Visibility inversion method and system based on multi-parameter fusion of light quantum radar

Abstract

The invention relates to the technical field of photoelectric detection and remote sensing, and discloses a visibility inversion method and system based on multi-parameter fusion of a light quantum radar. The method comprises the following steps: synchronously acquiring arrival time, spatial position, polarization state and wavefront phase information of echo photons through single photon detection, respectively extracting an intensity flicker index, a beam drift feature, a polarization evolution parameter and a Zernike low-order phase distortion coefficient, constructing a multi-parameter fusion feature vector, and inputting the multi-parameter fusion feature vector into a pre-trained encoder-decoder neural network model to obtain an encoder-decoder neural network model; and high-precision visibility inversion is realized. The system comprises a single photon detection module, a multi-dimensional feature extraction module and a visibility inversion module. According to the method, turbulence interference and real visibility change are separated, the inversion error is small under strong turbulence, and the method is superior to a traditional method.

Description

Technical Field

[0001] This invention belongs to the field of photoelectric detection and remote sensing technology, specifically relating to a visibility inversion method and system for multi-parameter fusion of quantum radar. Background Technology

[0002] With the increasing demand for slant-path atmospheric visibility detection in aviation, aerospace, and high-precision meteorological observations, traditional visibility inversion methods based on light intensity attenuation models face challenges. These methods typically assume a homogeneous atmospheric medium and a stable transmission path, estimating the extinction coefficient by measuring the ratio of the intensity of the emitted and received light signals, and then deriving visibility. However, in real slant-path conditions, atmospheric turbulence is prevalent, and the resulting random fluctuations in refractive index cause intensity flicker, phase distortion, and spatial drift in the light beam, severely disrupting the monotonic relationship between light intensity and path attenuation. Under conditions of long distances, low elevation angles, or high turbulence intensity, light intensity fluctuations can mask the true attenuation signal dominated by aerosol scattering, leading to deviations or even complete failure of the visibility inversion results.

[0003] Quantum radar, as an emerging active remote sensing technology, utilizes the unique properties of quantum light sources to explore detection capabilities beyond classical limits. Compared to traditional lidar, which relies on the intensity or phase information of classical light fields, quantum radar can extract deep features of environmental disturbances through the non-local correlation characteristics of entangled photon pairs. Time-frequency entangled photon pairs, due to their strong quantum correlation in both the time and frequency domains, can theoretically simultaneously encode the time delay information of path transmission and the frequency mismatch information caused by medium disturbances, providing a physical basis for multi-parameter joint sensing.

[0004] In existing technologies, both classical lidar and nascent quantum detection systems struggle to achieve simultaneous, decoupled inversion of visibility and turbulence intensity in slant-path turbulent environments. Conventional equipment only collects single intensity or echo time information, failing to distinguish the mixed effects of aerosol attenuation and turbulent scattering. Even with the introduction of partial polarization or coherence measurements, there is still a lack of ability to quantitatively characterize the quantum correlation degradation caused by turbulence. Currently, there is no effective algorithm to separate the photon time broadening parameter, which independently reflects visibility, and the correlation distortion parameter, which characterizes the path integral turbulence intensity, from the coincidence count and frequency correlation spectrum of entangled photons. In slant-path detection scenarios with strong turbulence interference, a novel inversion framework integrating quantum optics and atmospheric physics is needed to overcome the problems of parameter confusion, insufficient sensitivity, and poor environmental adaptability of traditional methods. Summary of the Invention

[0005] This invention provides a multi-parameter fusion visibility inversion method and system for quantum radar, aiming to solve the technical problem that atmospheric turbulence-induced light intensity scintillation and beam drift interfere with intensity-attenuation-based visibility measurements in slant-path visibility detection in aviation and aerospace applications, and that existing equipment cannot simultaneously and accurately quantify the effects of turbulence. This invention constructs a multi-dimensional quantum optical observation system that integrates photon arrival time distribution, photon counting statistics, polarization state evolution characteristics, and wavefront phase perturbation information, combined with a coupled physical model of turbulence intensity and visibility, to achieve high-precision, interference-resistant inversion of slant-path visibility.

[0006] This invention provides a visibility inversion method based on multi-parameter fusion of optical quantum radar, comprising: The signal of photons backscattered from the atmospheric slant path is received by a single-photon avalanche diode array, and the arrival timestamp, spatial coordinates, polarization state information and wavefront phase information of each photon are recorded simultaneously. Based on the photon arrival timestamp, a range-gated photon counting sequence segmented along the slant path is constructed, and a Poisson distribution is fitted to the photon counts within each range gate. The ratio of the photon count variance to the mean is extracted as the intensity scintillation index. Based on the spatial coordinates, the instantaneous offset of the beam centroid on the receiving plane is calculated, and the power spectral density analysis is performed on the offset sequence within a continuous time window to extract the energy proportion of the main frequency band as the beam drift characteristic parameter. Based on the polarization state information, the polarization ellipse parameters of the echo photons are reconstructed using Stokes parameters, and the polarization degree attenuation rate and polarization azimuth angle rotation are calculated as sensitive indicators of aerosol particle shape and concentration distribution. Based on the wavefront phase information, the phase distortion caused by atmospheric turbulence is reconstructed using a Shaker-Hartmann wavefront sensor, and the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial are calculated as a quantitative representation of the atmospheric coherence length. The intensity scintillation index, beam drift characteristic parameters, polarization attenuation rate, polarization azimuth angle rotation, and Zernike low-order coefficient root mean square value are used together to form a multi-parameter fused feature vector. The multi-parameter fused feature vector is input into a pre-trained visibility inversion neural network model, which consists of an encoder-decoder structure. The encoder is a hybrid network of multi-head attention mechanism and temporal convolution, and the decoder is a fully connected regression layer. The output is the visibility value corresponding to each distance gate on the slant path.

[0007] Preferably, based on the photon arrival timestamp, a range-gated photon counting sequence segmented along the slant path is constructed, and a Poisson distribution is fitted to the photon counts within each range gate. The ratio of the photon count variance to the mean is extracted as the intensity scintillation index, including: Based on the laser pulse emission time, according to the time interval Nanoseconds are used to divide time windows; The number of photons received within each distance gate is accumulated to form a photon counting sequence; For 100 consecutive laser pulse cycles, the mean and variance of the photon count within each distance gate are calculated, and the ratio of the variance to the mean is defined as the intensity scintillation index.

[0008] Preferably, based on the spatial coordinates, the instantaneous offset of the beam centroid on the receiving plane is calculated, and power spectral density analysis is performed on the offset sequence within a continuous time window to extract the main frequency band energy proportion as a beam drift characteristic parameter, including: Let the first The spatial coordinates of each detection unit are At any given moment The number of photons received is The total number of effective detection units is The instantaneous offset of the beam's centroid in the x-direction Instantaneous offset in the y-direction ; The offset sequence was subjected to power spectral density analysis using the Welch method, with a Hanning window function, a window length of 1 second, and an overlap rate of 50%. Define the main frequency band and calculate the proportion of the power spectrum integral within the main frequency band to the total power spectrum integral, which is used as the beam drift characteristic parameter.

[0009] Preferably, based on the polarization state information, the polarization ellipse parameters of the echo photon are reconstructed using Stokes parameters, and the polarization degree attenuation rate and polarization azimuth angle rotation are calculated, including: Photon counting intensity by horizontal polarization component Photon counting intensity of the vertical polarization component and the intensity of the +45 degree polarization component. With the intensity of the -45 degree polarization component right-hand circular polarization component intensity With the intensity of the left-hand circular polarization component Reconstructing Stokes parameters: Total light intensity Difference in intensity between horizontal and vertical polarization components ±45 degree polarization component intensity difference Intensity difference between right-handed and left-handed circular polarization components ; Calculate the polarization attenuation rate ; Calculate the polarization azimuth rotation amount .

[0010] Preferably, based on the wavefront phase information, a Shaker-Hartmann wavefront sensor is used to reconstruct the phase distortion caused by atmospheric turbulence, and the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial are calculated, including: Based on the local wavefront slope output by the Shaker-Hartmann sensor, the coefficients of the first 20 Zernike polynomials were fitted using the least squares method. Extracting the second to fifth order Zernike coefficients , , , These correspond to defocus, astigmatism, and coma, respectively. Calculate the root mean square value , For the first Zernike coefficient of order.

[0011] Preferably, the intensity scintillation index, beam drift characteristic parameters, polarization attenuation rate, polarization azimuth angle rotation, and the root mean square value of the Zernike low-order coefficients are used to construct a multi-parameter fused feature vector, including: The five feature parameters are arranged in the order of the distance gate to form a five-dimensional feature vector; The entire slant path forms a feature matrix with a length of 10000 and a dimension of 5, which serves as a multi-parameter fusion feature vector.

[0012] Preferably, the multi-parameter fused feature vector is input into a pre-trained visibility inversion neural network model. This visibility inversion neural network model consists of an encoder-decoder structure, wherein the encoder is a hybrid network of multi-head attention mechanism and temporal convolution, the decoder is a fully connected regression layer, and the output is the visibility value corresponding to each distance gate on the slant path, including: The encoder consists of four layers of alternating stacked multi-head self-attention layers and temporal convolutional layers. The activation function is a linear function, and the output dimension is equal to the number of distance gates.

[0013] Preferably, the training process of the visibility inversion neural network model includes: Slant path echo data with known visibility standard values ​​under different meteorological conditions are collected, and the corresponding multi-parameter fusion feature vectors are extracted as input, with the standard visibility value as the supervision label. End-to-end training is performed using the mean squared error loss function. Training continues until the validation set loss shows no decrease for 10 consecutive rounds.

[0014] Preferably, the Shaker-Hartmann microlens array is located on the receiving focal plane, and the microlens units correspond one-to-one with the single-photon detection units for sampling the local slope of the wavefront; Each microlens focuses a local wavefront onto a four-quadrant photodiode below it. By comparing the light intensity differences in the four quadrants, the wavefront slope of the local region is calculated, and the entire wavefront phase distribution is reconstructed.

[0015] This invention also provides a visibility inversion system for multi-parameter fusion of quantum radar, comprising: The single-photon detection module is used to receive the echo photon signal backscattered through the atmospheric slant path and simultaneously output the arrival timestamp, spatial coordinates, polarization state information and wavefront phase information of each photon. The intensity scintillation feature extraction module is used to construct a distance-gated photon counting sequence based on the photon arrival timestamp, and calculate the ratio of the variance to the mean of the photon counts within each distance gate to generate an intensity scintillation index. The beam drift feature extraction module is used to calculate the instantaneous offset of the beam centroid based on the spatial position coordinates, perform power spectral density analysis on the offset sequence, and extract the main frequency band energy ratio as the beam drift feature parameter. The polarization evolution feature extraction module is used to reconstruct Stokes parameters based on the polarization state information and calculate the polarization degree attenuation rate and polarization azimuth angle rotation. The wavefront phase perturbation feature extraction module is used to reconstruct atmospheric turbulence phase distortion based on the wavefront phase information and calculate the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial. The multi-parameter fusion module is used to combine the intensity scintillation index, beam drift characteristic parameters, polarization degree attenuation rate, polarization azimuth angle rotation amount, and Zernike low-order coefficient root mean square value into a multi-parameter fusion feature vector. The visibility inversion module is used to input the multi-parameter fused feature vector into a pre-trained visibility inversion neural network model and output the visibility value corresponding to each distance gate on the slant path. The visibility inversion neural network model includes an encoder and a decoder. The encoder is composed of alternating stacks of multi-head self-attention layers and temporal convolutional layers, used to capture the nonlinear coupling relationship between multiple parameters and temporal dynamic characteristics. The decoder consists of three fully connected layers with linear activation functions and an output dimension equal to the number of distance gates.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention achieves joint quantitative characterization of intensity scintillation, beam drift, polarization depolarization, and wavefront distortion caused by atmospheric turbulence by synchronously acquiring four-dimensional information of the arrival time, spatial location, polarization state, and wavefront phase of the photon-level echo signal. 2. This invention constructs a multi-parameter fusion feature vector that includes turbulence intensity and aerosol scattering characteristics, breaking through the limitations of traditional visibility retrieval based solely on light intensity attenuation. 3. The encoder-decoder neural network model used in this invention can automatically learn the nonlinear coupling mapping relationship between turbulence and visibility without relying on empirical extinction coefficient assumptions; 4. In slant path detection, this invention can separate turbulence interference from actual visibility changes, thereby improving inversion accuracy and robustness. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the coupled modeling of multi-parameter fusion feature vectors and visibility inversion neural networks in this invention; Figure 3 This is a flowchart illustrating the logical process of synchronous acquisition and preprocessing of four-dimensional information of photon-level echo signals in this invention. Figure 4 This is a diagram of the multi-channel parallel processing logic framework for extracting intensity scintillation index, beam drift characteristics, polarization evolution parameters, and wavefront phase perturbation characteristics in this invention. Figure 5 This is a schematic diagram of the internal principle framework of the visibility inversion neural network encoder-decoder structure and its multi-head attention and temporal convolution hybrid mechanism in this invention; Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the single-photon detection module and the atmospheric slant path in this invention. Detailed Implementation

[0018] refer to Figures 1 to 6 This invention provides a multi-parameter fusion visibility inversion method and system for quantum radar, aiming to solve the technical challenge of severe interference caused by atmospheric turbulence-induced light intensity scintillation and beam drift in visibility detection along slant paths in aviation and aerospace, which hinders intensity attenuation-based visibility measurements, and the difficulty of simultaneously and accurately quantifying the effects of turbulence in existing equipment. By constructing a multi-dimensional quantum optical observation system that integrates photon arrival time distribution, photon counting statistics, polarization state evolution characteristics, and wavefront phase perturbation information, and combining this with a coupled physical model of turbulence intensity and visibility, high-precision and interference-resistant inversion of slant path visibility is achieved.

[0019] The method includes the following steps: S1 receives the backscattered photon signal through the atmospheric slant path via a single-photon avalanche diode array, and simultaneously records the arrival timestamp, spatial coordinates, polarization state information, and wavefront phase information of each photon. S2, based on the photon arrival timestamp, construct a range-gated photon counting sequence segmented along the slant path, and fit the photon count within each range gate to a Poisson distribution, extracting the ratio of the photon count variance to the mean as the intensity scintillation index. S3. Based on the spatial position coordinates, calculate the instantaneous offset of the beam centroid on the receiving plane, perform power spectral density analysis on the offset sequence within a continuous time window, and extract the main frequency band energy ratio as the beam drift characteristic parameter. S4. Based on the polarization state information, the polarization ellipse parameters of the echo photon are reconstructed using Stokes parameters, and the polarization degree attenuation rate and polarization azimuth angle rotation are calculated as sensitive indicators of aerosol particle shape and concentration distribution. S5. Based on the wavefront phase information, the phase distortion caused by atmospheric turbulence is reconstructed using a Shaker-Hartmann wavefront sensor, and the root mean square values ​​of the coefficients of the second to fifth orders of the low-order terms of the Zernike polynomial are calculated as a quantitative representation of the atmospheric coherence length. S6, the intensity scintillation index, beam drift characteristic parameters, polarization degree attenuation rate, polarization azimuth angle rotation amount, and Zernike low-order coefficient root mean square value are combined to form a multi-parameter fusion feature vector. S7. Input the multi-parameter fused feature vector into the pre-trained visibility inversion neural network model. The visibility inversion neural network model consists of an encoder-decoder structure. The encoder is a hybrid network of multi-head attention mechanism and temporal convolution, and the decoder is a fully connected regression layer. The output is the visibility value corresponding to each distance gate on the slant path. S8, the training process of the visibility inversion neural network model includes: collecting slant path echo data of known visibility standard values ​​under different meteorological conditions, extracting the corresponding multi-parameter fusion feature vector as input, using the standard visibility value as the supervision label, and performing end-to-end training using the mean square error loss function until the validation set loss converges.

[0020] In the above method, the specific implementation process of step S1 is as follows: The transmitting unit uses a single-frequency laser with a pulse width of 500 picoseconds, a repetition frequency of 10 kHz, and a center wavelength of 1550 nm. After beam expansion and collimation, a probe beam with a divergence angle of 0.5 milliradians is formed. This probe beam is incident on the atmospheric medium along a slant path, and undergoes Mie scattering and Rayleigh scattering at different altitudes. Some backscattered photons return to the receiving end. The receiving unit consists of a Cassegrain telescope with a focal length of 300 mm, a quarter-wave plate, a polarizing beam splitter prism, two single-photon avalanche diode arrays, and a Shaker-Hartmann microlens array. The echo light is first focused to the focal plane by the Cassegrain telescope, and then passes through a polarization analysis optical path composed of a quarter-wave plate and a polarizing beam splitter prism, which decomposes the incident light into horizontal and vertical polarization components, which are received by two independent single-photon avalanche diode arrays respectively. Each single-photon avalanche diode array contains 3200 detector units arranged in 32 rows and 32 columns, with each unit measuring 200 μm × 200 μm. Upon receiving a single photon, each detector unit outputs an electrical pulse signal with a precise timestamp, achieving a time resolution better than 100 picoseconds. Simultaneously, a Shaker-Hartmann microlens array, located on the same focal plane, has microlens units corresponding one-to-one with the single-photon detector units, used to sample the local wavefront slope. Each microlens focuses the local wavefront onto a four-quadrant photodiode below it. By comparing the intensity differences across the four quadrants, the wavefront slope of the local region is calculated, thereby reconstructing the entire wavefront phase distribution. The timestamps, spatial coordinates, polarization channel identifiers, and wavefront slope data of all detector units are synchronously recorded via a high-speed data acquisition card at a sampling frequency of 10 kHz, ensuring complete capture of information from all echo photons within each laser pulse cycle.

[0021] In step S2, the range-gated photon counting sequence is constructed based on the laser pulse emission time, dividing the time window into 1-nanosecond intervals, corresponding to a spatial resolution of 15 centimeters and a maximum detection range of 15 kilometers, resulting in 10,000 range gates. The number of photons received within each range gate is accumulated, forming a photon counting sequence of length 10,000. This photon counting sequence is processed using a sliding window over 100 consecutive laser pulse cycles, calculating the mean and variance of the photon count within each range gate. Since single-photon detection conforms to Poisson statistical properties, under ideal conditions without turbulence interference, the photon count variance should equal the mean. However, atmospheric turbulence causes light intensity scintillation, making the actual variance greater than the mean. Therefore, an intensity scintillation index is defined. The ratio of variance to mean is expressed mathematically as follows: ; For the first The variance of the photon count within the distance gate, For the first The average photon count within a range gate. The intensity scintillation index directly reflects the modulation intensity of light intensity by atmospheric turbulence within the range. The larger the value, the stronger the turbulence, and the greater the error in the traditional visibility inversion based on average light intensity.

[0022] In step S3, the instantaneous offset of the beam centroid is calculated based on the spatial coordinates and photon counts of all effective detector units. Let the first... The spatial coordinates of each detection unit are At any given moment The number of photons received is The total number of effective detection units is The instantaneous offset of the beam's centroid in the x-direction Instantaneous offset in the y-direction ; The instantaneous centroid offset sequence of the beam was sampled at a frequency of 10 kHz to form a continuous time series. Power spectral density analysis was performed on this time series using the Welch method with a Hanning window function, a window length of 1 second, and an overlap rate of 50%. Power spectral density reflects the energy distribution of beam drift at different frequencies. Beam drift caused by atmospheric turbulence is mainly concentrated in the low-frequency band; the dominant frequency band was defined as 0.1 Hz to 10 Hz. The proportion of the power spectral integral within this dominant frequency band to the total power spectral integral was calculated as a beam drift characteristic parameter. This beam drift characteristic parameter quantifies the degree to which turbulence disrupts beam pointing stability and is an indicator for evaluating the optical transmission quality of the slant path.

[0023] In step S4, the polarization state information is processed through Stokes parametric reconstruction. This is achieved by counting the photon intensity of the horizontal polarization component. Photon counting intensity of the vertical polarization component and the intensity of the +45 degree polarization component. With the intensity of the -45 degree polarization component right-hand circular polarization component intensity With the intensity of the left-hand circular polarization component Reconstructing Stokes parameters: Total light intensity Difference in intensity between horizontal and vertical polarization components ±45 degree polarization component intensity difference Intensity difference between right-handed and left-handed circular polarization components ; Calculate the polarization attenuation rate ; Calculate the polarization azimuth rotation amount .

[0024] Non-spherical aerosol particles in the atmosphere can cause depolarization and rotation of the polarization state of echo light, leading to a decrease in DOP and Changes in these parameters are directly related to the concentration, particle size distribution, and shape factor of aerosols, providing independent physical constraints for visibility inversion in addition to intensity.

[0025] In step S5, the wavefront phase information is reconstructed from the local wavefront slope data output by the Shaker-Hartmann sensor. The first 20 Zernike polynomial coefficients are fitted using the least squares method. The second to fifth orders correspond to low-order aberrations such as defocus, astigmatism, and coma, which are mainly caused by atmospheric turbulence. The root mean square values ​​of these four coefficients are calculated: ; For the first Zernike coefficient of order.

[0026] Value and atmospheric coherence length It is inversely proportional and is a direct optical parameter for quantifying turbulence intensity.

[0027] In step S6, the five characteristic parameters mentioned above—intensity scintillation index, beam drift characteristic parameter, polarization attenuation rate, polarization azimuth angle rotation, and Zernike low-order coefficient root mean square value—are arranged in range gate order to form a five-dimensional feature vector. Each range gate corresponds to a feature vector, and the entire slant path forms a feature matrix with a length of 10000 and a dimension of 5.

[0028] In step S7, the feature matrix is ​​input into a pre-trained visibility inversion neural network model. The encoder of this model consists of four alternately stacked multi-head self-attention layers and temporal convolutional layers. The multi-head self-attention mechanism has 8 heads and is used to capture the nonlinear coupling relationships between different feature dimensions; the temporal convolutional layers have a kernel size of 7, a stride of 1, and 128 channels, used to extract local temporal dynamic characteristics along the distance gate direction. The high-dimensional feature representation output by the encoder is fed into the decoder, which consists of three fully connected layers with 256, 128, and 1 neurons respectively. The activation function is a linear function, and the final output dimension is 10000, representing the visibility value corresponding to each distance gate, in meters.

[0029] In step S8, the model training uses a real-world dataset. The system of this invention and standard visibility meters, such as forward scattering meters or transmissometers, are simultaneously deployed in typical scenarios such as airport runways and high-altitude observation stations to collect echo data under different weather conditions, including clear skies, fog, haze, rain, snow, and varying turbulence intensities. Each set of data includes raw photon event records and corresponding reference visibility values. Steps S1 to S6 are performed on each set of data to generate a multi-parameter fused feature vector as input, with the reference visibility value as the label. A batch size of 32, a learning rate of 0.001, an AdamW optimizer, a weight decay coefficient of 0.001, and a mean squared error loss function are used. Training continues until the validation set loss shows no decrease for 10 consecutive epochs, at which point the model converges.

[0030] The system includes a single-photon detection module, an intensity scintillation feature extraction module, a beam drift feature extraction module, a polarization evolution feature extraction module, a wavefront phase perturbation feature extraction module, a multi-parameter fusion module, and a visibility inversion module. The single-photon detection module, as described above, performs synchronous acquisition of four-dimensional information. The intensity scintillation feature extraction module receives timestamp data streams and performs range gate partitioning, photon count accumulation, sliding window statistics, and intensity scintillation index calculation. The beam drift feature extraction module receives spatial position coordinates and photon counts, calculates the beam centroid offset in real time, and uses a fast Fourier transform library for power spectral density analysis, outputting the energy percentage of the main frequency band. The polarization evolution feature extraction module receives photon count sequences from two polarization channels, switches the polarization optical element states according to a preset time sequence, reconstructs the Stokes parameters, and calculates the DOP and... The wavefront phase perturbation feature extraction module receives slope data from the Shaker-Hartmann sensor, runs the Zernike fitting algorithm, and outputs the root mean square values ​​of low-order coefficients. The multi-parameter fusion module aligns the outputs of the above five modules according to the distance gate and concatenates them into a feature vector. The visibility inversion module loads the trained neural network model, performs forward inference, and outputs a high-resolution visibility profile.

[0031] During operation, all modules of the system interact with each other via shared memory or a high-speed bus, ensuring that the processing latency is less than the laser pulse period. Anomaly handling mechanisms include: triggering an automatic calibration process when the dark count rate of the single-photon detection module exceeds a threshold; filling data with nearest-neighbor distance gating or the average of historical data when the feature extraction module detects missing or outlier values; and initiating post-processing correction based on a physical model when the visibility value output by the neural network exceeds a physically reasonable range, ensuring the reliability of the results.

[0032] In summary, this embodiment improves the accuracy and robustness of slant path visibility inversion by synchronously acquiring photon-level four-dimensional information, constructing multi-parameter fusion features, and using deep neural networks to learn the complex mapping relationship between turbulence and visibility, separating turbulence interference from actual visibility changes.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A visibility inversion method based on multi-parameter fusion in quantum radar, characterized in that, include: The signal of photons backscattered from the atmospheric slant path is received by a single-photon avalanche diode array, and the arrival timestamp, spatial coordinates, polarization state information and wavefront phase information of each photon are recorded simultaneously. Based on the photon arrival timestamp, a range-gated photon counting sequence segmented along the slant path is constructed, and a Poisson distribution is fitted to the photon counts within each range gate. The ratio of the photon count variance to the mean is extracted as the intensity scintillation index. Based on the spatial coordinates, the instantaneous offset of the beam centroid on the receiving plane is calculated, and the power spectral density analysis is performed on the offset sequence within a continuous time window to extract the energy proportion of the main frequency band as the beam drift characteristic parameter. Based on the polarization state information, the polarization ellipse parameters of the echo photons are reconstructed using Stokes parameters, and the polarization degree attenuation rate and polarization azimuth angle rotation are calculated as sensitive indicators of aerosol particle shape and concentration distribution. Based on the wavefront phase information, the phase distortion caused by atmospheric turbulence is reconstructed using a Shaker-Hartmann wavefront sensor, and the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial are calculated as a quantitative representation of the atmospheric coherence length. The intensity scintillation index, beam drift characteristic parameters, polarization attenuation rate, polarization azimuth angle rotation, and Zernike low-order coefficient root mean square value are used together to form a multi-parameter fused feature vector. The multi-parameter fused feature vector is input into a pre-trained visibility inversion neural network model. This visibility inversion neural network model consists of an encoder-decoder structure. The encoder is a hybrid network of multi-head attention mechanism and temporal convolution, and the decoder is a fully connected regression layer. The output is the visibility value corresponding to each distance gate on the slant path.

2. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 1, characterized in that, Based on the photon arrival timestamp, a range-gated photon counting sequence segmented along the slant path is constructed, and a Poisson distribution is fitted to the photon counts within each range gate. The ratio of the photon count variance to the mean is extracted as the intensity scintillation index, including: Based on the laser pulse emission time, according to the time interval Nanoseconds are used to divide time windows; The number of photons received within each distance gate is accumulated to form a photon counting sequence; For 100 consecutive laser pulse cycles, the mean and variance of the photon count within each distance gate are calculated, and the ratio of the variance to the mean is defined as the intensity scintillation index.

3. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 2, characterized in that, Based on the spatial coordinates, the instantaneous offset of the beam centroid on the receiving plane is calculated. Power spectral density analysis is performed on the offset sequence within a continuous time window, and the energy proportion of the dominant frequency band is extracted as a beam drift characteristic parameter, including: Let the first The spatial coordinates of each detection unit are At any given moment The number of photons received is The total number of effective detection units is The instantaneous offset of the beam's centroid in the x-direction Instantaneous offset in the y-direction ; The offset sequence was subjected to power spectral density analysis using the Welch method, with a Hanning window function, a window length of 1 second, and an overlap rate of 50%. Define the main frequency band and calculate the proportion of the power spectrum integral within the main frequency band to the total power spectrum integral, which is used as the beam drift characteristic parameter.

4. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 3, characterized in that, Based on the polarization state information, the polarization ellipse parameters of the echo photon are reconstructed using Stokes parameters, and the polarization degree attenuation rate and polarization azimuth angle rotation are calculated, including: Photon counting intensity by horizontal polarization component Photon counting intensity of the vertical polarization component and the intensity of the +45 degree polarization component. With the intensity of the -45 degree polarization component right-hand circular polarization component intensity With the intensity of the left-hand circular polarization component Reconstructing Stokes parameters: Total light intensity Difference in intensity between horizontal and vertical polarization components ±45 degree polarization component intensity difference Intensity difference between right-handed and left-handed circular polarization components ; Calculate the polarization attenuation rate ; Calculate the polarization azimuth rotation amount .

5. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 4, characterized in that, Based on the wavefront phase information, the phase distortion caused by atmospheric turbulence is reconstructed using a Shaker-Hartmann wavefront sensor, and the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial are calculated, including: Based on the local wavefront slope output by the Shaker-Hartmann sensor, the coefficients of the first 20 Zernike polynomials were fitted using the least squares method. Extracting the second to fifth order Zernike coefficients , , , These correspond to defocus, astigmatism, and coma, respectively. Calculate the root mean square value , For the first Zernike coefficient of order.

6. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 5, characterized in that, The intensity scintillation index, beam drift characteristic parameters, polarization attenuation rate, polarization azimuth angle rotation, and the root mean square value of the Zernike low-order coefficients are combined to form a multi-parameter fused feature vector, including: The five feature parameters are arranged in the order of the distance gate to form a five-dimensional feature vector; The entire slant path forms a feature matrix with a length of 10000 and a dimension of 5, which serves as a multi-parameter fusion feature vector.

7. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 6, characterized in that, The multi-parameter fused feature vector is input into a pre-trained visibility inversion neural network model. This visibility inversion neural network model consists of an encoder-decoder structure. The encoder is a hybrid network of multi-head attention mechanism and temporal convolution, and the decoder is a fully connected regression layer. The output is the visibility value corresponding to each distance gate on the slant path, including: The encoder consists of four layers of alternating stacked multi-head self-attention layers and temporal convolutional layers. The activation function is a linear function, and the output dimension is equal to the number of distance gates.

8. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 7, characterized in that, The training process of the visibility inversion neural network model includes: Slant path echo data with known visibility standard values ​​under different meteorological conditions are collected, and the corresponding multi-parameter fusion feature vectors are extracted as input, with the standard visibility value as the supervision label. End-to-end training is performed using the mean squared error loss function. Training continues until the validation set loss shows no decrease for 10 consecutive rounds.

9. The visibility inversion method for multi-parameter fusion of quantum radar according to claim 8, characterized in that, The Shaker-Hartmann microlens array is located on the receiving focal plane, and the microlens unit corresponds one-to-one with the single-photon detection unit to sample the local slope of the wavefront. Each microlens focuses a local wavefront onto a four-quadrant photodiode below it. By comparing the light intensity differences in the four quadrants, the wavefront slope of the local region is calculated, and the entire wavefront phase distribution is reconstructed.

10. A visibility inversion system for multi-parameter fusion in quantum radar, characterized in that, include: The single-photon detection module is used to receive the echo photon signal backscattered through the atmospheric slant path and simultaneously output the arrival timestamp, spatial coordinates, polarization state information and wavefront phase information of each photon. The intensity scintillation feature extraction module is used to construct a distance-gated photon counting sequence based on the photon arrival timestamp, and calculate the ratio of the variance to the mean of the photon counts within each distance gate to generate an intensity scintillation index. The beam drift feature extraction module is used to calculate the instantaneous offset of the beam centroid based on the spatial position coordinates, perform power spectral density analysis on the offset sequence, and extract the main frequency band energy ratio as the beam drift feature parameter. The polarization evolution feature extraction module is used to reconstruct Stokes parameters based on the polarization state information and calculate the polarization degree attenuation rate and polarization azimuth angle rotation. The wavefront phase perturbation feature extraction module is used to reconstruct atmospheric turbulence phase distortion based on the wavefront phase information and calculate the root mean square values ​​of the second to fifth order coefficients of the Zernike polynomial. The multi-parameter fusion module is used to combine the intensity scintillation index, beam drift characteristic parameters, polarization degree attenuation rate, polarization azimuth angle rotation amount, and Zernike low-order coefficient root mean square value into a multi-parameter fusion feature vector. The visibility inversion module is used to input the multi-parameter fused feature vector into a pre-trained visibility inversion neural network model and output the visibility value corresponding to each distance gate on the slant path. The visibility inversion neural network model includes an encoder and a decoder. The encoder is composed of alternating stacks of multi-head self-attention layers and temporal convolutional layers, used to capture the nonlinear coupling relationship between multiple parameters and temporal dynamic characteristics. The decoder consists of three fully connected layers with linear activation functions and an output dimension equal to the number of distance gates.

Images