Bionic polarized light navigation enhancement method and system for extreme weather

Abstract

The application discloses a kind of bionic polarized light navigation enhancement method and system for extreme weather, including, obtain sky polarization original data and auxiliary navigation information, according to the failure criterion of underexposure, low polarization degree and local inconsistency to identify the invalid part in data, generate the mask image of the identification effective and invalid area;Original data, auxiliary information and mask image are input into a pre-trained, with the physical model constraint of internal polarized light field physical information neural network;The network uses its physical reasoning ability to predictively repair invalid area, combine the original data of effective area, output predicted polarization pattern diagram;Based on the polarization pattern diagram, the heading information of carrier is solved;The application integrates physical model into neural network, can fundamentally reconstruct the polarization information that a large area is missing due to extreme weather, improves the robustness and safety of navigation method under all-weather conditions by model parameter adaptive learning and output navigation confidence.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and autonomous navigation technology, and in particular to a biomimetic polarized light navigation enhancement method and system for extreme weather conditions. Background Technology

[0002] Biomimetic polarized light navigation technology, as an autonomous navigation method that mimics the use of polarized light patterns in the sky by organisms such as desert ants for positioning and orientation, shows broad application prospects in fields such as drones, autonomous robots, and vehicles due to its passive, error-free, and high-precision characteristics. Specifically, this technology utilizes the predictable Rayleigh scattering of sunlight by atmospheric molecules under clear weather conditions to form a stable, all-sky polarized light distribution pattern with the solar / anti-solar meridian as its axis of symmetry. The navigation system captures this pattern using a polarization sensor, accurately calculates the solar azimuth angle, and then determines the vehicle's heading, achieving high-precision navigation.

[0003] However, due to the very definition of biomimetic polarization-based navigation technology, its application scenarios are severely limited by weather conditions. In extreme weather conditions such as dense fog, torrential rain, sandstorms, or thick cloud cover, the atmosphere is filled with large aerosols or water droplets much larger than the wavelength of light, causing light transmission to shift from Rayleigh scattering to complex Mie scattering and multiple scattering. This fundamental change in physical processes not only severely disrupts the original stable polarization pattern but also leads to polarization distortion, a sharp drop in signal-to-noise ratio, and even the emergence of depolarized regions where polarization information is completely lost over vast areas of the sky. Existing enhancement methods mostly focus on filtering signals or fusing multi-source data. These methods simply treat signal degradation under extreme weather conditions as noise superposition, lacking in-depth modeling of the complex scattering physics behind it. Therefore, when faced with large-area information loss rather than just noise interference, traditional methods cannot fundamentally repair and reconstruct the damaged polarization pattern, leading to a precipitous drop in navigation accuracy or even complete failure, greatly limiting the all-weather, highly robust application of biomimetic polarization-based navigation technology. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a biomimetic polarization-enhanced navigation method for extreme weather conditions, addressing the problems mentioned in the background art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a biomimetic polarization-guided navigation enhancement method for extreme weather conditions, comprising:

[0008] Acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on preset failure criteria, identify invalid parts in the raw sky polarization data and generate mask images that identify valid and invalid data.

[0009] The original sky polarization data, auxiliary navigation information, and mask image are input into a pre-trained physical information neural network. Based on the inherent polarization light field physical model of the physical information neural network, information prediction is performed on the invalid region, and combined with the original data of the valid region, a predicted polarization pattern map is output.

[0010] Based on the predicted polarization mode diagram, the heading information of the carrier is calculated.

[0011] As a preferred embodiment of the biomimetic polarization navigation enhancement method for extreme weather described in this invention, the original sky polarization data is a Stokes vector image; the auxiliary navigation information includes time, geographical location information, and carrier attitude information; the failure criteria include underexposure criteria, polarization degree below a preset threshold criteria, and data local inconsistency criteria.

[0012] As a preferred embodiment of the biomimetic polarization navigation enhancement method for extreme weather described in this invention, the physical information neural network adopts an encoder-decoder structure and is provided with a skip connection for transmitting multi-scale spatial features between the encoder and the decoder.

[0013] In a preferred embodiment of the biomimetic polarization-guided navigation enhancement method for extreme weather described in this invention, the training of the physical information neural network is accomplished by minimizing a hybrid loss function, wherein the hybrid loss function is configured such that the output of the physical information neural network simultaneously satisfies the following conditions during training:

[0014] First condition: Within the valid data area marked by the mask image, the output of the physical information neural network is consistent with the original sky polarization data;

[0015] The second condition is that the output of the physical information neural network conforms to the rules defined by the polarization field physical model within the entire mask image domain.

[0016] As a preferred embodiment of the biomimetic polarized light navigation enhancement method for extreme weather described in this invention, the polarized light field physical model is defined as a partial differential equation that simultaneously satisfies the following physical effects:

[0017] The effects of polarized light diffusion and smoothing in space;

[0018] Polarized light experiences intensity attenuation and depolarization effects due to atmospheric scattering;

[0019] The source effect of the entire polarization mode directionality determined by the position of the sun.

[0020] As a preferred embodiment of the biomimetic polarization light navigation enhancement method for extreme weather described in this invention, it further includes:

[0021] In the physical model of the polarized light field, the coefficients of the attenuation and depolarization effects and the intensity of the source effect are used as learnable parameters of the model.

[0022] The learnable parameters are combined with the weights of the physical information neural network itself during the training process for optimization.

[0023] As a preferred embodiment of the biomimetic polarization light navigation enhancement method for extreme weather described in this invention, it further includes:

[0024] By evaluating the degree of agreement between the predicted polarization pattern and the physical model of the polarized light field, as well as the overall polarization intensity of the predicted polarization pattern, the confidence level of the solved heading information of the carrier is calculated and output.

[0025] Secondly, the present invention provides a biomimetic polarization navigation enhancement system for extreme weather conditions, comprising:

[0026] The data acquisition and preprocessing module is configured to acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on a preset failure criterion, it identifies invalid parts in the raw sky polarization data and generates a mask image that identifies the valid and invalid areas of the data.

[0027] The polarization mode prediction and repair module is configured to input the original sky polarization data, auxiliary navigation information, and mask image into a pre-trained physical information neural network. Based on the polarization light field physical model inherent in the physical information neural network, it performs information prediction on the invalid region and outputs a predicted polarization mode map by combining the original data of the valid region.

[0028] The heading information calculation module is configured to calculate the heading information of the carrier based on the predicted polarization mode diagram.

[0029] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described method.

[0030] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the above-described method.

[0031] Compared with existing technologies, the beneficial effects of the invention are:

[0032] 1. This invention integrates the physical model of polarized light field into the physical information neural network in the form of inherent constraints, giving the model the ability to make physical inferences. At the same time, in the face of large-area loss of polarization information under extreme weather conditions, this invention no longer performs data filtering or interpolation, but performs predictive repair and reconstruction based on physical laws. This fundamentally solves the navigation failure problem caused by the reliance on the integrity of input data in traditional methods, and expands the applicable scenarios of biomimetic polarized light navigation technology.

[0033] 2. By using key parameters (such as the depolarization effect coefficient) in the polarization field physical model as learnable parameters, and optimizing them together with the weights of the neural network during training, the physical model can learn autonomously from the data and accurately characterize the specific impact of different weather conditions (such as fog, rain, and clouds) on polarization light transmission. This overcomes the shortcomings of traditional physical models, such as poor universality and inability to accurately match changing environments, thereby improving the accuracy of navigation solutions under various severe weather conditions.

[0034] 3. In addition, this invention not only provides heading calculation results, but also outputs the confidence level of the current navigation result by quantitatively evaluating the degree of conformity between the repaired polarization mode and physical laws. This confidence level can provide decision-making basis for multi-sensor fusion systems (such as fusion with IMU and GNSS), making the entire navigation process more intelligent and safer. Attached Figure Description

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0036] Figure 1 This is a flowchart illustrating the overall process of a biomimetic polarized light navigation enhancement method for extreme weather, as described in one embodiment of the present invention. Detailed Implementation

[0037] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0038] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used 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 different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0040] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0041] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0042] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0043] Example 1

[0044] Reference Figure 1This is the first embodiment of the present invention, which provides a biomimetic polarization-enhanced navigation method for extreme weather conditions, comprising:

[0045] S1. Acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on preset failure criteria, identify invalid parts in the raw sky polarization data and generate mask images that identify valid and invalid data.

[0046] Furthermore, raw data on sky polarization can be acquired through a biomimetic polarization navigation sensor mounted on a carrier (such as a drone or vehicle).

[0047] Specifically, this biomimetic polarization navigation sensor uses Division of Focal Plane (DoFP) technology and is equipped with a fisheye lens, enabling it to capture polarization information of the entire sky in a single exposure.

[0048] Specifically, the acquired raw sky polarization data is a Stokes vector image, which consists of multiple channels, including at least the following channels:

[0049] S0 channel: represents the total light intensity at all points in the sky; S1 channel: represents the intensity difference of linearly polarized light in the 0° and 90° directions; S2 channel: represents the intensity difference of linearly polarized light in the 45° and 135° directions.

[0050] It should be noted that the above three channels together describe the linear polarization state of sky light;

[0051] In addition, while acquiring the raw sky polarization data, the biomimetic polarization navigation system integrates a time module, a GPS / BDS receiver, and an inertial measurement unit (IMU) to simultaneously acquire auxiliary navigation information that is spatiotemporally aligned with each frame of Stokes vector image. This auxiliary navigation information includes time information, geographical location information, and carrier attitude information.

[0052] Specifically, time information (such as UTC time) and geographical location information (latitude and longitude) are used to calculate the theoretical position of the sun on the celestial sphere (i.e., the solar azimuth and altitude angles), representing the physical source driving the directionality of polarization modes;

[0053] Specifically, the carrier attitude information (roll angle, pitch angle, yaw angle) is used to transform the Stokes vector image captured in the sensor coordinate system to the geographic coordinate system through coordinate transformation, thereby eliminating the impact of carrier turbulence and tilt on sky polarization mode observations.

[0054] It should be noted that due to the influence of extreme weather, the acquired Stokes vector images may contain a large amount of unreliable or even completely erroneous information. Therefore, it is necessary to establish a preset failure criterion and conduct data quality assessment and invalid region identification pixel by pixel.

[0055] Furthermore, the failure criteria include a combination of the following three factors:

[0056] Underexposure criteria: Check the pixel values ​​of the S0 channel. If the value of a pixel reaches or approaches the sensor's maximum range (e.g., 255 for an 8-bit image), it is considered oversaturated (usually caused by direct sunlight or strong reflections). If its value is lower than the preset minimum response threshold, it is considered underexposure (usually caused by extremely thick cloud cover). Since the S1 and S2 values ​​in this area are distorted, they are also identified as invalid.

[0057] Polarization degree below preset threshold criterion: Calculate the degree of linear polarization (DoLP) of each pixel; due to the strong depolarization effect produced by multiple scattering in weather conditions such as dense fog and heavy rain, resulting in extremely low DoLP values, a polarization degree threshold (e.g., 0.02) can be set; if the DoLP value of a pixel is lower than this polarization degree threshold, it is considered that its polarization signal is weak, the signal-to-noise ratio is too low, and it is easily dominated by noise, so it is identified as invalid;

[0058] Preferably, the polarization threshold is no longer a fixed empirical value, but is determined using a dynamic adaptive method: For the current Stokes vector image, a statistical histogram of the polarization degree of all pixels is calculated. Since the DoLP value of invalid regions (such as thick clouds and dense fog) generally approaches zero, while the DoLP value of valid regions has a wide distribution range, the histogram usually presents a bimodal or "L"-shaped distribution. At this time, by using automatic threshold segmentation algorithms such as Otsu's method or K-means clustering to analyze the histogram, the threshold point that can best separate low DoLP noise regions and high DoLP signal regions can be found, and this threshold point is used as the polarization threshold of the current pixel. This adaptive method allows the polarization threshold to be dynamically adjusted according to real-time weather changes, making it more robust.

[0059] Specifically, the linear polarization degree of each pixel is calculated using the following formula:

[0060]

[0061] Local data inconsistency criterion: Calculate the angle of polarization (AoP) image. Because physically, the polarization angle of the sky should change smoothly in space. If there is an abnormal abrupt change in the polarization angle gradient or variance in the local neighborhood of a certain pixel, which is far beyond the range of change under normal weather conditions, it indicates that the point may be affected by local reflection, sensor noise or complex scattering. Its data consistency is poor and should be identified as invalid.

[0062] Specifically, the variation range under normal weather conditions is calibrated offline: a standard angle of polarization (AoP) image dataset containing N (e.g., N=1000) images under clear or partly cloudy weather conditions is collected. For each pixel in each image in the dataset, the gradient magnitude within an M×M (e.g., M=5) neighborhood window is calculated. All calculated gradient magnitudes are summarized to construct a global gradient magnitude distribution histogram. The 99.5% quantile of this distribution is taken as the upper limit threshold for anomaly detection. In real-time processing, if the local gradient magnitude of the AoP of a pixel exceeds this upper limit threshold, it is determined to be invalid.

[0063] Specifically, the formula for calculating the polarization angle image is expressed as follows:

[0064]

[0065] It should be noted that, except for the above-mentioned invalid identification, all other cases are valid;

[0066] Furthermore, based on the failure criteria, a single-channel mask image with the same size as the Stokes vector image is generated, and a value is assigned to each pixel position in the mask image: if the original data at that position is determined to be valid, a reference value (1) is assigned to the corresponding position in the mask image; if it is determined to be invalid, another reference value (0) is assigned.

[0067] It should be noted that the mask image clearly identifies the valid and invalid regions in the original data;

[0068] S2. Input the original sky polarization data, auxiliary navigation information, and mask image into a pre-trained physical information neural network. Based on the inherent polarization light field physical model of the physical information neural network, perform information prediction on the invalid region and combine it with the original data of the valid region to output a predicted polarization pattern map.

[0069] It should be noted that a physical information neural network model is constructed to achieve intelligent repair and reconstruction of polarization information damaged under extreme weather conditions;

[0070] Furthermore, the Stokes vector image (raw sky polarization data) obtained in step S1, the generated mask image, and the processed auxiliary navigation information are stitched together as a unified input to the physical information neural network model. The auxiliary navigation information (calculated sun position) is mainly used to parameterize the subsequent polarization light field physical model, rather than being directly used as the image channel input.

[0071] Specifically, the physical information neural network adopts an encoder-decoder structure and is equipped with skip connections that transmit multi-scale spatial features between the encoder and the decoder. The advantage of this configuration is that it is suitable for image-to-image conversion tasks, such as the classic U-Net architecture.

[0072] Specifically, the encoder structure consists of several (at least one) convolutional layers and downsampling layers (such as max pooling layers) stacked together. It is mainly responsible for receiving multi-channel input data, extracting features from low-level texture to high-level semantics through layer-by-layer convolution, and continuously expanding the receptive field during downsampling to capture the global distribution features of polarization patterns.

[0073] Specifically, the decoder structure consists of several deconvolutional and convolutional layers (corresponding to the number of convolutional and downsampling layers in the encoder structure). It is mainly responsible for progressively decoding the semantic features extracted by the encoder, restoring them to the original image resolution, and generating the repaired polarization pattern map.

[0074] Specifically, the skip connection is set after each downsampling step of the encoder, directly transmitting its output feature map and concatenating it to the corresponding resolution layer of the decoder. This allows the decoder to directly utilize the shallow, high-resolution features from the encoder without depth compression when reconstructing the image. For this invention, this is an auxiliary innovation because it ensures that in the effective data area marked by the mask image, the output of the physical information neural network can retain the original and reliable polarization details to the greatest extent, preventing the input data from being lost during the encoding and decoding process.

[0075] Furthermore, a training dataset is constructed for this physical information neural network: a large number of samples containing paired data are collected as the training dataset, where each pair of samples includes a partially damaged original Stokes vector image acquired under extreme weather conditions; a theoretically lossless ideal polarization pattern map (as the "ground truth") generated by physical simulation software (such as a radiative transfer model based on the Monte Carlo method) at the same time and geographical location corresponding to the Stokes vector image; a mask image generated from the original partially damaged Stokes vector image; and auxiliary navigation information from the original Stokes vector image.

[0076] Furthermore, pre-training is performed based on the obtained training dataset. The pre-training process is accomplished by minimizing a mixture loss function, which is configured to drive the output of the physical information neural network model to simultaneously satisfy the following two complementary conditions during the model's pre-training process via backpropagation. This ensures that the physical information neural network model is not only a data fitter but also a "follower" of physical laws:

[0077] The first condition is the data fidelity constraint, which means that within the effective data area marked by the mask image, the predicted polarization pattern map output by the neural network should be as consistent as possible with the input, undisturbed original sky polarization data. The loss in the process of maintaining consistency is usually called data loss. The calculation steps are as follows: calculate the mean square error between the output of the physical information neural network model and the original Stokes vector image only at the pixel point with a mask value of 1, so as to ensure that the physical information neural network model makes full use of all reliable observation data.

[0078] The second condition is the physical law consistency constraint, that is, in the entire image domain (including the valid and invalid data areas), the output of the physical information neural network model must conform to the laws defined by the preset polarization light field physical model. The part of the loss generated in this process is called physical loss.

[0079] Furthermore, during the pre-training process, the training dataset is input into the physical information neural network model in batches. For each batch, forward propagation is performed to obtain the prediction map, the mixed loss is calculated, and then the gradient of the mixed loss function with respect to the weights and learnable parameters of the physical neural network model is calculated through the backpropagation algorithm (Adam optimizer), and all parameters are updated. This process is repeated until the model's performance on the validation set converges.

[0080] Specifically, the calculated mixing loss L is expressed as:

[0081] L=α×DL+β×PL

[0082] Where α and β are hyperparameters used to balance the importance of data fidelity and consistency with physical laws; DL represents data loss and PL represents physical loss.

[0083] Furthermore, in the early stages of model training, α = 1.0 and β = 0.1 are set to allow the neural network to first learn to fit effective data. As training progresses, the value of β is gradually increased (for example, by using an annealing strategy to eventually reach β = 1.0) to strengthen the constraints of physical laws. The initial learning rate of the model is set to 1e-4, and a learning rate decay strategy is adopted. For example, whenever the validation set loss does not decrease within 10 epochs, the learning rate is multiplied by 0.5. The batch size of the model is set to 8, 16, or 32 depending on the hardware (such as GPU memory) capabilities.

[0084] Furthermore, the physical model of polarized light field is defined as a partial differential equation, which represents a mathematical expression of the fundamental physical processes of sky-polarized light propagating in a complex atmospheric medium. This equation couples the following physical effects:

[0085] The effect of polarized light spreading and smoothing in space is characterized by the Laplace operator term. The Laplace operator term ensures that the repaired polarization pattern is continuous and smooth in space, avoiding unnatural sharp abrupt changes at the edges of invalid regions, which conforms to the natural properties of light field distribution.

[0086] Polarized light experiences intensity attenuation and depolarization due to atmospheric scattering, which is characterized by an attenuation term. This attenuation term describes the phenomenon that the total light intensity (S0) and the intensity difference between linearly polarized light (S1, S2) decrease as the transmission path increases when light passes through media such as dense fog and clouds.

[0087] The source effect of the entire polarization mode directionality determined by the position of the Sun is characterized by a source term, which is the driving force for the formation of the polarization mode. Its shape and direction are uniquely determined by the azimuth and altitude angles of the Sun on the celestial sphere (calculated through auxiliary navigation information).

[0088] It should be noted that this source effect defines the axis of symmetry and the overall distribution trend of the polarization mode, which is the basis for ensuring the correctness of navigation solutions;

[0089] Furthermore, the physical model of this polarized light field can be expressed as:

[0090] S(x,y)=[S0(x,y),S1(x,y),S2(x,y)] T

[0091] The system of partial differential equations distributed in the image spatial domain (x,y) can be expressed in general form as follows:

[0092]

[0093] in, This represents the effect of polarized light spreading and smoothing in space, and is a Laplace operator term. This term acts on every component of the Stokes vector; S(x,y) represents the intensity attenuation and depolarization effect of polarized light caused by atmospheric scattering, and is the attenuation term; The source effect representing the directionality of the entire polarization mode determined by the Sun's position is the source term. Its intensity and direction are determined by the Sun's theoretical position in the celestial coordinate system (altitude angle θ). s and azimuth The source term is determined and can be modeled as an ideal polarization mode distribution function that conforms to Rayleigh scattering theory, serving as the basic form driving the formation of the entire physical field; A and B are the attenuation coefficient and the source intensity coefficient, respectively, serving as learnable parameters of the model. During the training process, the model will autonomously learn the values ​​of A and B that best describe the current weather conditions (such as fog concentration and cloud thickness) based on the input real data, thereby enabling the physical model itself to have the ability to adapt to the environment.

[0094] Furthermore, the physical loss is calculated by substituting the output of the physical information neural network model (i.e., the predicted Stokes vector image) into the above partial differential equation and calculating the residual of the partial differential equation. If the output of the physical information neural network model fully conforms to physical laws, the residual of the partial differential equation is zero. Therefore, by minimizing the L2 norm of the residual of the partial differential equation, the physical information neural network model can be forced to learn to generate a physically valid polarization mode.

[0095] Furthermore, to achieve a high degree of adaptability to different extreme weather conditions, the coefficients of attenuation and depolarization effects and the intensity of the source effect can be used as learnable parameters in the polarization field physical model. These learnable parameters are then combined with the weights of the physical information neural network during training for optimization. Since traditional physical models typically use fixed, idealized parameters, they are difficult to adapt to real and changing weather conditions. This invention addresses this by no longer treating physical coefficients in the partial differential equations (e.g., the attenuation coefficient characterizing atmospheric turbidity) as constants, but instead incorporating them as part of the physical information neural network model. These coefficients are optimized during training along with convolutional kernel weights, biases, etc. This means that the physical information neural network model can autonomously learn the physical parameters that best describe the current weather conditions (e.g., light fog, heavy fog, thin clouds) from the data. This makes the physical model itself environmentally adaptable, greatly improving the accuracy and physical realism of the repair results.

[0096] It should be noted that the pre-trained physical information neural network, after receiving real-time, partially damaged Stock vector images, masks and auxiliary information, can use its learned physical reasoning ability to make predictions of invalid areas marked by the mask in accordance with physical laws, and seamlessly integrate them with the original data of the valid areas to output a physically consistent predicted polarization pattern map, so as to provide a reliable basis for subsequent heading calculation.

[0097] S3. Based on the predicted polarization mode diagram, calculate the heading information of the carrier;

[0098] Furthermore, the solar meridian is extracted from the predicted polarization pattern map;

[0099] It needs to be explained that one of the most fundamental and stable features of the sky polarization pattern is its symmetry, that is, with the great circle containing the sun / anti-sun as the axis of symmetry, which is called the solar meridian;

[0100] Preferably, the solar meridian is extracted using a symmetry search algorithm, as follows:

[0101] At the center of the image coordinate system of the predicted polarization pattern map, several candidate axes of symmetry are generated with a preset angular resolution (e.g., 0.5 degrees), covering an angle range from 0° to 180°; then, for each candidate axis of symmetry, all valid pixels p in the image are traversed. i And calculate p i The point p′ symmetric about the candidate axis of symmetry i If p′ i If it is still within the valid range of the image, then read p. i and p′ i The polarization angle AoP(p) at the location i ) and AoP(p′ i According to the principle of symmetry, AoP(p′) i )=-AoP(p i ) or AoP(p′ i )+AoP(p i If ) = 0, calculate its symmetry error or similarity measure; sum the errors (or similarity measures) of all pixels to get the total score of the candidate axis; find the angle that minimizes (for error) or maximizes (for similarity) the total score of the candidate axis, and the line corresponding to the angle is the extracted solar meridian;

[0102] Specifically, the symmetry error formula is expressed as:

[0103] e i =|AoP(p′) i )+AoP(p i )|

[0104] Among them, e i This is represented as symmetry error;

[0105] Specifically, the similarity measurement formula is expressed as:

[0106] s i =cos(2×(AoP(p′)) i )+AoP(p i )))

[0107] Among them, s i Represented as a similarity measure;

[0108] It should be noted that since the predicted polarization pattern map has been physically repaired, its polarization pattern is complete and has a low noise level. Therefore, the accuracy and robustness of the solar meridian extraction process can be improved compared to the process performed directly on the original damaged data.

[0109] Furthermore, the carrier's heading information is calculated using the extracted solar meridian. The heading angle is defined as the angle between the forward axis of the carrier's coordinate system and the geographic north direction. The calculation process is as follows:

[0110] S301. Determine the sun's azimuth in the image coordinate system: The direction of the extracted solar meridian in the image coordinate system is directly used as the azimuth angle of the sun relative to the carrier's head.

[0111] S302. Obtain the theoretical solar azimuth angle: Using the time and geographical location information obtained synchronously in step S1, calculate the theoretical azimuth angle of the sun in the geographic coordinate system at this moment (i.e., the angle of the sun relative to geographic north) through astronomical algorithms.

[0112] S303. Calculate the heading angle: Since the heading angle of the vehicle is equal to the difference between the theoretical solar azimuth angle and the solar azimuth in the image coordinate system, the current heading information of the vehicle can be calculated by using this difference.

[0113] Furthermore, this invention also calculates and outputs a confidence index characterizing the reliability of the currently solved heading information by evaluating the degree of conformity between the predicted polarization pattern diagram and the physical model of the polarized optical field, as well as the overall polarization intensity of the predicted polarization pattern diagram. This allows the navigation information to not only provide the result but also inform the user how reliable the result is. The confidence index calculation integrates the following assessments of physical consistency and signal quality:

[0114] The physical consistency assessment quantifies the degree of conformity between the predicted polarization pattern map and the polarization field physical model. Specifically, the output predicted polarization pattern map is substituted back into the polarization field physical model in the form of partial differential equations. The global norm of the physical residual over the entire image domain is calculated. The smaller the calculated physical residual value, the closer the solution generated by the physical information neural network model is to the solution of the polarization field physical model. Correspondingly, the higher the physical authenticity, the more the residual value is normalized in reverse and used as the score of physical consistency.

[0115] Signal quality assessment quantifies the overall polarization intensity of the predicted polarization pattern map. This is achieved by calculating the average linear polarization degree (DoLP) of all pixels on the predicted polarization pattern map. Since the depolarization effect is significant under extreme weather conditions, the DoLP is generally low. Therefore, after repair by the method of this invention, a high-quality predicted pattern map should restore a reasonable polarization intensity. Thus, the average DoLP value can be used as an intuitive indicator to measure the overall quality of the repaired signal. The higher the value, the stronger the polarization signal and the better the signal-to-noise ratio of the navigation solution.

[0116] Furthermore, the obtained physical consistency score and signal quality score are normalized and fused using a preset weighting function. Finally, a confidence index in the range [0,1] is calculated and output. For example, a confidence index of 0.95 indicates that the current headway information has very high reliability; while a confidence index of 0.3 warns the user that although a headway value is given, the result has low reference value due to the current harsh environment or poor model repair effect.

[0117] Specifically, the preset weighting function is represented by a linear weighted summation:

[0118] CI=ω×(Pcs)+(1-ω)×(Sqc)

[0119] Where Pcs represents the physical consistency score, Sqc represents the signal quality assessment score, and ω represents the weighting coefficient, which ranges from [0,1]. This value can be determined by performing a grid search on the test dataset, thus making the confidence level have the strongest negative correlation with the actual heading calculation error.

[0120] It should be noted that this confidence level can provide decision-making support for upper-level applications (such as flight control systems and multi-sensor fusion navigation systems), making the entire navigation system more intelligent and safer.

[0121] Furthermore, this embodiment also provides a biomimetic polarized light navigation enhancement system for extreme weather conditions, including:

[0122] The data acquisition and preprocessing module is configured to acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on a preset failure criterion, it identifies invalid parts in the raw sky polarization data and generates a mask image that identifies the valid and invalid areas of the data.

[0123] The polarization mode prediction and repair module is configured to input the original sky polarization data, auxiliary navigation information, and mask image into a pre-trained physical information neural network. Based on the polarization light field physical model inherent in the physical information neural network, it performs information prediction on the invalid region and outputs a predicted polarization mode map by combining the original data of the valid region.

[0124] The heading information calculation module is configured to calculate the heading information of the carrier based on the predicted polarization mode diagram.

[0125] This embodiment also provides a computer device applicable to the use of biomimetic polarization navigation enhancement methods for extreme weather conditions, including:

[0126] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement the biomimetic polarization navigation enhancement method for extreme weather conditions proposed in the above embodiments.

[0127] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices 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 the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0128] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the biomimetic polarization navigation enhancement method for extreme weather as proposed in the above embodiments.

[0129] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0130] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0131] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0132] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0134] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0135] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A biomimetic polarization-guided navigation enhancement method for extreme weather conditions, characterized in that, include: Acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on preset failure criteria, identify invalid parts in the raw sky polarization data and generate mask images that identify the valid and invalid regions of the data. The original sky polarization data, auxiliary navigation information, and mask image are input into a pre-trained physical information neural network. Based on the inherent polarization light field physical model of the physical information neural network, information prediction is performed on the invalid region, and combined with the original data of the valid region, a predicted polarization pattern map is output. Based on the predicted polarization mode diagram, the heading information of the carrier is calculated; Also includes: By evaluating the degree of agreement between the predicted polarization pattern and the physical model of the polarized light field, as well as the overall polarization intensity of the predicted polarization pattern, the confidence level of the solved heading information of the carrier is calculated and output.

2. The biomimetic polarized light navigation enhancement method for extreme weather as described in claim 1, characterized in that, The raw sky polarization data is a Stokes vector image; the auxiliary navigation information includes time, geographic location information, and vehicle attitude information. The failure criteria include underexposure criteria, polarization degree below a preset threshold criteria, and data local inconsistency criteria.

3. The biomimetic polarized light navigation enhancement method for extreme weather as described in claim 1, characterized in that, The physical information neural network adopts an encoder-decoder structure and is equipped with skip connections that transmit multi-scale spatial features between the encoder and the decoder.

4. The biomimetic polarized light navigation enhancement method for extreme weather as described in claim 1, characterized in that, The training of the physical information neural network is accomplished by minimizing a hybrid loss function, which is configured such that the output of the physical information neural network simultaneously satisfies the following conditions during training: First condition: Within the valid data area marked by the mask image, the output of the physical information neural network is consistent with the original sky polarization data; The second condition is that the output of the physical information neural network conforms to the rules defined by the polarization field physical model within the entire mask image domain.

5. The biomimetic polarized light navigation enhancement method for extreme weather as described in claim 4, characterized in that, The physical model of the polarized light field is defined as a partial differential equation that simultaneously satisfies the following physical effects: The effects of polarized light diffusion and smoothing in space; Polarized light experiences intensity attenuation and depolarization effects due to atmospheric scattering; The source effect of the entire polarization mode directionality determined by the position of the sun.

6. The biomimetic polarized light navigation enhancement method for extreme weather as described in claim 5, characterized in that, Also includes: In the physical model of the polarized light field, the coefficients of the attenuation and depolarization effects and the intensity of the source effect are used as learnable parameters of the model. The learnable parameters are combined with the weights of the physical information neural network itself during the training process for optimization.

7. A biomimetic polarized light navigation enhancement system for extreme weather, based on the biomimetic polarized light navigation enhancement method for extreme weather as described in any one of claims 1 to 6, characterized in that, include: The data acquisition and preprocessing module is configured to acquire raw sky polarization data and simultaneously acquire auxiliary navigation information. Based on a preset failure criterion, it identifies invalid parts in the raw sky polarization data and generates a mask image that identifies the valid and invalid areas of the data. The polarization mode prediction and repair module is configured to input the original sky polarization data, auxiliary navigation information, and mask image into a pre-trained physical information neural network. Based on the polarization light field physical model inherent in the physical information neural network, it performs information prediction on the invalid region and outputs a predicted polarization mode map by combining the original data of the valid region. The heading information calculation module is configured to calculate the heading information of the carrier based on the predicted polarization mode diagram.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images