A method and system for intelligent identification of pollution sources using quantum radar

Abstract

This invention provides a method and system for intelligent identification of pollution sources using quantum radar. The method includes achieving precise positioning through preprocessing of photonic quantum data frames, construction of a multi-dimensional feature space, and iterative window optimization. Abnormal signals are filtered based on a quantum background noise Poisson statistical model to generate a preprocessed data frame. A quantum feature space is constructed by fusing photon wavelength, polarization state, and arrival time to generate a back-projected probability distribution map. A composite displacement vector is constructed to update window parameters by calculating the first, second, and third central moments and spatial gradient of the probability distribution within the window. When the magnitude of the displacement vector is less than the convergence threshold, the location, diffusion range, and direction of the pollution source are output.

Description

Technical Field

[0001] This application belongs to the field of identification, and in particular relates to an intelligent identification method and system for tracing pollution sources using quantum radar. Background Technology

[0002] In air pollution monitoring, quantum radar, with its wide monitoring range and high accuracy, has been increasingly widely adopted in recent years. However, after acquiring feature data of the monitoring area, the core issue in source tracing and identification is how to automatically and quickly locate the target center from the data. The CamShift algorithm, based on color or feature histograms, is a classic method. This algorithm constructs a feature back-projection probability map of the target and iteratively searches for the region with the highest probability density using the MeanShift concept. It can also adjust the size and orientation of the search window based on the target's second moment, making it suitable for tracking targets with changing morphologies. However, the CamShift algorithm relies on the first moment of the probability distribution within the window when updating the search window position. When the pollution source diffusion pattern is irregular, causing a significant skew or tailing on the probability distribution map, the centroid position may deviate significantly from the probability peak, resulting in positioning errors. In scenarios with low contrast between the target and background or blurred boundaries, relying solely on centroid-driven displacement may lead to slower convergence or getting trapped in local optima, affecting the real-time performance and accuracy of source tracing. Summary of the Invention

[0003] To address the aforementioned problems, in the first aspect, this invention proposes a method for intelligent identification of pollution sources using quantum radar, comprising the following steps:

[0004] Acquire photonic quantum data frames of the monitoring area, perform compliance checks on the signal strength of each data point within the photonic quantum data frames based on a preset quantum background noise Poisson statistical model, and set the strength of data points that do not meet the test to zero to obtain preprocessed data frames;

[0005] A multidimensional quantum feature space is constructed based on the photon wavelength, polarization state and arrival time of the pollution source to be traced. The multidimensional quantum features of each data point in the monitoring area are calculated based on the preprocessed data frame, and a quantum feature back projection probability distribution map of the monitoring area is generated.

[0006] An initial search window is set on the probability distribution map, and the search window is iteratively updated. Each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient; and calculating and updating the size and orientation of the search window according to the second central moment.

[0007] When the magnitude of the composite displacement vector is less than the preset convergence threshold, the search window is determined to have converged, and the center position of the converged window is determined as the location of the pollution source. The size and direction of the converged window are determined as the diffusion range and direction of the pollution source.

[0008] Optionally, the step of performing a conformity check on the signal intensity of each data point within the quantum data frame based on a preset quantum background noise Poisson statistical model, and setting the intensity of data points that do not conform to the check to zero, includes:

[0009] Establish the null hypothesis that the data points belong to background noise;

[0010] Set the significance level to α;

[0011] Calculate the p-value of the signal strength at each data point under the Poisson statistical model;

[0012] When the p-value is less than α, the null hypothesis is rejected and the intensity of the data points is retained.

[0013] When the p value is greater than or equal to α, the null hypothesis is accepted and the intensity of the data point is set to zero.

[0014] Optionally, the construction of a multidimensional quantum feature space based on the photon wavelength, polarization state, and arrival time of the pollution source to be traced includes:

[0015] A three-dimensional Cartesian coordinate system is constructed with the photon wavelength λ as the first axis, the polarization state angle θ as the second axis, and the arrival time t as the third axis, as the multidimensional quantum feature space.

[0016] Optionally, the step of calculating the multidimensional quantum features of each data point within the monitoring area based on the preprocessed data frame, and generating a quantum feature back-projection probability distribution map of the monitoring area, includes:

[0017] Using sample data from the pollution source to be traced, a three-dimensional target feature histogram is statistically generated in the multidimensional quantum feature space;

[0018] For each data point in the preprocessed data frame, the multidimensional quantum features are mapped onto the target feature histogram, the corresponding histogram statistics are found and obtained, and the statistics are used as the probability value of the data point in the back-projected probability distribution map.

[0019] Optionally, setting the initial search window on the probability distribution map includes:

[0020] Traverse the probability distribution graph to find the data point with the highest probability value;

[0021] Set the center of the initial search window at the location of the data point;

[0022] Set the initial search window size to a rectangular window that covers a preset proportion β of the total area of ​​the monitored region.

[0023] Optionally, the step of constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient includes:

[0024] Let the centroid displacement vector determined by the first moment be denoted as... The skewness vector determined by the third central moments is denoted as... And the spatial gradient vector is denoted as ;

[0025] The composite displacement vector V is obtained by weighted summation;

[0026] The updated center position is obtained by adding the composite displacement vector V to the center position of the current search window.

[0027] Optionally, the step of calculating and updating the size and orientation of the search window based on the second-order central moment includes:

[0028] From the second-order central moment , , Construct the covariance matrix;

[0029] Calculate the eigenvalues ​​of the covariance matrix. , and the corresponding feature vector , ;

[0030] Update the orientation of the search window to the feature vector. , The direction it points to;

[0031] Update the major and minor axes of the search window to... and , where C is a preset coefficient.

[0032] Optionally, determining that the search window has converged when the magnitude of the composite displacement vector is less than a preset convergence threshold includes:

[0033] After each iteration update, the magnitude L of the composite displacement vector is calculated;

[0034] When the modulus L is less than the preset convergence threshold When the search window converges, the iteration stops.

[0035] In the second aspect, the present invention proposes a pollution source tracing and intelligent identification system utilizing quantum radar, comprising the following modules:

[0036] The acquisition module is used to acquire the photonic quantum data frame of the monitoring area, perform a conformity test on the signal intensity of each data point in the photonic quantum data frame based on a preset quantum background noise Poisson statistical model, and set the intensity of the data points that do not conform to the test to zero, so as to obtain a preprocessed data frame.

[0037] The generation module is used to construct a multidimensional quantum feature space based on the photon wavelength, polarization state and arrival time of the pollution source to be traced, calculate the multidimensional quantum features of each data point in the monitoring area based on the preprocessed data frame, and generate a quantum feature back projection probability distribution map of the monitoring area.

[0038] A construction module is used to set an initial search window on the probability distribution map and iteratively update the search window. Each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient; and calculating and updating the size and orientation of the search window according to the second central moment.

[0039] The determination module is used to determine the convergence of the search window when the magnitude of the composite displacement vector is less than the preset convergence threshold, and to determine the center position of the converged window as the location of the pollution source, and to determine the size and direction of the converged window as the diffusion range and direction of the pollution source.

[0040] Preferably, the step of performing a conformity check on the signal intensity of each data point within the quantum data frame based on a preset quantum background noise Poisson statistical model, and setting the intensity of data points that do not conform to the check to zero, includes:

[0041] Establish the null hypothesis that the data points belong to background noise;

[0042] Set the significance level to α;

[0043] Calculate the p-value of the signal strength at each data point under the Poisson statistical model;

[0044] When the p-value is less than α, the null hypothesis is rejected and the intensity of the data points is retained.

[0045] When the p value is greater than or equal to α, the null hypothesis is accepted and the intensity of the data point is set to zero.

[0046] Preferably, the construction of a multidimensional quantum feature space based on the photon wavelength, polarization state, and arrival time of the pollution source to be traced includes:

[0047] A three-dimensional Cartesian coordinate system is constructed with the photon wavelength λ as the first axis, the polarization state angle θ as the second axis, and the arrival time t as the third axis, as the multidimensional quantum feature space.

[0048] Preferably, the step of calculating the multidimensional quantum features of each data point within the monitoring area based on the preprocessed data frame and generating a quantum feature back-projection probability distribution map of the monitoring area includes:

[0049] Using sample data from the pollution source to be traced, a three-dimensional target feature histogram is statistically generated in the multidimensional quantum feature space;

[0050] For each data point in the preprocessed data frame, the multidimensional quantum features are mapped onto the target feature histogram, the corresponding histogram statistics are found and obtained, and the statistics are used as the probability value of the data point in the back-projected probability distribution map.

[0051] Preferably, setting the initial search window on the probability distribution map includes:

[0052] Traverse the probability distribution graph to find the data point with the highest probability value;

[0053] Set the center of the initial search window at the location of the data point;

[0054] Set the initial search window size to a rectangular window that covers a preset proportion β of the total area of ​​the monitored region.

[0055] Preferably, the step of constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient includes:

[0056] Let the centroid displacement vector determined by the first moment be denoted as... The skewness vector determined by the third central moments is denoted as... And the spatial gradient vector is denoted as ;

[0057] The composite displacement vector V is obtained by weighted summation;

[0058] The updated center position is obtained by adding the composite displacement vector V to the center position of the current search window.

[0059] Preferably, the step of calculating and updating the size and orientation of the search window based on the second-order central moment includes:

[0060] From the second-order central moment , , Construct the covariance matrix;

[0061] Calculate the eigenvalues ​​of the covariance matrix. , and the corresponding feature vector , ;

[0062] Update the orientation of the search window to the feature vector. , The direction it points to;

[0063] Update the major and minor axes of the search window to... and , where C is a preset coefficient.

[0064] Preferably, determining that the search window has converged when the magnitude of the composite displacement vector is less than a preset convergence threshold includes:

[0065] After each iteration update, the magnitude L of the composite displacement vector is calculated;

[0066] When the modulus L is less than the preset convergence threshold When the search window converges, the iteration stops.

[0067] This invention preprocesses photonic quantum data by constructing a quantum background noise Poisson statistical model, which filters out background noise and improves the signal-to-noise ratio of the target signal. Utilizing multidimensional quantum features constructed from photon wavelength, polarization state, and arrival time enhances the specificity of identifying particular pollution sources and improves anti-interference capabilities. During the search window update process, the third-order central moment representing the degree of distribution skew and the spatial gradient indicating the direction of probability growth are input into the iterative calculation of the center position. This reliably corrects the centroid deviation from the peak value caused by the irregular diffusion pattern of the pollution source and directly points the search direction to the point of highest probability, thereby improving the accuracy of pollution source localization and accelerating the convergence speed. This enables rapid and reliable determination of the pollution source location, diffusion range, and direction. Attached Figure Description

[0068] Figure 1 A flowchart of the first embodiment;

[0069] Figure 2 This is a schematic diagram illustrating the Poisson statistical compliance test for quantum background noise. Detailed Implementation

[0070] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0071] In the first embodiment, the present invention proposes a method for intelligent identification of pollution sources using quantum radar, such as... Figure 1 This includes the following steps:

[0072] S1, acquire the photonic quantum data frame of the monitoring area, perform a conformity test on the signal intensity of each data point in the photonic quantum data frame based on the preset quantum background noise Poisson statistical model, and set the intensity of the data points that do not conform to the test to zero to obtain the preprocessed data frame;

[0073] A photon count image of the monitored area is acquired using a single-photon detector array as a quantum data frame, where the intensity value of each pixel represents the number of photons received at that location. The preset quantum background noise Poisson statistical model is as follows: , where parameters The average photon count for the background region is obtained through statistical measurements in a contaminated reference region. The conformity test involves setting a confidence level, such as 99.9%, and applying the Poisson cumulative distribution function. Calculate the corresponding photon number threshold Iterate through each data point in the quantum data frame. If the photon count value of that data point is less than or equal to... If the value is greater than 0, then the point is determined to be background noise, and the intensity value is set to 0; if it is greater than 0, then the point is determined to be background noise. If the signal is positive, it is identified as a signal, and the original intensity value is retained. This is then used to generate a preprocessed data frame, such as... Figure 2 .

[0074] In some embodiments, the step of performing a conformity check on the signal intensity of each data point within the quantum data frame based on a preset quantum background noise Poisson statistical model, and setting the intensity of data points that do not conform to the check to zero to obtain a preprocessed data frame includes:

[0075] Establish the null hypothesis that the data points belong to background noise;

[0076] Set the significance level to α;

[0077] Calculate the p-value of the signal strength at each data point under the Poisson statistical model;

[0078] When the p-value is less than α, the null hypothesis is rejected and the intensity of the data points is retained.

[0079] When the p value is greater than or equal to α, the null hypothesis is accepted and the intensity of the data point is set to zero.

[0080] A statistical model representing quantum background noise is established. The events of background photons arriving at the detector are random and independent, and the count value per unit time or unit area follows a Poisson distribution. By sampling areas within the monitoring region where no pollution source signal is identified, the average photon count rate of the background noise can be estimated. For example, if a total photon count of 50 is obtained by measuring a 100-pixel background area, the background noise model can be set to have a mean of 50. The Poisson distribution is used. Optionally, the present invention is a quantum radar system under low ambient light and weak photon detection conditions. The noise mainly comes from dark counting and single-photon events, and its statistical characteristics can approximately follow a Poisson distribution.

[0081] Hypothesis testing is performed on each data point (pixel) within the quantum data frame. For any pixel, if the measured photon count is k, the null hypothesis is established. The count value k of this pixel comes from the background noise distribution. Set a significance level, for example... The probability of observing k or more photons, i.e., a greater number of photons, under this null hypothesis is calculated using the following formula: If the count value of a pixel is k=5, in If the calculated p-value under the model is much less than 0.01, the null hypothesis is rejected, the point is considered not to be background noise, and its original intensity value is retained as a valid signal. Conversely, if the count value of another pixel is k=1 and the calculated p-value is greater than 0.01, the null hypothesis is accepted, the point is considered to belong to background noise, and its intensity value is set to zero. This process iterates through all data points, generating a preprocessed data frame that has passed the noise compliance test.

[0082] S2, construct a multidimensional quantum feature space based on the photon wavelength, polarization state and arrival time of the pollution source to be traced, calculate the multidimensional quantum features of each data point in the monitoring area based on the preprocessed data frame, and generate a quantum feature back projection probability distribution map of the monitoring area;

[0083] Taking sulfur dioxide gas as an example, it emits characteristic fluorescence under specific laser excitation. These fluorescent photons possess specific wavelength ranges, polarization states, and delay times relative to the excitation pulse. A three-dimensional quantum feature space is constructed, with the three dimensions being photon wavelength, polarization angle, and arrival time, respectively. A known pollution source sample area is selected, and the distribution of photons within this area in the aforementioned three-dimensional feature space is statistically analyzed, forming a three-dimensional feature histogram, which is the target model q. For the entire monitoring area, the distribution of all photons in the same feature space is calculated, yielding the image model p. The ratio R = q / p is calculated, reflecting the correlation strength between different feature combinations and the target pollution source. For each pixel in the preprocessed data frame, the features of each photon contained in that data point are substituted into the ratio R to find the corresponding value, and these values ​​are summed or averaged to obtain the probability value of that pixel. The probability values ​​of all pixels together constitute the quantum feature back-projection probability distribution map.

[0084] In some embodiments, constructing a multidimensional quantum feature space based on the photon wavelength, polarization state, and arrival time of the pollution source to be traced includes:

[0085] A three-dimensional Cartesian coordinate system is constructed with the photon wavelength λ as the first axis, the polarization state angle θ as the second axis, and the arrival time t as the third axis, as the multidimensional quantum feature space.

[0086] The space is constructed as a three-dimensional Cartesian coordinate system, defined by three mutually orthogonal coordinate axes. The first axis, the X-axis, represents the wavelength of the photon. The unit is usually nanometers. The second axis, the Y-axis, represents the polarization angle of the photon. The unit is degrees or radians, and it is measured relative to a reference direction. The third axis, the Z-axis, represents the relative time t for a photon to reach the detector, and the unit can be nanoseconds or picoseconds.

[0087] When a photon is detected by the detection system, three key physical parameters are measured simultaneously. For example, when a photon is detected, its wavelength is measured to be 532 nm, its polarization angle to be 60°, and its arrival time to be 15.2 ns relative to the start of the data frame. This photon event is mapped to a unique coordinate point in this three-dimensional quantum feature space. Another detected photon, with parameters of wavelength 405 nm, polarization angle 120°, and arrival time 18.5 ns, is mapped to another point. By continuously measuring and mapping sample photons from a specific pollution source, a point cloud is formed in this three-dimensional space. The distribution shape and density of the point cloud constitute the quantum characteristic fingerprint of the pollution source.

[0088] In some embodiments, the step of calculating the multidimensional quantum features of each data point within the monitoring area based on the preprocessed data frame and generating a quantum feature back-projection probability distribution map of the monitoring area includes:

[0089] Using sample data from the pollution source to be traced, a three-dimensional target feature histogram is statistically generated in the multidimensional quantum feature space;

[0090] For each data point in the preprocessed data frame, the multidimensional quantum features are mapped onto the target feature histogram, the corresponding histogram statistics are found and obtained, and the statistics are used as the probability value of the data point in the back-projected probability distribution map.

[0091] A target feature model, namely a three-dimensional target feature histogram, is established by collecting a large amount of photon sample data from known pollution sources to be traced. The aforementioned three-dimensional quantum feature space is then discretized, divided into small cubic units, or bins, the size of which is determined based on the characteristic bandwidth of the pollution source to be traced. For example, the wavelength axis... Each segment is 10nm long, with polarization axes Each segment is divided into 5° intervals, and the time axis t is divided into 2ns intervals. All sample photon data points are then processed according to... Coordinates are placed into corresponding bins, and the number of photons in each bin is counted. The count value of each bin is divided by the total number of sample photons, and then normalized to obtain the probability value of each bin. The three-dimensional grid containing the probability values ​​is the target feature histogram.

[0092] For each spatial location pixel (x,y) in the preprocessed data frame of the entire monitoring area, extract the quantum features of the photons detected at that location. Using the feature coordinates, the bin to which the feature belongs is located in the pre-constructed 3D target feature histogram. For example, the photon feature of pixel (128, 256) is (535, 62, 16.1), which falls into the bin with a wavelength of 530-540nm, polarization of 60-65°, and time of 16-18ns. The normalized probability value corresponding to this bin is read from the histogram, assumed to be 0.08. Thus, on the generated 2D back-projection probability distribution map, the pixel value at coordinates (128, 256) is assigned a value of 0.08. This process is repeated for all pixels within the monitoring area to form a 2D probability map, where the brightness or value of each pixel represents the degree of matching between the photon feature at that location and the feature of the pollution source to be traced.

[0093] S3, set an initial search window on the probability distribution map, and iteratively update the search window, wherein each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient; and calculating and updating the size and orientation of the search window according to the second central moment.

[0094] Select an initial rectangular search window on the probability distribution plot. In the k-th iteration, calculate the zeroth and first moments of the probability distribution P(x,y) within the window to obtain the centroid position. Calculate the third-order central moments and construct the skewness vector S, with components and... and Proportional to each other, used to indicate the skewness of the probability distribution. Simultaneously, the probability distribution plot is calculated using methods such as the Sobel operator at the center of the current window. gradient vector at P. Construct a composite displacement vector ,in , , The preset weighting coefficients are used. The new window center position is updated to... Calculate the second-order central moments. , , The major axis, minor axis, and orientation angle of the probability distribution ellipse are then determined, and the width, height, and rotation angle of the search window for the next iteration are adjusted based on these parameters.

[0095] In some embodiments, setting an initial search window on the probability distribution graph includes:

[0096] Traverse the probability distribution graph to find the data point with the highest probability value;

[0097] Set the center of the initial search window at the location of the data point;

[0098] Set the initial search window size to a rectangular window that covers a preset proportion β of the total area of ​​the monitored region.

[0099] A global search is performed on the generated two-dimensional back-projected probability distribution map. This probability distribution map can be viewed as a two-dimensional matrix, where each element P(x,y) represents the probability of a spatial location (x,y) matching a pollution source feature. The algorithm examines each element value in the matrix to determine the global maximum value. For example, in a 1024x1024 pixel probability distribution map, after traversing and comparing, the probability value is found to be the maximum, 0.95, at coordinates (520, 610). Therefore, the center position of the initial search window is determined to be... .

[0100] The initial search window size is determined by a preset scaling factor. and the total area of ​​the monitoring area Jointly decided. Assume the total area of ​​the monitoring area is... Pixels, preset ratio That is, the initial window should cover 5% of the total area. Therefore, the area of ​​the initial window is... Should be Pixels. The window can be set as a square window with a side length of 229 pixels. Therefore, the initial search window is set as a rectangular area centered at (520, 610) with a size of approximately 229x229 pixels.

[0101] In some embodiments, constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient includes:

[0102] Let the centroid displacement vector determined by the first moment be denoted as... The skewness vector determined by the third central moments is denoted as... And the spatial gradient vector is denoted as ;

[0103] By weighted summation The composite displacement vector is calculated. ,in , , Preset weighting coefficients;

[0104] Add the composite displacement vector to the center position of the current search window. This yields the updated center position.

[0105] Calculate the probability distribution within the current search window Moment: Zeroth moment First-order moment and Thus, the centroid coordinates are obtained. The centroid displacement vector is the vector pointing from the geometric center of the window to the probabilistic centroid. The second-order central moments are calculated based on the centroid. and and derive the standard deviation. , Calculate the third-order central moments. and To construct the skewness vector The standardized skewness coefficient is multiplied by the characteristic length. , Take half the length of the window's diagonal: , ,in This is used to avoid division by zero. Simultaneously, the 3×3 Sobel operator is used to calculate the center of the window. Gradient at: and horizontal kernel respectively and vertical core Convolution yields gradient components and This forms a gradient vector pointing in the direction of the fastest probability growth. .

[0106] After obtaining the three vectors mentioned above, they are combined into a composite displacement vector by weighted summation. For example, let the preset weight be... The three vectors obtained by calculation are as follows: , , The composite displacement vector is then... The center position is updated by adding the current center position of the search window to the composite displacement vector. If the current center is (520, 610), the new center position is updated to (522, 610.5).

[0107] In some embodiments, calculating and updating the size and orientation of the search window based on the second-order central moment includes:

[0108] From the second-order central moment , , Construct the covariance matrix;

[0109] Calculate the eigenvalues ​​of the covariance matrix. , and the corresponding feature vector , ;

[0110] Update the orientation of the search window to the feature vector. , The direction it points to;

[0111] Update the major and minor axes of the search window to... and , where C is a preset coefficient.

[0112] Within the current search window, based on the probability distribution P(x,y) and the centroid... Calculate the second central moment: ,as well as The three moments mentioned above are used to construct a 2x2 covariance matrix, and the model is as follows: For example, calculations yielded... , , Then the covariance matrix is ​​constructed.

[0113] By solving the characteristic equation, two eigenvalues ​​are obtained. and and their corresponding two orthogonal eigenvectors and The eigenvector represents the main direction of the data distribution, while the eigenvalue represents the variance or dispersion along that direction. Assume that a large eigenvalue is calculated... The corresponding feature vector is The smaller eigenvalue is The corresponding feature vector is Based on the above results, update the geometry of the search window from a rectangle to an ellipse. The direction of the ellipse is determined by the feature vectors, with the principal axis aligned with... Consistent, secondary shaft direction is the same as Consistent. The size of the ellipse is determined by the eigenvalues. If the preset coefficient C=2, the length of the major semi-axis is updated to... The length of the short half-axis has been updated to .

[0114] S4. When the magnitude of the composite displacement vector is less than the preset convergence threshold, the search window is determined to have converged, and the center position of the converged window is determined as the location of the pollution source. The size and direction of the converged window are determined as the diffusion range and direction of the pollution source.

[0115] Specifically, after each iteration of updating the window position, the Euclidean norm of the composite displacement vector ΔP, i.e., the magnitude ||ΔP||, is calculated. This magnitude is compared with a preset convergence threshold ε, for example, ε = 0.5 pixels. If ||ΔP|| is less than ε, the search is considered converged, and the iteration process terminates. The iteration is also forcibly terminated when the number of iterations reaches a preset upper limit, such as 30. After the iteration terminates, the center coordinates of the search window obtained from the last update are... This pinpoints the precise location of the pollution source. The dimensions of this convergence window, such as width and height, represent the extent of the pollution plume's spread, while the rotation angle indicates the main direction of pollution diffusion.

[0116] In some embodiments, determining that the search window has converged when the magnitude of the composite displacement vector is less than a preset convergence threshold includes:

[0117] After each iteration update, the magnitude L of the composite displacement vector is calculated;

[0118] When the modulus L is less than the preset convergence threshold When the search window converges, the iteration stops.

[0119] The core model is based on the movement magnitude of the search window center position. In each iteration, a composite displacement vector is generated to update the window center position. Before updating the position, the Euclidean magnitude of the vector, i.e., its size, needs to be calculated. The magnitude L represents the expected distance the window center will move in the current iteration step. For example, in the k-th iteration, the calculated composite displacement vector is... If the pixel is a modulus, then the modulus is Pixel.

[0120] Pre-set a very small positive number as the convergence threshold. ,For example Pixels. The threshold represents the minimum acceptable amount of movement; any movement less than this value is considered to indicate that the window position is substantially stable. This is done each time the modulus is calculated. Afterwards, the modulus result will be compared with... Comparison. In the example above, Since 1.0 > 0.1, the convergence condition is not met, so the algorithm will continue to the next iteration. If in the subsequent m-th iteration, the calculated displacement vector is... , module length Since 0.086 < 0.1, the condition is satisfied at this point. When the condition is met, the algorithm determines that the search window has converged. The iteration process then stops, and the current position, size, and orientation of the search window are output as the source tracing result.

[0121] In a second embodiment, the present invention also provides a pollution source tracing and intelligent identification system utilizing quantum radar, comprising the following modules:

[0122] The acquisition module is used to acquire the photonic quantum data frame of the monitoring area, perform a conformity test on the signal intensity of each data point in the photonic quantum data frame based on a preset quantum background noise Poisson statistical model, and set the intensity of the data points that do not conform to the test to zero, so as to obtain a preprocessed data frame.

[0123] The generation module is used to construct a multidimensional quantum feature space based on the photon wavelength, polarization state and arrival time of the pollution source to be traced, calculate the multidimensional quantum features of each data point in the monitoring area based on the preprocessed data frame, and generate a quantum feature back projection probability distribution map of the monitoring area.

[0124] A construction module is used to set an initial search window on the probability distribution map and iteratively update the search window. Each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient; and calculating and updating the size and orientation of the search window according to the second central moment.

[0125] The determination module is used to determine the convergence of the search window when the magnitude of the composite displacement vector is less than the preset convergence threshold, and to determine the center position of the converged window as the location of the pollution source, and to determine the size and direction of the converged window as the diffusion range and direction of the pollution source.

[0126] In this specification, relational terms such as "first" and "second" are used merely 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 a process, method, article, or apparatus. Unless otherwise limited, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the associated listed items.

[0127] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0128] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent identification of pollution sources using quantum radar, characterized in that, Includes the following steps: Acquire photonic quantum data frames of the monitoring area, perform compliance checks on the signal strength of each data point within the photonic quantum data frames based on a preset quantum background noise Poisson statistical model, and set the strength of data points that do not meet the test to zero to obtain preprocessed data frames; A multidimensional quantum feature space is constructed based on the photon wavelength, polarization state and arrival time of the pollution source to be traced. The multidimensional quantum features of each data point in the monitoring area are calculated based on the preprocessed data frame, and a quantum feature back projection probability distribution map of the monitoring area is generated. An initial search window is set on the probability distribution graph, and the search window is iteratively updated. Each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient to update the center position of the search window; and calculating and updating the size and orientation of the search window according to the second central moment. When the magnitude of the composite displacement vector is less than the preset convergence threshold, the search window is determined to have converged, and the center position of the converged window is determined as the location of the pollution source. The size and direction of the converged window are determined as the diffusion range and direction of the pollution source.

2. The method according to claim 1, characterized in that, The step of performing a conformity check on the signal intensity of each data point within the quantum data frame based on a preset quantum background noise Poisson statistical model, and setting the intensity of data points that do not conform to the check to zero, includes: Establish the null hypothesis that the data points belong to background noise; Set the significance level to α; Calculate the p-value of the signal strength at each data point under the Poisson statistical model; When the p-value is less than α, the null hypothesis is rejected and the intensity of the data points is retained. When the p value is greater than or equal to α, the null hypothesis is accepted and the intensity of the data point is set to zero.

3. The method according to claim 1, characterized in that, The construction of a multidimensional quantum feature space based on the photon wavelength, polarization state, and arrival time of the pollution source to be traced includes: A three-dimensional Cartesian coordinate system is constructed with the photon wavelength λ as the first axis, the polarization state angle θ as the second axis, and the arrival time t as the third axis, as the multidimensional quantum feature space.

4. The method according to claim 1, characterized in that, The step of calculating the multidimensional quantum features of each data point within the monitoring area based on the preprocessed data frame, and generating a quantum feature back-projection probability distribution map of the monitoring area, includes: Using sample data from the pollution source to be traced, a three-dimensional target feature histogram is statistically generated in the multidimensional quantum feature space; For each data point in the preprocessed data frame, the multidimensional quantum features are mapped onto the target feature histogram, the corresponding histogram statistics are found and obtained, and the statistics are used as the probability value of the data point in the back-projected probability distribution map.

5. The method according to claim 1, characterized in that, Setting an initial search window on the probability distribution graph includes: Traverse the probability distribution graph to find the data point with the highest probability value; Set the center of the initial search window at the location of the data point; Set the initial search window size to a rectangular window that covers a preset proportion β of the total area of ​​the monitored region.

6. The method according to claim 1, characterized in that, The step of constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first-order moment, the skewness vector determined by the third-order central moment, and the spatial gradient includes: Let the centroid displacement vector determined by the first moment be denoted as... The skewness vector determined by the third central moments is denoted as... And the spatial gradient vector is denoted as ; The composite displacement vector V is obtained by weighted summation; The updated center position is obtained by adding the composite displacement vector V to the center position of the current search window.

7. The method according to claim 1, characterized in that, The step of calculating and updating the size and orientation of the search window based on the second-order central moment includes: From the second-order central moment , , Construct the covariance matrix; Calculate the eigenvalues ​​of the covariance matrix. , and the corresponding feature vector , ; Update the orientation of the search window to the feature vector. , The direction it points to; Update the major and minor axes of the search window to... and , where C is a preset coefficient.

8. The method according to any one of claims 1-7, characterized in that, The step of determining that the search window has converged when the magnitude of the composite displacement vector is less than a preset convergence threshold includes: After each iteration update, the magnitude L of the composite displacement vector is calculated; When the modulus L is less than the preset convergence threshold When the search window converges, the iteration stops.

9. A pollution source tracing and intelligent identification system utilizing quantum radar, characterized in that, Includes the following modules: The acquisition module is used to acquire the photonic quantum data frame of the monitoring area, perform a conformity test on the signal intensity of each data point in the photonic quantum data frame based on a preset quantum background noise Poisson statistical model, and set the intensity of the data points that do not conform to the test to zero, so as to obtain a preprocessed data frame. The generation module is used to construct a multidimensional quantum feature space based on the photon wavelength, polarization state and arrival time of the pollution source to be traced, calculate the multidimensional quantum features of each data point in the monitoring area based on the preprocessed data frame, and generate a quantum feature back projection probability distribution map of the monitoring area. A construction module is used to set an initial search window on the probability distribution map and iteratively update the search window. Each iteration update includes: calculating the first moment, second central moment, and third central moment of the probability distribution map within the current search window, as well as the spatial gradient of the probability distribution map at the center of the window; constructing a composite displacement vector to update the center position of the search window based on the centroid position determined by the first moment, the skewness vector determined by the third central moment, and the spatial gradient; and calculating and updating the size and orientation of the search window according to the second central moment. The determination module is used to determine the convergence of the search window when the magnitude of the composite displacement vector is less than the preset convergence threshold, and to determine the center position of the converged window as the location of the pollution source, and to determine the size and direction of the converged window as the diffusion range and direction of the pollution source.

Images