise simulation using a non-uniform polarized light transmission model and an optical simulation engine more realistically reflects the complex factors in the actual environment, significantly improving the reliability and accuracy of transmission characteristic testing in complex environments.

[0103] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for testing multi-spectral polarized light transmission characteristics based on passive imaging, characterized by: include: The multi-spectral polarization image dataset is spatiotemporally aligned and weightedly fused using a multi-dimensional image fusion method to generate a polarization coupling matrix. The polarization coupling matrix is ​​input into the non-uniform polarized light transmission model. The multi-scale convolution layer extracts depolarization features. The attention mechanism layer analyzes the polarization correlation between spectrum bands and outputs a multi-spectral polarized light transmission characteristic test solution. The multi-spectral polarized light transmission characteristic test scheme is input into the optical simulation engine, driving the LabVIEW unit to perform polarized scattered light imaging simulation in the smoke environment and obtain the simulated polarization transmission characteristic distribution cloud map; The simulated polarization transmission characteristic distribution cloud map is compared with the multi-spectral polarization image dataset, and the residuals between the actual polarization values ​​and the simulated polarization values ​​are analyzed to obtain the adjustment calibration parameters and optimize the non-uniform polarized light transmission model.

2. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: The multi-spectral polarization image dataset includes multi-spectral polarization image data, environmental parameters, and device metadata.

3. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: Generating the polarization coupling matrix specifically includes the following steps: Wavelet decomposition is used to suppress noise in multi-spectral polarization image datasets, and SIFT algorithm and linear interpolation are applied to perform cross-spectral spatial registration and temporal synchronization to obtain multi-spectral polarization intensity tensors. A multi-dimensional image fusion method is used to assign dynamic weights to the multi-spectral polarization intensity tensor, which is coupled with environmental parameters. The polarization coupling matrix is ​​output through tensor reorganization.

4. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: The specific construction process of the non-uniform polarized light transmission model is as follows: Residual connections and gating mechanisms are used to perform hierarchical parameter fusion of multi-scale convolutional layers and attention mechanism layers to construct a non-uniform polarization light transmission model.

5. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: The output multi-spectral polarized light transmission characteristic test solution specifically includes the following steps: The multi-scale convolution layer uses multi-branch convolution kernels to perform cross-scale debiased feature extraction and residual fusion to generate debiased feature tensors; The attention mechanism layer dynamically weights the polarization correlation through a gating mechanism to obtain the spectral segment attention weight matrix; The depolarization feature tensor and the spectral attention weight matrix are spliced ​​in the feature channel to form a polarization coupling feature map; ReLU activation and batch normalization are used to perform feature channel compression and task adaptation mapping on the polarization coupling feature map to generate a multi-spectral polarization light transmission characteristic test solution.

6. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: The step of obtaining the simulated polarization transmission characteristic distribution cloud map specifically includes the following steps: The Stokes parameters and polarization parameters in the multi-spectral polarization light transmission characteristics test scheme are extracted using the amplitude-division polarization modulation method and input into the optical simulation engine through the Matlab-Stokes interface; The optical simulation engine drives the PID controller of the LabVIEW unit to adjust the smoke concentration gradient and simulate polarized scattered light imaging. It also uses the four-quadrant difference algorithm to calculate the depolarization rate and obtain the polarization distribution parameters across the spectral band. Collect real-time light intensity distribution data, perform three-dimensional spatial sampling and fusion on the real-time light intensity distribution data and cross-spectral polarization distribution parameters, and perform polarization characteristic mapping to form a simulated polarization transmission characteristic distribution cloud map.

7. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 6, wherein: The obtaining and adjusting calibration parameters specifically includes the following steps: The three-dimensional photon path is traced on the simulated polarization transmission characteristic distribution cloud map using the three-dimensional ray tracing method to obtain the simulated polarization value; Perform four-channel light intensity interpolation and frequency domain demodulation on the multi-spectral polarization image dataset to obtain the actual polarization value; The actual polarization value and the simulated polarization value are differentially compared and calculated to obtain the residual distribution parameters; the momentum gradient descent algorithm is used to perform gradient iterative correction on the residual distribution parameters to generate adjustment calibration parameters.

8. The method for testing multi-spectral polarized light transmission characteristics based on passive imaging according to claim 1, wherein: The optimization of the non-uniform polarized light transmission model specifically includes the following steps: The adjusted calibration parameters are input into the inhomogeneous polarization light transmission model, and the Markov chain Monte Carlo method is used to perform parameter inversion to obtain the posterior confidence interval of the parameters. According to the posterior confidence interval of the parameters, the weight parameters of the non-uniform polarization light transmission model are incrementally updated, and the optimized non-uniform polarization light transmission model is output.

Images