Neutron gamma pulse rapid matching method and system

Abstract

The invention belongs to the technical field of radiation measurement, and particularly relates to a neutron gamma pulse rapid matching method and system. The method comprises the following steps: acquiring a digital pulse sequence; determining a parabolic virtual baseline according to the weighted sum of the pulse baseline noise and the peak amplitude, and solving an intersection point of the pulse first-order differential signal and the baseline as a weighted zero crossing point time feature; calculating a ratio of waveform area first-order moments before and after a peak point as an asymmetric torque characteristic; constructing a basic transition probability matrix by using a historical sample, and performing nonlinear disturbance according to the asymmetric torque characteristic of the current pulse to generate an instantaneous transition probability matrix; and combining the two features to form a two-dimensional vector, calculating a weighted mahalanobis distance based on an instantaneous transition probability matrix, and completing matching according to a short distance principle. According to the invention, the problem of insufficient precision and stability caused by noise sensitivity and fixed matching criterion of the existing pulse discrimination technology is solved.

Description

Technical Field

[0001] This invention belongs to the field of radiation measurement technology, specifically relating to a fast matching method and system for neutron gamma pulses. Background Technology

[0002] In the field of nuclear radiation detection, the accurate differentiation between neutrons and gamma rays is a core requirement for ensuring the accuracy of radiation measurements, and pulse shape discrimination technology is a key means to achieve this goal. Currently, the mainstream discrimination techniques mainly rely on the charge comparison method and the zero-crossing method, but both have significant limitations in practical applications, making it difficult to balance stability, accuracy, and processing efficiency.

[0003] The core principle of the charge comparison method is to classify components by calculating the ratio of the charge of the fast and slow components of a pulse signal. Its advantages lie in its simplicity and low hardware implementation cost, making it widely used in early radiation detection systems. However, this method is highly dependent on the integrity of the pulse signal: in high count rate scenarios, adjacent pulses are prone to superposition, i.e., pulse stacking, leading to deviations in the calculation of the charge of the fast and slow components. Simultaneously, factors such as temperature fluctuations and power supply noise in the detector's operating environment can cause baseline drift, further compromising the stability of the charge ratio. Ultimately, this results in a significant decrease in discrimination accuracy, failing to meet the requirements of high radiation intensity scenarios.

[0004] The zero-crossing rule distinguishes neutrons from gamma rays by extracting the time difference between the zero-crossing points of bipolar pulses through specific shaping processing of the pulse signal. Compared to the charge comparison method, it is faster and theoretically more suitable for real-time detection scenarios. However, this method is extremely sensitive to electronic noise: when the detector output signal-to-noise ratio fluctuates, the zero-crossing time will produce significant jitter, especially when processing low-amplitude pulses. The jitter amplitude may even exceed the time difference between neutrons and gamma rays themselves, leading to a large number of misclassifications. In addition, the pulse shaping process requires additional signal processing steps, which may cause signal delays. Furthermore, the fixed shape parameters are difficult to adapt to the differences in pulse shape produced by particles of different energies, further limiting its applicability.

[0005] To improve discrimination accuracy, the industry has attempted to introduce more complex feature extraction algorithms, such as wavelet transform and frequency domain analysis, or machine learning models, such as support vector machines and artificial neural networks. These methods, by mining multi-dimensional features of pulse signals or constructing complex classification models, have indeed improved discrimination results to some extent, especially in low signal-to-noise ratio and multi-energy mixed radiation scenarios, outperforming traditional methods. However, this comes at the cost of a sharp increase in computational complexity, impacting processing capabilities. Furthermore, most existing discrimination methods, whether based on simple threshold criteria or complex machine learning models, typically rely on static core matching or classification logic. That is, once the model is trained or the parameters are set, the discrimination criteria remain fixed. However, in actual detection scenarios, the detector's operating state dynamically drifts with environmental factors, and the pulse patterns generated by neutrons and gamma rays of different energies exhibit subtle and continuous changes. Static discrimination criteria cannot adapt to these dynamic changes, leading to insufficient long-term system stability, requiring frequent manual parameter calibration, increasing maintenance costs, and failing to meet the needs of unattended monitoring scenarios. Summary of the Invention

[0006] Therefore, the purpose of this invention is to propose a fast matching method and system for neutron gamma pulses, in order to solve the technical problems of existing pulse discrimination technology being sensitive to noise and having fixed matching criteria, resulting in insufficient accuracy and stability.

[0007] To address the above problems, the technical solution of the fast matching method for neutron gamma pulses proposed in this invention is as follows: A fast matching method for neutron gamma pulses includes the following steps: Acquire the digital pulse sequence output by the neutron gamma detector; For each pulse signal in the digital pulse sequence, a parabolic virtual baseline is determined based on the weighted sum of the mean and peak amplitude of the baseline noise before the pulse; the pulse signal is differentiated by first order, and the intersection of the differentiated signal and the parabolic virtual baseline is solved, and the time corresponding to the intersection is used as the weighted zero-crossing time feature of the pulse. Using the peak time of the pulse signal as the time axis, the first moment of the waveform area before and after the peak point relative to the time axis is calculated, and the ratio of the two is used as the asymmetric torque feature of the pulse; a normalized confusion matrix is ​​constructed using a historical pulse sample dataset, and the normalized confusion matrix is ​​used as the basic transition probability matrix; based on the asymmetric torque feature of the current pulse, the diagonal elements of the basic transition probability matrix are nonlinearly perturbed to generate an instantaneous transition probability matrix related to the current pulse shape; The weighted zero-crossing time features and asymmetric torque features are combined to form a two-dimensional feature vector of the pulse to be matched. Based on the instantaneous transition probability matrix, Mahalanobis distance weights are set for the preset feature cluster centers of neutrons and gammas. The Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers is calculated. The matching of neutron or gamma pulses is completed according to the principle of closer distance.

[0008] Furthermore, the method for determining the parabolic virtual baseline is as follows: Let the mean of the pre-pulse baseline noise be . Peak amplitude is The peak time is The pulse start time is The magnitude of the parabola's vertex is ; Parabola vertex amplitude The calculation formula is: ;in, , and These are preset weighting coefficients; by The vertex of the parabola is such that the parabola passes through the pulse initiation point. Constructing a parabolic virtual baseline : .

[0009] Furthermore, the mean baseline noise before the pulse is obtained by extracting a preset number of sampling points before the leading edge of the pulse signal and calculating the mean amplitude of these sampling points.

[0010] Furthermore, the method for determining the weighted zero-crossing time characteristics of the pulse is as follows: The first-order differential signal of the pulse signal is calculated using the central difference method. Solve for all intersection points by simultaneously solving the equations of the first-order differential signal and the parabolic virtual baseline; Select the unique intersection point of the time coordinates between the start point and the peak point of the pulse signal among all intersection points, and use the time corresponding to this intersection point as the weighted zero-crossing time feature of the pulse.

[0011] Furthermore, the method for determining the asymmetric torque characteristics of the pulse is as follows: For the waveform portion preceding the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the preceding segment; For the waveform portion following the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the latter part; The ratio of the first moment of the latter segment to the first moment of the former segment is calculated and used as the asymmetric torque characteristic of the pulse.

[0012] Furthermore, the method for determining the basic transition probability matrix is ​​as follows: A support vector machine classifier is trained using the aforementioned historical pulse sample dataset; The historical pulse sample dataset is input into the support vector machine classifier for classification. The classification results are statistically analyzed, and a 2x2 confusion matrix is ​​generated. The confusion matrix is ​​defined as follows: Elements in the confusion matrix Indicates the true category is The pulse is classified into categories. The number of samples; Normalize each row of the confusion matrix to obtain the basic transition probability matrix; define the basic transition probability matrix as follows. Elements in the basic transition probability matrix The calculation method is as follows: .

[0013] Furthermore, the method for determining the instantaneous transition probability matrix is ​​as follows: The standard asymmetric torque characteristic reference values ​​for neutron pulses and gamma pulses are defined, denoted as follows: and Let the asymmetric torque characteristic of the current pulse be... Calculate the absolute value of the difference between it and the two reference values ​​respectively. , ,in, ; Substituting the absolute value of the difference into the exponential decay function, two independent disturbance adjustment coefficients are calculated. , , The preset attenuation constant; Update the basic transition probability matrix using the perturbation adjustment coefficients. The diagonal elements are used to obtain an intermediate matrix. The elements in the intermediate matrix are calculated as follows: ; ; For the intermediate matrix Each row is normalized to generate the instantaneous transition probability matrix.

[0014] Furthermore, the Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers is calculated as follows: Let the two-dimensional feature vector of the pulse to be matched be... The preset feature cluster centers for neutrons and gammas are respectively and The characteristic covariance matrix of historical pulse samples is The instantaneous transition probability matrix is ; Calculate separately Weighted Mahalanobis distance to neutron and gamma cluster centers ,in, The calculation formula is as follows: , The diagonal elements of the instantaneous transition probability matrix serve as weighting factors for the Mahalanobis distance.

[0015] Furthermore, the process of obtaining the digital pulse sequence output by the neutron-gamma detector is as follows: an organic scintillator detector is used in conjunction with a photomultiplier tube to convert the scintillation light generated by the interaction of neutrons or gamma rays with the scintillator into an analog electrical pulse signal. Then, a high-speed analog-to-digital converter is used to sample the analog electrical pulse signal at a sampling rate of 500MHz and a sampling precision of 12 bits, converting the analog electrical pulse signal into time series data composed of a series of discrete sampling point amplitudes, thus obtaining the digital pulse sequence.

[0016] The technical solution of the neutron gamma pulse fast matching system proposed in this invention is as follows: A neutron-gamma pulse fast matching system includes a processor and a memory; the memory stores a computer program, which, when run by the processor, implements a neutron-gamma pulse fast matching method according to any of the above technical solutions.

[0017] The beneficial effects of this invention are as follows: By constructing a parabolic virtual baseline that integrates the mean of the pre-pulse baseline noise and the peak amplitude, and combining this with the first-order differential solution to obtain weighted zero-crossing time features, this invention effectively suppresses the interference of electronic noise and baseline drift, significantly improving the stability of time features in low signal-to-noise ratio and low-amplitude pulse scenarios. Simultaneously, by calculating the ratio of the first-order moments of the waveform areas before and after the peak time as an asymmetric torque feature, this invention accurately captures the essential differences between the tail morphology of neutron and gamma pulses, providing a highly discriminative basis for distinguishing between the two types of pulses. Furthermore, by generating basic transfer through training with historical samples... The probability matrix is ​​generated by nonlinear perturbation based on the asymmetric torque characteristics of the current pulse to produce an instantaneous transition probability matrix. This allows the discrimination criterion to dynamically adapt to subtle changes in pulse shape and detector operating state drift, significantly enhancing the reliability of identifying pulses with blurred boundaries. Finally, Mahalanobis distance weights are set based on the instantaneous transition probability matrix, and classification matching is completed by combining two-dimensional feature vectors. This avoids the high computational complexity caused by complex algorithms, ensures real-time processing capabilities in high count rate scenarios, and comprehensively considers discrimination accuracy, stability, and scenario adaptability through feature complementarity and dynamic weight adjustment. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the fast matching method for neutron gamma pulses of the present invention. Figure 2 A schematic diagram of the acquired digital pulse sequence; Figure 3 This is a schematic diagram showing how the pulse waveform is divided into a pre-wave section and a post-wave section using the time axis at the peak point. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Specific embodiments of the fast matching method for neutron gamma pulses proposed in this invention: like Figure 1 As shown, the fast matching method for neutron gamma pulses includes the following steps: S1, acquire the digital pulse sequence output by the neutron gamma detector.

[0021] Specifically, this step uses an organic scintillator detector in conjunction with a photomultiplier tube to convert the scintillation light generated by the interaction of neutrons or gamma rays with the scintillator into an analog electrical pulse signal. Then, a high-speed analog-to-digital converter samples the analog electrical pulse signal at a sampling rate of 500MHz and a sampling precision of 12 bits, converting the analog electrical pulse signal into time-series data composed of a series of discrete sampling point amplitudes, i.e., a digital pulse sequence, such as... Figure 2 As shown.

[0022] S2, for each pulse signal in the digital pulse sequence, a parabolic virtual baseline is determined based on the weighted sum of the mean and peak amplitude of the baseline noise before the pulse; the pulse signal is differentiated by first order, and the intersection of the differentiated signal and the parabolic virtual baseline is solved, and the time corresponding to the intersection is used as the weighted zero-crossing time feature of the pulse.

[0023] This step replaces the traditional fixed threshold or zero level with a parabolic virtual baseline. This parabolic virtual baseline is dynamically generated, and its vertex position is determined by the mean and peak amplitude of the current pulse's baseline noise. This means that the parabolic virtual baseline can automatically adapt to pulses of different amplitudes and varying noise levels. The first derivative represents the rate of change of the pulse signal. Solving for the intersection of the differential signal and this dynamic parabolic virtual baseline aims to find a time point that stably reflects the pulse's rising edge characteristics, a point that is less susceptible to noise and baseline drift.

[0024] In this embodiment, the method for determining the parabolic virtual baseline is as follows: Let the mean of the pre-pulse baseline noise be . Peak amplitude is The peak time is The pulse start time is The magnitude of the parabola's vertex is ; Parabola vertex amplitude The calculation formula is: ;in, , and These are preset weighting coefficients; by The vertex of the parabola is such that the parabola passes through the pulse initiation point. Constructing a parabolic virtual baseline : .

[0025] Specifically, by setting a trigger threshold, individual pulse waveform data are identified and extracted from the digital pulse sequence. The average amplitude of the 50 sampling points before the leading edge of the pulse signal is calculated to obtain the mean baseline noise before the pulse. Find the maximum amplitude, i.e., the peak amplitude, by traversing the entire pulse waveform data. Set weighting coefficients. and ,For example =0.2, =0.8, and the magnitude of the parabola's vertex can be obtained by calculating the weighted sum. .

[0026] For example, the start time of a pulse The peak time is 50 ns. Mean baseline noise before pulse is 100 ns. For 10 amplitude units, peak amplitude The amplitude is 510 units. Calculate the amplitude at the vertex of the parabola using the formula. The calculated result is 410 amplitude units. After obtaining the vertex amplitude, the vertex coordinates of the parabola can be determined as time 100 ns and amplitude 410. The parabola must also pass through the pulse's starting point, with coordinates of time 50 ns and amplitude 10. Using these two points, i.e., the vertex coordinates and the starting point coordinates, a unique equation for a downward-opening parabola can be determined. Substituting these values ​​into the given formula yields a parabolic virtual baseline. For a time The quadratic function forms a smooth curve below the rising edge of the pulse, which serves as the threshold for subsequent processing.

[0027] In this embodiment, the method for determining the weighted zero-crossing time characteristics of the pulse is as follows: The first-order differential signal of the pulse signal is calculated using the central difference method. Solve for all intersection points by simultaneously solving the equations of the first-order differential signal and the parabolic virtual baseline; Select the unique intersection point of the time coordinates between the start point and the peak point of the pulse signal among all intersection points, and use the time corresponding to this intersection point as the weighted zero-crossing time feature of the pulse.

[0028] Specifically, for a discrete digital pulse signal, the center difference method can be used to calculate the slope at each sampling point. For example, for a time point... At the sampling point, the differential value is approximately equal to Signal amplitude minus The signal amplitude at each sampling point is then divided by twice the sampling time interval. Performing this operation on all sampling points of the entire pulse signal yields a new first-order differential signal curve. This differential signal curve is then compared to a parabolic virtual baseline. By solving the system of equations simultaneously, we can find the intersection points of the two function curves, which means finding all time points where the differential signal value equals the parabolic virtual baseline value. The solution process may yield one or more intersection points. All intersection points are filtered, retaining only those with time coordinates located at the pulse start point. and peak point An intersection point between the two points marks the end of the steepest rising region of the pulse signal, and is an important time characteristic point of the pulse, namely the weighted zero-crossing time characteristic of the pulse.

[0029] Of course, in other embodiments, the differential signal of the pulse signal can also be calculated using the first-order difference method.

[0030] S3, using the peak time of the pulse signal as the time axis, calculate the first moment of the waveform area before and after the peak point relative to the time axis, and use the ratio of the two as the asymmetric torque feature of the pulse; construct a normalized confusion matrix using the historical pulse sample dataset, and use the normalized confusion matrix as the basic transition probability matrix; according to the asymmetric torque feature of the current pulse, perform nonlinear perturbation on the diagonal elements of the basic transition probability matrix to generate an instantaneous transition probability matrix related to the current pulse shape.

[0031] This step comprises three subtasks: extracting asymmetric torque features, constructing the basic transition probability matrix, and generating the instantaneous transition probability matrix. The waveforms of neutron pulses and gamma pulses, especially the tail portion, exhibit significant differences. The first moment is a method of quantifying weighted area; calculating the ratio of the first moment before and after the peak is used to quantify the degree of asymmetry in the pulse shape, i.e., the asymmetric torque feature, indicating the length of the tail. Neutron pulses typically have longer tails, with a relatively larger first moment in the latter part, thus resulting in a correspondingly larger value for the asymmetric torque feature, effectively distinguishing the two types of pulses.

[0032] In this step, the method for determining the asymmetric torque characteristics of the pulse is as follows: For the waveform portion preceding the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the preceding segment; For the waveform portion following the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the latter part; The ratio of the first moment of the latter segment to the first moment of the former segment is calculated and used as the asymmetric torque characteristic of the pulse.

[0033] For example, if the peak time of a pulse is 100 ns, then the time axis t=100 is the reference axis, dividing the pulse waveform into two parts: the waveform before the peak point and the waveform after the peak point, as shown below. Figure 3 As shown.

[0034] In this embodiment, the method for determining the basic transition probability matrix is ​​as follows: A support vector machine classifier is trained using the aforementioned historical pulse sample dataset; The historical pulse sample dataset is input into the support vector machine classifier for classification. The classification results are statistically analyzed, and a 2x2 confusion matrix is ​​generated. The confusion matrix is ​​defined as follows: Elements in the confusion matrix Indicates the true category is The pulse is classified into categories. The number of samples; Normalize each row of the confusion matrix to obtain the basic transition probability matrix; define the basic transition probability matrix as follows. Elements in the basic transition probability matrix The calculation method is as follows: .

[0035] Specifically, this process requires a historical dataset containing a large number of labeled pulses, for example, 1000 known neutron pulses and 1000 known gamma pulses. A support vector machine (SVM) classifier model is trained using this dataset to learn patterns that distinguish between the two types of pulses. After training, these 2000 historical pulse samples are fed back into the newly trained SVM classifier for prediction. The prediction results are statistically analyzed and compared with the true labels to generate a confusion matrix. Assuming the classification results are: 950 of the 1000 true neutron pulses are correctly identified as neutrons, and 50 are incorrectly identified as gammas; 80 of the 1000 true gamma pulses are incorrectly identified as neutrons, and 920 are correctly identified as gammas, a confusion matrix is ​​obtained. Each row of this confusion matrix is ​​then normalized, i.e., each element of the row is divided by the sum of all elements in that row. Dividing 950 and 50 in the first row by the sum of 1000 yields 0.95 and 0.05 respectively; dividing 80 and 920 in the second row by the sum of 1000 yields 0.08 and 0.92 respectively. Therefore, the resulting fundamental transition probability matrix... The first row contains 0.95 and 0.05, and the second row contains 0.08 and 0.92. The basic transition probability matrix records the basic misclassification probability of the support vector machine classifier for each class.

[0036] In this embodiment, the instantaneous transition probability matrix is ​​determined as follows: The standard asymmetric torque characteristic reference values ​​for neutron pulses and gamma pulses are defined, denoted as follows: and Let the asymmetric torque characteristic of the current pulse be... Calculate the absolute value of the difference between it and the two reference values ​​respectively. , ,in, ; Substituting the absolute value of the difference into the exponential decay function, two independent disturbance adjustment coefficients are calculated. , , The preset attenuation constant; Update the basic transition probability matrix using the perturbation adjustment coefficients. The diagonal elements are used to obtain an intermediate matrix. The elements in the intermediate matrix are calculated as follows: ; ; For the intermediate matrix Each row is normalized to generate an instantaneous transition probability matrix.

[0037] Specifically, this method sets standard asymmetric torque characteristic reference values ​​for two types of pulses, for example, 1.5 for neutron pulses and 0.8 for gamma pulses. When a new pulse arrives, the asymmetric torque characteristic of the current pulse is calculated, assumed to be 1.4. The difference between this value and the two reference values ​​is calculated: 0.1 for neutron pulses and 0.6 for gamma pulses. This indicates that the shape of the pulse is closer to that of a typical neutron pulse. The two difference values ​​are converted into perturbation adjustment coefficients using an exponential decay function. Assuming a decay constant of 5, the perturbation adjustment coefficient for neutron pulses is 0.607, and for gamma pulses it is 0.050. After obtaining the perturbation adjustment coefficients, they are used to adjust the diagonal elements of the basic transition probability matrix. The diagonal elements of the basic transition probability matrix are 0.95 and 0.92, respectively. After adjustment, the diagonal elements are approximately 0.577 and 0.046, while the off-diagonal elements remain unchanged. Each row of the adjusted matrix is ​​normalized again to obtain the instantaneous transition probability matrix, which reflects the impact of the current impulse feature on the classification confidence.

[0038] S4. Combine the weighted zero-crossing time features and asymmetric torque features to form a two-dimensional feature vector of the pulse to be matched; based on the instantaneous transition probability matrix, set the Mahalanobis distance weight for the preset feature cluster centers of neutrons and gammas, calculate the Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers, and complete the matching of neutron or gamma pulses according to the principle of closer distance.

[0039] In this step, the Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers is calculated as follows: Let the two-dimensional feature vector of the pulse to be matched be... The preset feature cluster centers for neutrons and gammas are respectively and The characteristic covariance matrix of historical pulse samples is The instantaneous transition probability matrix is ; Calculate separately Weighted Mahalanobis distance to neutron and gamma cluster centers ,in, The calculation formula is as follows: , The diagonal elements of the instantaneous transition probability matrix serve as weighting factors for the Mahalanobis distance.

[0040] Compare the magnitudes of the two weighted Mahalanobis distances. < If the signal is positive, the current pulse will be matched as a neutron; otherwise, it will be matched as a gamma.

[0041] This invention suppresses noise and baseline drift interference, improves stability under low signal-to-noise ratio conditions, enhances matching reliability, and significantly improves the ability to distinguish boundary-ambiguous pulses.

[0042] Specific embodiments of the neutron gamma pulse fast matching system proposed in this invention: The neutron-gamma pulse fast matching system includes a processor and a memory; the memory stores a computer program, which, when run by the processor, implements the neutron-gamma pulse fast matching method in the above embodiments.

[0043] The neutron-gamma pulse fast matching system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. Their setup and functions are known in the art and will not be described in detail here.

[0044] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0045] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A fast matching method for neutron gamma pulses, characterized in that, Includes the following steps: Acquire the digital pulse sequence output by the neutron gamma detector; For each pulse signal in the digital pulse sequence, a parabolic virtual baseline is determined based on the weighted sum of the mean and peak amplitude of the baseline noise before the pulse; the pulse signal is differentiated by first order, and the intersection of the differentiated signal and the parabolic virtual baseline is solved, and the time corresponding to the intersection is used as the weighted zero-crossing time feature of the pulse. Using the peak time of the pulse signal as the time axis, the first moment of the waveform area before and after the peak point relative to the time axis is calculated, and the ratio of the two is used as the asymmetric torque feature of the pulse; a normalized confusion matrix is ​​constructed using a historical pulse sample dataset, and the normalized confusion matrix is ​​used as the basic transition probability matrix; based on the asymmetric torque feature of the current pulse, the diagonal elements of the basic transition probability matrix are nonlinearly perturbed to generate an instantaneous transition probability matrix related to the current pulse shape; The weighted zero-crossing time features and asymmetric torque features are combined to form a two-dimensional feature vector of the pulse to be matched. Based on the instantaneous transition probability matrix, Mahalanobis distance weights are set for the preset feature cluster centers of neutrons and gammas. The Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers is calculated. The matching of neutron or gamma pulses is completed according to the principle of closer distance.

2. The fast matching method for neutron gamma pulses according to claim 1, characterized in that, The method for determining the parabolic virtual baseline is as follows: Let the mean of the pre-pulse baseline noise be . Peak amplitude is The peak time is The pulse start time is The magnitude of the parabola's vertex is ; Parabola vertex amplitude The calculation formula is: ;in, , and These are preset weighting coefficients; by The vertex of the parabola is such that the parabola passes through the pulse initiation point. Constructing a parabolic virtual baseline : .

3. The fast matching method for neutron gamma pulses according to claim 2, characterized in that, The mean baseline noise before the pulse is obtained by taking a preset number of sampling points before the leading edge of the pulse signal and calculating the mean amplitude of these sampling points.

4. The fast matching method for neutron gamma pulses according to claim 1, characterized in that, The method for determining the weighted zero-crossing time characteristics of the pulse is as follows: The first-order differential signal of the pulse signal is calculated using the central difference method. Solve for all intersection points by simultaneously solving the equations of the first-order differential signal and the parabolic virtual baseline; Select the unique intersection point of the time coordinates between the start point and the peak point of the pulse signal among all intersection points, and use the time corresponding to this intersection point as the weighted zero-crossing time feature of the pulse.

5. The fast matching method for neutron gamma pulses according to claim 1, characterized in that, The method for determining the asymmetric torque characteristics of the pulse is as follows: For the waveform portion preceding the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the preceding segment; For the waveform portion following the time axis, the product of the amplitude of each sampling point and the time distance from the sampling point to the time axis is accumulated to obtain the first moment of the latter part; The ratio of the first moment of the latter segment to the first moment of the former segment is calculated and used as the asymmetric torque characteristic of the pulse.

6. The fast matching method for neutron gamma pulses according to claim 1, characterized in that, The method for determining the basic transition probability matrix is ​​as follows: A support vector machine classifier is trained using the aforementioned historical pulse sample dataset; The historical pulse sample dataset is input into the support vector machine classifier for classification. The classification results are statistically analyzed, and a 2x2 confusion matrix is ​​generated. The confusion matrix is ​​defined as follows: Elements in the confusion matrix Indicates the true category is The pulse is classified into categories. The number of samples; Normalize each row of the confusion matrix to obtain the basic transition probability matrix; define the basic transition probability matrix as follows. Elements in the basic transition probability matrix The calculation method is as follows: .

7. The fast matching method for neutron gamma pulses according to claim 6, characterized in that, The method for determining the instantaneous transition probability matrix is ​​as follows: The standard asymmetric torque characteristic reference values ​​for neutron pulses and gamma pulses are defined, denoted as follows: and Let the asymmetric torque characteristic of the current pulse be... Calculate the absolute value of the difference between it and the two reference values ​​respectively. , ,in, ; Substituting the absolute value of the difference into the exponential decay function, two independent disturbance adjustment coefficients are calculated. , , The preset attenuation constant; Update the basic transition probability matrix using the perturbation adjustment coefficients. The diagonal elements are used to obtain an intermediate matrix. The elements in the intermediate matrix are calculated as follows: ； ； For the intermediate matrix Each row is normalized to generate the instantaneous transition probability matrix.

8. The fast matching method for neutron gamma pulses according to claim 7, characterized in that, The method for calculating the Mahalanobis distance from the two-dimensional feature vector to the weighted neutron and gamma cluster centers is as follows: Let the two-dimensional feature vector of the pulse to be matched be... The preset feature cluster centers for neutrons and gammas are respectively and The characteristic covariance matrix of historical pulse samples is The instantaneous transition probability matrix is ; Calculate separately Weighted Mahalanobis distance to neutron and gamma cluster centers ,in, The calculation formula is as follows: , The diagonal elements of the instantaneous transition probability matrix serve as weighting factors for the Mahalanobis distance.

9. The fast matching method for neutron gamma pulses according to claim 1, characterized in that, The process of obtaining the digital pulse sequence output by the neutron-gamma detector is as follows: an organic scintillator detector is used in conjunction with a photomultiplier tube to convert the scintillation light generated by the interaction of neutrons or gamma rays with the scintillator into an analog electrical pulse signal. Then, a high-speed analog-to-digital converter is used to sample the analog electrical pulse signal at a sampling rate of 500MHz and a sampling precision of 12 bits. The analog electrical pulse signal is converted into time series data composed of a series of discrete sampling point amplitudes to obtain the digital pulse sequence.

10. A fast matching system for neutron gamma pulses, characterized in that, It includes a processor and a memory; the memory stores a computer program that, when run by the processor, implements the neutron gamma pulse fast matching method according to any one of claims 1-9.

Images