Neutron net counting method and system based on background counting spectrum adaptive deduction

Abstract

The application discloses a neutron net counting method and system based on background counting spectrum adaptive deduction, and relates to the technical field of spectrum analysis. The method comprises the following steps: collecting neutron counting spectrum data and detector working condition data to be measured, and obtaining a measured counting spectrum through energy channel calibration, dead time correction and time normalization; constructing a multi-source background template library based on historical background counting spectrum; determining a deduction coefficient according to the spectral shape similarity and dynamic allocation weight of the measured counting spectrum and each basic background spectrum template, obtaining a target neutron net counting spectrum and target neutron net counting; determining a net counting total uncertainty and outputting a confidence interval according to counting statistical error, background deduction error and model matching error; and judging background drift according to the reconstruction residual and net counting variation characteristics, and updating the multi-source background template library in combination with the detector working condition data. The application can improve the stability and traceability of the neutron net counting result under the condition of time-varying background.

Description

Technical Field

[0001] This invention relates to the field of energy spectrum analysis technology, and in particular to a neutron net counting method and system based on adaptive subtraction of background counting spectrum. Background Technology

[0002] In the fields of nuclear radiation detection and neutron energy spectrum analysis, accurately obtaining the net neutron count of a target is a prerequisite for tasks such as nuclear material identification, dose assessment, and rare event search. Due to the complexity of the neutron detection environment, the measured count spectrum is often mixed with various background interferences, including ambient radiation, detector inherent radioactivity, and electronic system noise. To eliminate these interferences, existing background subtraction methods mostly employ background measurements within a fixed time window and direct subtraction, or rely on conventional background estimation techniques based on smoothing filters to obtain the net count.

[0003] However, existing technologies have significant limitations in practical applications. First, the background noise in complex measurement environments is not constant but fluctuates dynamically with changes in detector temperature, equipment aging, and environmental conditions. Traditional fixed background subtraction methods cannot adapt to this time-varying characteristic and are prone to undersubtraction or oversubtraction during periods of background fluctuation. Second, existing background fitting models based on single features struggle to effectively distinguish between physical background overlap, environmental interference, and electronic noise, easily leading to the loss of effective signals in low-background or dynamically high-noise scenarios. Furthermore, traditional methods typically only provide a single-point estimated count value after background subtraction, lacking dynamic quantitative evaluation of the current background model's matching degree and overall error, making it impossible to determine the reliability of the final output net count. In summary, existing methods, due to their poor adaptability, limited functionality, and lack of dynamic error tracking, are insufficient to meet the high stability and high reliability measurement requirements under complex operating conditions. Summary of the Invention

[0004] This invention provides a neutron net counting method and system based on adaptive subtraction of background counting spectrum, which solves the problems of poor adaptability and low interference stripping accuracy of traditional fixed background subtraction in the prior art.

[0005] The first aspect of this invention proposes a neutron net counting method based on adaptive subtraction of the background counting spectrum, comprising the following steps: Collect the neutron count spectrum data to be tested and the detector operating condition data. Use the detector operating condition data to preprocess the neutron count spectrum data to be tested to obtain the normalized measured count spectrum. The historical background counting spectrum is obtained, and the historical background counting spectrum is divided into background components from multiple different sources. An independent basic background spectrum template is constructed for each background component to form a multi-source background template library. Calculate the spectral similarity between the normalized measured count spectrum and each basic background spectral template in the multi-source background template library, and determine the current subtraction coefficient of each basic background spectral template based on the spectral similarity. The target neutron net count spectrum is obtained by subtracting each basic background spectrum template, which is weighted according to each current deduction coefficient, from the normalized measured count spectrum. The net count spectrum of the target neutrons is statistically analyzed to obtain the net count of the target neutrons. The error in the process of obtaining the net count spectrum of the target neutrons is evaluated to obtain the error synthesis term. The error synthesis term is synthesized into the total net count uncertainty. The net count confidence interval is generated and output based on the net count of the target neutrons and the total net count uncertainty. The residual characteristics are obtained by calculating the difference between the normalized measured count spectrum and the sum of each basic background spectrum template after weighting according to each current deduction coefficient; the fluctuation of the net target neutron count within the historical time window is statistically analyzed to obtain the variation characteristics; the background drift is assessed based on the residual characteristics and variation characteristics, and when the background drift is determined to have occurred, the multi-source background template library is updated in combination with the detector operating data.

[0006] In one alternative implementation, the detector operating data specifically includes the detector operating temperature, signal amplification gain status, and system dead time. The measured neutron count spectrum data is preprocessed using detector operating data to obtain a normalized measured count spectrum, specifically including: The energy channel alignment calibration of the neutron count spectrum data under test is performed using the detector operating temperature and signal amplification gain state. The system dead time is used to compensate for the count loss of the neutron count spectrum data after channel alignment calibration, and then smoothing and filtering are performed to output the normalized measured count spectrum.

[0007] In one optional implementation, the historical background counting spectrum is divided into background components from multiple different sources, and an independent basic background spectrum template is constructed for each background component, forming a multi-source background template library, specifically including: The historical background counting spectrum is separated into the environmental radiation background component, the detector intrinsic background component, the overlapping neutron background component, and the electronic noise background component. For each background component, its attenuation characteristics or random fluctuation characteristics are extracted. A clustering algorithm is used to establish a separate energy spectrum profile feature matrix for the background component with the corresponding characteristics. Each energy spectrum profile feature matrix is ​​stored as an independent basic background spectrum template in a multi-source background template library.

[0008] In one optional implementation, the spectral similarity between the normalized measured count spectrum and each basic background spectral template in the multi-source background template library is calculated, and the current subtraction coefficient for each basic background spectral template is determined based on the spectral similarity, specifically including: The normalized measured count spectrum is multiplied by the inner product of each basic background spectrum template in the multi-source background template library, and the corresponding spectral similarity is quantized and output. The spectral similarity is combined with the dynamically assigned weights generated for the normalized measured count spectrum to perform normalization mapping, generating the proportional weights of each basic background spectrum template, and the proportional weights are used as the current subtraction coefficients of each basic background spectrum template. The net target neutron count spectrum is obtained by subtracting the basic background spectrum templates, weighted according to each current deduction coefficient, from the normalized measured count spectrum. Specifically, this includes: The full width at half maximum (FWHM) of the energy peak and the variance of the counting statistics are analyzed in the normalized measured count spectrum. The energy deduction window is automatically constrained and calculated based on the full width at half maximum (FWHM) of the energy peak and the variance of the counting statistics. Within the energy subtraction window, the target neutron net count spectrum is output by subtracting each basic background spectrum template, which is weighted according to each current subtraction coefficient, from the normalized measured count spectrum.

[0009] In one alternative implementation, the dynamic weight allocation is obtained through a pre-trained multi-layer feedforward neural network model, and the specific acquisition process is as follows: The feature tensor of the normalized measured count spectrum, the environmental feature tensor of the historical background count spectrum, and the detector operating condition data are extracted and concatenated into a multi-dimensional composite input feature vector. The multidimensional composite input feature vector is fed into a multi-layer feedforward neural network model, and nonlinear correction feature extraction is performed through hidden network layers. The output layer of the multi-layer feedforward neural network model outputs dynamically assigned weights for each basic background spectrum template in the multi-source background template library.

[0010] In one alternative implementation, the error in obtaining the net count spectrum of the target neutron is evaluated to obtain an error synthesis term, which is then synthesized into a net total count uncertainty, specifically including: The statistical variance of the net count of target neutrons is calculated using the characteristic laws of the Poisson distribution, and the counting statistical error term is obtained; Based on the distribution deviation of each current deduction coefficient, calculate the background deduction error term introduced by the deviation of the background model deduction ratio; The root mean square difference between the normalized measured count spectrum and the sum of the basic background spectrum templates after weighting according to each current deduction coefficient is calculated to obtain the background model matching error term; The count statistics error term, background subtraction error term, and background model matching error term are combined as error synthesis terms to form the net total count uncertainty.

[0011] In one optional implementation, the difference between the normalized measured count spectrum and the sum of each basic background spectrum template weighted according to each current deduction coefficient is calculated to obtain residual characteristics; the fluctuation of the net target neutron count within the historical time window is statistically analyzed to obtain variation characteristics; the residual characteristics and variation characteristics are used to assess whether background drift has occurred; when background drift is determined to have occurred, the multi-source background template library is updated in conjunction with detector operating data, specifically including: The normalized measured count spectrum is subtracted from the sum of the basic background spectrum templates after being weighted according to each current subtraction coefficient, to form the reconstructed residual sequence on the discrete energy channel, and the reconstructed residual sequence is used as the residual feature; The ratio of the standard deviation to the mean of the net target neutron count over multiple consecutive historical time windows is used to obtain the variation index, which is then used as the variation characteristic. When the dispersion of the reconstructed residual sequence exceeds the first drift detection threshold, or the variation index exceeds the second drift detection threshold, background drift is determined to have occurred. The system retrieves a backup background template from the multi-source background template library, replaces it with a template that matches the determined background drift type and detector operating data, and updates the multi-source background template library.

[0012] In one optional implementation, after retrieving and replacing a backup background template that matches the determined background drift type and detector operating data, the method further includes: If background drift is determined to occur and there is no matching backup background template in the multi-source background template library, extract multiple normalized measured count spectra obtained continuously under background drift state to construct a feature matrix. A nonnegative matrix factorization algorithm is used to perform blind source decomposition on the feature matrix to extract independent anomalous background basis vectors. Independent anomalous background basis vectors are encapsulated to generate dynamically supplemented basic background spectrum templates, and the corresponding detector operating data are labeled. The dynamically supplemented basic background spectrum templates are then compiled into a multi-source background template library.

[0013] The acquisition of neutron count spectrum data to be tested includes: Deploy at least three sliding data acquisition windows with different integration time durations concurrently in the system; The captured neutron pulse signals are synchronously accumulated and recorded using various integral time sliding acquisition windows to form multi-scale parallel corresponding neutron count spectrum data. This allows for the identification of transient interference events based on the triggering characteristics under different integral time sliding acquisition windows, and the blocking of transient interference events from updating the multi-source background template library.

[0014] A second aspect of this invention proposes a neutron net counting system based on adaptive subtraction of the background counting spectrum, the system comprising: The preprocessing module is configured to collect the neutron count spectrum data to be measured and the detector operating condition data, and use the detector operating condition data to preprocess the neutron count spectrum data to be measured to obtain the normalized measured count spectrum. The background multi-source modeling module is configured to acquire historical background counting spectra, divide the historical background counting spectra into background components from multiple different sources, and build an independent basic background spectrum template for each background component to form a multi-source background template library. The adaptive subtraction module is configured to calculate the spectral similarity between the normalized measured count spectrum and each basic background spectrum template in the multi-source background template library, determine the current subtraction coefficient of each basic background spectrum template based on the spectral similarity, and subtract each basic background spectrum template weighted according to each current subtraction coefficient from the normalized measured count spectrum to obtain the target neutron net count spectrum. The error assessment and output module is configured to statistically analyze the net target neutron count spectrum to obtain the net target neutron count, assess the error in the process of obtaining the net target neutron count spectrum to obtain the error synthesis term, synthesize the error synthesis term into the total net count uncertainty, generate the net count confidence interval based on the net target neutron count and the total net count uncertainty, and output it. The closed-loop feedback update module is configured to calculate the difference between the normalized measured count spectrum and the sum of each basic background spectrum template after weighting according to each current deduction coefficient, to obtain residual characteristics; to statistically analyze the fluctuation of the net target neutron count within the historical time window, to obtain variation characteristics; to assess whether background drift has occurred based on the residual characteristics and variation characteristics; and when background drift is determined to have occurred, to update the multi-source background template library in conjunction with detector operating data.

[0015] A third aspect of this invention provides an electronic device, comprising: a memory and a processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the neutron net counting method based on adaptive subtraction of background counting spectrum as proposed in the foregoing embodiments.

[0016] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the neutron net counting method based on adaptive subtraction of background counting spectrum as proposed in the foregoing embodiments.

[0017] Beneficial Effects: This invention provides a neutron net counting method and system based on adaptive subtraction of the background counting spectrum. By subdividing historical background spectrum components and constructing a multi-source background template library, and dynamically determining the subtraction coefficient by combining the spectral similarity of the measured spectrum, it overcomes the limitations of existing technologies, such as poor environmental adaptability and low interference stripping accuracy caused by fixed background subtraction. This scheme utilizes dynamic weighting to achieve adaptive subtraction, flexibly responding to time-varying fluctuations in detection conditions and environmental background, and improving the accuracy of useful signal extraction in complex scenarios. Simultaneously, a comprehensive error evaluation mechanism covering statistical fluctuations and model bias is constructed, and the net counting result including confidence intervals is output, providing a quantitative and reliable basis for subsequent decision-making. Furthermore, by calculating residuals and variation characteristics in real time, a closed-loop monitoring and feedback update mechanism for background drift is established, enabling the system to autonomously track dynamic environmental changes, avoiding model failure caused by long-term operation, and ensuring the stability and accuracy of the measurement system under different operating conditions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 A flowchart illustrating the steps of a neutron net counting method based on adaptive subtraction of background counting spectrum provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the functional modules of a neutron net counting system based on adaptive subtraction of background counting spectrum provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to the accompanying drawings.

[0021] Reference Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0022] like Figure 1As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (Wi-Fi). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0023] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0024] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and computer programs.

[0025] exist Figure 1 In the illustrated electronic device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be set in the electronic device. The electronic device calls the neutron net counting system based on background count spectrum adaptive subtraction stored in the memory 1005 through the processor 1001, and executes the neutron net counting method based on background count spectrum adaptive subtraction provided in the embodiment of the present invention.

[0026] Combined with appendix Figure 2 As shown, this embodiment provides a neutron net count method based on adaptive subtraction of the background count spectrum. This method effectively removes various time-varying background interferences during neutron measurement through a closed-loop mechanism of multi-source background modeling, similarity matching, adaptive subtraction, uncertainty assessment, and dynamic updating, outputting a high-confidence neutron net count. The method of this embodiment fully includes the following steps S101 to S106: S101: Collect the neutron count spectrum data to be tested and the detector operating condition data. Use the detector operating condition data to preprocess the neutron count spectrum data to be tested to obtain the normalized measured count spectrum. In the fields of nuclear radiation detection and neutron energy spectrum analysis, front-end detectors (such as helium-3 proportional counters or scintillator detectors) are inevitably affected by the equipment's operating status and environmental physical parameters during the process of capturing neutrons and converting them into electrical pulses. Therefore, obtaining accurate raw data and its associated operating conditions is fundamental to all subsequent adaptive processing.

[0027] Specifically, in order to capture features at different time scales, the process of acquiring the neutron count spectrum data to be tested includes the following sub-steps: S1011, concurrently deploy at least three sliding acquisition windows with different integration time durations; S1012 uses each integral time sliding acquisition window to synchronously accumulate and record the captured neutron pulse signal, forming multi-scale parallel corresponding neutron count spectrum data to identify transient interference events based on the triggering characteristics under the integral time sliding acquisition window of different durations, and to block the update of the multi-source background template library by the transient interference events.

[0028] In practice, a sliding window can be set, for example, a first sliding window (e.g., at the 10-millisecond level, used to capture microphone effect noise or electromagnetic pulse interference within a very short time frame), a second sliding window (e.g., at the 1-second level, used to reflect normal neutron flux statistical fluctuations), and a third sliding window (e.g., at the 10-second or longer level, used to smooth long-term statistical background trends). Through multi-scale parallel recording, when an unnatural spike in the count rate occurs within a very short time window, while the mean of the long time window has not been significantly affected, it can be determined that transient interference from non-nuclear events (such as false signals caused by physical vibrations) has been captured. In this case, the data within that time period will be marked as unreliable, forcibly preventing it from entering the subsequent background template update process, thereby protecting the purity of the background template.

[0029] Meanwhile, in one optional implementation, the detector operating data specifically includes the detector operating temperature, signal amplification gain status, and system dead time. These three parameters are the core variables that directly affect the spectrum shape and count values.

[0030] The measured neutron count spectrum data is preprocessed using detector operating data to obtain a normalized measured count spectrum. This process includes the following sub-steps: S1013 utilizes the detector's operating temperature and signal amplification gain state to perform channel alignment calibration on the neutron count spectrum data under test. Because the detector and its downstream electronic amplification circuitry are temperature-sensitive, temperature drift can cause an overall linear or nonlinear shift in pulse amplitude, which manifests as a left-right shift of the energy peak center channel in the energy spectrum. In this step, the channel shift at the current operating temperature and gain state is determined using a pre-established temperature-gain drift compensation lookup table or compensation curve. Subsequently, the channel data of each channel in the neutron count spectrum under test are reverse-shifted and compensated to the standard channel at the reference temperature, ensuring that the characteristic peaks of specific energies (such as the 764 keV energy peak released by the reaction of neutrons with helium-3) are tightly aligned with the fixed channel, eliminating the comparison error caused by spectral drift.

[0031] S1014 uses the system dead time to perform count loss compensation correction on the neutron count spectrum data after channel alignment calibration, and performs smoothing filtering to output the normalized measured count spectrum.

[0032] In neutron multiplexing measurements or high count rate environments, the pulse resolution capability of the detector and its electronic systems has a physical limit, namely the system dead time. This can cause some arriving pulses to overlap and fail to be recorded independently. In this embodiment, dead time correction is solved by a textual conversion using a non-paralyzed or paralyzed dead time theoretical model: Specifically, the measured apparent count rate is multiplied by the system dead time to obtain a time duty cycle, then this time duty cycle is subtracted to obtain the effective recording time ratio, and finally the apparent count rate is divided by this effective recording time ratio to restore the true count rate after compensating for missed pulses. After dead time compensation, in order to suppress the high-frequency statistical fluctuation noise caused by the Poisson distribution, a moving average filtering algorithm or a multinomial smoothing algorithm is used to smooth the spectrum, and its total integral area or time base is normalized and scaled to finally output a normalized measured count spectrum that eliminates operating condition distortion.

[0033] S102, Obtain the historical background counting spectrum, divide the historical background counting spectrum into background components from multiple different sources, and construct an independent basic background spectrum template for each background component to form a multi-source background template library; The background is not an indivisible, single whole, but a complex aggregate composed of multiple components with different physical origins and temporal characteristics. The reason why traditional fixed deduction methods fail is precisely because they cannot separate these independently changing components.

[0034] In one alternative implementation, this step specifically includes the following sub-steps: S1021, the historical background counting spectrum is separated into the environmental radiation background component, the detector intrinsic background component, the overlapping neutron background component, and the electron noise background component; specifically: (1) Background radiation component: mainly from secondary neutrons produced by cosmic rays penetrating the atmosphere, and radiation released from the surrounding environment (such as concrete walls, natural radioactive nuclides such as uranium, thorium, and potassium in the soil). This component is affected by meteorological conditions (such as air pressure and the washing effect of rainfall) and has a slow time-varying characteristic.

[0035] (2) Intrinsic background component of the detector: mainly originates from trace amounts of radioactive impurities (such as trace amounts of alpha radioactive decay) inherent in the detector's outer shell material, internal filling gas, or photomultiplier tube glass. This component exhibits extremely high stability throughout the detector's lifetime, and its spectral shape remains unchanged.

[0036] (3) Overlapping neutron background component: In time-of-flight (TOF) neutron energy spectrum measurements or certain specific neutron source pulse measurements, slow neutrons from the previous pulse cycle may leak or overlap into the current cycle, forming a tail or constant background, the intensity of which is strongly correlated with the source excitation frequency and historical intensity.

[0037] (4) Electronic noise background component: Electronic thermal noise originating from the front-end preamplifier, main amplifier and analog-to-digital conversion circuit, usually concentrated at the low energy end of the energy spectrum, and exhibits obvious random fluctuation characteristics with environmental electromagnetic interference and equipment temperature.

[0038] S1022, for each background component, extract its attenuation characteristics or random fluctuation characteristics, use a clustering algorithm to establish a separate energy spectrum profile feature matrix for the background component with corresponding characteristics, and store each energy spectrum profile feature matrix as an independent basic background spectrum template in the multi-source background template library.

[0039] For the large number of historical background data fragments separated above, multi-dimensional feature vectors are constructed according to their evolutionary behavior over time (i.e., the time constant of physical decay or the variance distribution of statistical fluctuations). Then, unsupervised clustering algorithms (such as density-based noisy spatial clustering algorithms, DBSCAN, or K-means clustering) are used to divide tens of thousands of historical spectra into several highly cohesive clusters. The arithmetic mean or median of all spectra within each cluster is extracted as a typical profile of that background component, and converted into a one-dimensional or multi-dimensional energy spectrum contour feature matrix. Finally, the multiple feature matrices corresponding to various components together constitute a multi-source background template library, providing the basic materials for subsequent dynamic, modular background reconstruction.

[0040] S103, calculate the spectral similarity between the normalized measured count spectrum and each basic background spectral template in the multi-source background template library, and determine the current subtraction coefficient of each basic background spectral template based on the spectral similarity. Given the normalized measured count spectrum of the currently collected unknown components, this step aims to use mathematical calculations to determine the proportion of various basic backgrounds contained within it, i.e., to solve for the weights of each template (current deduction coefficients).

[0041] In one alternative implementation, this step specifically includes: S1031, perform inner product dot multiplication operation between the normalized measured count spectrum and each basic background spectrum template in the multi-source background template library, and quantize the corresponding spectral similarity. This approach utilizes the physical meaning of the vector space model. The normalized measured count spectrum is considered a high-dimensional vector in a multi-dimensional space (its dimension equal to the number of channels in the energy spectrum). Similarly, each basic background spectral template is also considered a high-dimensional vector of the same dimension. By multiplying the measured spectral vector with the count values ​​of the corresponding channels of each template vector and summing the results (i.e., performing inner product dot product operations), a scalar value is obtained. The larger this value, the more closely the measured spectrum matches the template in terms of contour undulations and peak positions, thus quantifying the spectral similarity in purely geometric morphology.

[0042] S1032, the dynamic weight allocation is obtained through a pre-trained multi-layer feedforward neural network model. The specific acquisition process is as follows: extract the feature tensor of the normalized measured count spectrum, the environmental feature tensor of the historical background count spectrum, and the detector operating condition data, and concatenate them into a multi-dimensional composite input feature vector; feed the multi-dimensional composite input feature vector into the multi-layer feedforward neural network model (FNN), and perform nonlinear correction feature extraction through the hidden network layer; output the dynamic weight allocation for each basic background spectrum template in the multi-source background template library through the output layer of the multi-layer feedforward neural network model.

[0043] Besides purely geometric spectral similarity, the context of the physical environment (i.e., operating conditions and historical patterns) also determines the true proportions of each background component. This embodiment introduces an artificial intelligence deep learning model to capture this complex nonlinear mapping relationship. Specifically, the normalized measured spectrum and historical environmental background are first abstracted by dimensionality reduction, extracting statistical feature tensors such as peak-to-valley ratio and low-to-high energy band count ratio. These tensors are then combined with the aforementioned detector operating condition data, such as temperature, gain, and dead time, and concatenated into a long, one-dimensional, multi-dimensional composite input feature vector. Subsequently, this input vector is fed into a multi-layer feedforward neural network model pre-trained using a large amount of Monte Carlo simulation data and historical real-world labeled data. The hidden layers of this model contain a large number of neurons and nonlinear activation functions (e.g., the linear rectified function, ReLU), which can fully fit the implicit physical interweaving patterns between operating parameters and background components (e.g., the nonlinear pattern of increased electronic noise component weight due to temperature increase). Finally, the network output layer (usually combined with the normalized exponential function Softmax) outputs a set of coefficients that sum to one. This is the dynamic weight allocation, which represents the theoretical probability or weight of each template predicted by the neural network based on the physical context.

[0044] S1033, combine the spectral similarity with the dynamically allocated weights generated for the normalized measured count spectrum to perform normalization mapping, generate the proportional weights of each basic background spectrum template, and use the proportional weights as the current subtraction coefficients of each basic background spectrum template.

[0045] The inner product similarity of geometric appearance and the neural network prediction weights of physical context each have their own emphasis. This step fuses the two with weights (for example, multiplying them and then normalizing so that the sum of the fusion weights of all templates approximates the total scale of the current background), thus obtaining a comprehensive and balanced proportional weight. This proportional weight takes into account both the actual appearance of the current spectrum and does not violate the current physical operating conditions of the detector. Finally, this set of finely calculated proportional weights is locked as the current subtraction coefficients for the subsequent subtraction operation.

[0046] S104, subtract each basic background spectrum template weighted according to each current deduction coefficient from the normalized measured count spectrum to obtain the target neutron net count spectrum; Once the content ratio of each background component is determined, an adaptive stripping deduction operation can be performed. To avoid under- or over-deduction caused by fixed deduction, this embodiment creatively introduces adaptive parameter constraints.

[0047] In one alternative implementation, this step specifically includes: S1041, analyze the full width at half maximum (FWHM) of the energy peak and the variance of the counting statistics of the normalized measured count spectrum, and automatically constrain and calculate the energy subtraction window based on the full width at half maximum (FWHM) of the energy peak and the variance of the counting statistics; In nuclear energy spectra, the neutron response signal of interest typically exhibits a Gaussian energy peak distribution. Fixed-width subtraction methods cannot accommodate peak broadening or narrowing caused by temperature or gain drift. In this step, the center and boundary of the neutron characteristic peak are first located using the zero-crossing point of the first derivative and the minimum value of the second derivative. Then, the full width at half the maximum peak intensity, i.e., the Free Wide Wavelength (FWHM), is calculated. Simultaneously, the variance of the counting statistical fluctuations within this energy region is used to assess the local noise level. Then, using the peak center as a reference, the window is extended to the left and right by several times the FWHM distance, and the extension step size is dynamically fine-tuned based on the variance amplitude. This automatically defines an energy subtraction window that encompasses the entire effective neutron signal while maximally rejecting unwanted background noise.

[0048] S1042, within the energy subtraction window, subtracts each basic background spectrum template, weighted according to each current subtraction coefficient, from the normalized measured count spectrum step by step, and outputs the target neutron net count spectrum.

[0049] After defining the energy subtraction window, the subtraction operation is performed only within the energy channels covered by that window. The specific execution logic is as follows: the first basic background spectrum template in the multi-source background template library is multiplied by its corresponding current subtraction coefficient to obtain the weighted component spectrum; the value of the weighted component spectrum is subtracted from the value of the corresponding energy channel in the normalized measured count spectrum. This subtraction operation is then repeated for the second, third, and all activated basic background spectrum templates. If a physically unreasonable negative value appears in the count value of a certain energy channel during the subtraction process, it is forcibly set to zero and smoothed. After this meticulous, step-by-step stripping, only the signal purely caused by neutrons is retained in the measured spectrum, which was originally mixed with a large amount of interference, forming the final output target net neutron count spectrum.

[0050] S105, Statistically analyze the net count spectrum of the target neutrons to obtain the net count of the target neutrons, evaluate the error in the process of obtaining the net count spectrum of the target neutrons to obtain the error synthesis term, synthesize the error synthesis term into the total net count uncertainty, generate the net count confidence interval based on the net count of the target neutrons and the total net count uncertainty, and output it. For searching for faint, rare events or identifying high-precision nuclear materials, simply providing a net count value is far from sufficient, because no measurement is absolutely accurate. The confidence boundary of that net count value must be quantified simultaneously.

[0051] In one alternative implementation, the evaluation process specifically includes the following sub-steps: S1051, using the characteristic laws of the Poisson distribution, calculate the statistical variance of the net count of target neutrons to obtain the counting statistical error term; Radioactive decay and nuclear reactions essentially follow the Poisson stochastic statistical law. According to the mathematical properties of the Poisson distribution, the statistical variance of an observed event is numerically equal to the expected value of its number of observations. Therefore, by summing the count values ​​within the region of interest in the net neutron count spectrum of the target, we obtain the net neutron count of the target. Then, by directly taking the value of this net neutron count as the variance estimate and taking its square root, we can obtain the unavoidable error amplitude caused by random statistical fluctuations, i.e., the counting statistical error term.

[0052] S1052, combine the distribution deviation of each current deduction coefficient to calculate the background deduction error term introduced by the deviation of the background model deduction ratio; The subtraction factor itself is obtained through model prediction and inevitably contains uncertainty. Using the error propagation principle, the variance or covariance of the weight of each basic background spectrum template in this measurement is calculated, and then these distribution deviations are mapped and amplified to the net count value to assess the systematic error limit caused by over- or under-subtraction of background. This is the background subtraction error term.

[0053] S1053, calculate the root mean square (RMS) of the difference between the normalized measured count spectrum and the sum of each basic background spectrum template after weighting according to each current deduction coefficient, and obtain the background model matching error term; This measures the degree to which the reconstructed background closely matches the real background. All weighted base background spectrum templates are superimposed to form the reconstructed total background spectrum. In the pure background energy region where there is no neutron signal, the normalized measured count spectrum is subtracted from the reconstructed total background spectrum channel by channel. Then, all differences are squared, the average of the squares is calculated, and finally, the square root is taken to obtain the RMS (Residual Model Score). This value reflects the residual error in model fit caused by the model's failure to perfectly fit the actual spectrum in terms of shape details; that is, the background model-matching error term.

[0054] S1054, the counting statistics error term, background subtraction error term, and background model matching error term are combined as error synthesis terms to form the net total counting uncertainty.

[0055] These three categories of errors originate from different, unrelated physical and mathematical mechanisms. Therefore, according to the evaluation rules for combined standard uncertainty, the squares of the count statistical error term, the squares of the background subtraction error term, and the squares of the background model matching error term are added together. Then, the square root of the sum is taken to obtain a single comprehensive index that fully covers fluctuations, calculation biases, and model defects—net total count uncertainty.

[0056] S1055 generates and outputs the net count confidence interval (CI) based on the net target neutron count and the total net count uncertainty.

[0057] Finally, a coverage factor is set according to the user's accuracy requirements (for example, a coverage factor of 2 is usually chosen, corresponding to approximately 95% confidence probability). Using the net neutron count of the target as the central estimate, and adding and subtracting the product of the coverage factor and the total uncertainty of the net count, an upper and lower limit for the value can be defined. The system ultimately outputs not only a net count in point form, but also a confidence interval in band form, providing a solid and reliable basis for subsequent radionuclide identification algorithms or security warning threshold judgments, avoiding false positives or false negatives caused by statistical fluctuations.

[0058] S106, calculate the difference between the normalized measured count spectrum and the sum of each basic background spectrum template after weighting according to each current deduction coefficient to obtain the residual characteristics; statistically analyze the fluctuation of the net target neutron count within the historical time window to obtain the variation characteristics; evaluate whether background drift has occurred based on the residual characteristics and variation characteristics, and when it is determined that background drift has occurred, update the multi-source background template library in combination with the detector operating data.

[0059] In long-term mobile probes or continuous underground dark matter detection experiments lasting several months, the environmental background is far from static (for example, rainfall can cause the leaching and deposition of radon progeny in the air, leading to a several-fold increase in surface radioactivity). If the template library is static and rigid, the model will quickly become ineffective. This step establishes a closed-loop mechanism for the model.

[0060] In one alternative implementation, the drift detection and update process specifically includes: S1061, subtract the sum of the normalized measured count spectrum and the sum of the basic background spectrum templates after weighting according to each current deduction coefficient to form the reconstructed residual sequence on the discrete energy channel, and use the reconstructed residual sequence as the residual feature; Here, we also construct the reconstructed total background spectrum and subtract it from the measured spectrum. Unlike the previous calculation of a single RMS value, we retain the sequence of difference values ​​across all energy channels. If the current physical environment introduces a completely new interference spectrum that does not exist in the template library, then obvious local bumps or abnormal fluctuations in specific energy regions will appear in this reconstructed residual sequence. This multidimensional morphological feature is the residual feature.

[0061] S1062, the ratio of the standard deviation to the mean of the net target neutron count within multiple consecutive historical time windows is used to obtain the variation index, and the variation index is used as the variation feature. This metric is evaluated based on the stability of the macroscopic time series. If the background subtraction algorithm works correctly, during the background measurement phase without real neutron source intrusion, the net target neutron count after subtraction should always exhibit small random oscillations around a mean close to zero. This step calculates the standard deviation of the net count over dozens of consecutive time windows and divides it by the mean (i.e., calculates the coefficient of variation). When the background undergoes gradual or abrupt drift, and the model fails to subtract it in time, this coefficient of variation will show a significantly amplified trend, forming a variation characteristic reflecting the model's time series failure.

[0062] S1063, when the dispersion of the reconstructed residual sequence exceeds the first drift detection threshold, or the variation index exceeds the second drift detection threshold, background drift is determined to have occurred; A dual-insurance judgment mechanism is established: the first drift identification threshold is responsible for capturing foreign object intrusion in the energy spectrum morphology (such as new scattering spectra introduced by changes in the shielding structure), and the second drift identification threshold is responsible for capturing uncontrolled divergence at the overall count level. As long as either condition is met, the system alarm is sounded, officially declaring that the current working environment has deviated from the coverage of the original multi-source background template library, and background drift is judged to have occurred.

[0063] S1064, retrieve the multi-source background template library, call up the backup background template that matches the determined background drift type and detector operating data, and replace it to update the multi-source background template library.

[0064] Upon occurrence of drift, a conservative strategy is first implemented. If this drift has occurred before (e.g., the environment where the device is located has experienced a regular temperature rise), then the dormant area of ​​the multi-source background template library may already contain a backup background template for this operating condition. In this case, the matching backup template is activated directly by indexing the detector's operating condition data, replacing the currently inapplicable template, thus achieving a smooth and rapid model switching.

[0065] More importantly, in the event of unprecedented unknown background drift, one alternative implementation includes, after S1064: S1065, if it is determined that background drift has occurred and there is no matching backup background template in the multi-source background template library, extract multiple normalized measured count spectra obtained continuously under the background drift state to construct a feature matrix. When the system recognizes that this is a completely new background environment, it begins to collect all measured count spectra of the environment within a certain period of time (such as tens of minutes), arranges them in columns, and combines them to form a large two-dimensional feature matrix.

[0066] S1066 uses the Non-negative Matrix Factorization (NFF) algorithm to perform blind source decomposition on the feature matrix, and extracts independent anomalous background basis vectors; Since radiation counting is not inherently negative, the Non-Negative Source Matrix (NMF) algorithm is the most suitable blind source separation method in line with physical laws. The core logic of the NMF algorithm is to iteratively decompose the high-dimensional feature matrix into the product of two low-rank non-negative matrices: a reference matrix (representing the individual pure reference spectra) and a coefficient matrix (representing the mixing ratio of each reference at different time points). During the algorithm's iteration, the original feature matrix is ​​continuously compared with the product of the two low-rank matrices through division, and the reference and coefficient matrices are updated according to the comparison ratio until the reconstruction error converges to a minimum. Through this decomposition, a pure, independent one-dimensional vector representing this nascent interference—the independent anomalous background basis vector—can be extracted from the chaotic mixed background, much like peeling back layers.

[0067] S1067 encapsulates the independent anomalous background basis vectors to generate a dynamically supplemented basic background spectrum template, marks the corresponding detector operating data, and compiles the dynamically supplemented basic background spectrum template into the multi-source background template library.

[0068] Finally, the anomalous background basis vector purified by the NMF algorithm is normalized and transformed into a standard spectral template format. Simultaneously, environmental labels such as the detector operating temperature and signal amplification gain status at the time of its creation are firmly bound to it like an identification document, forming a complete, dynamically supplementary basic background spectral template. This new template is then formally incorporated into the multi-source background template library for use in the current subtraction coefficient calculation process and will be kept on long-term standby.

[0069] This invention also provides a neutron net counting system based on adaptive subtraction of background counting spectrum, referring to... Figure 3 The diagram illustrates a functional block diagram of a neutron net counting system 300 based on adaptive subtraction of background counting spectrum according to the present invention. The system may include the following modules: The preprocessing module 301 is configured to collect the neutron count spectrum data to be measured and the detector operating condition data, and use the detector operating condition data to preprocess the neutron count spectrum data to be measured to obtain the normalized measured count spectrum. Background multi-source modeling module 302 is configured to acquire historical background counting spectrum, divide the historical background counting spectrum into background components from multiple different sources, and construct an independent basic background spectrum template for each background component to form a multi-source background template library. The adaptive subtraction module 303 is configured to calculate the spectral similarity between the normalized measured count spectrum and each basic background spectrum template in the multi-source background template library, determine the current subtraction coefficient of each basic background spectrum template based on the spectral similarity, and subtract each basic background spectrum template weighted according to each current subtraction coefficient from the normalized measured count spectrum to obtain the target neutron net count spectrum. The error assessment and output module 304 is configured to perform statistics on the net count spectrum of the target neutrons to obtain the net count of the target neutrons, assess the error in the process of obtaining the net count spectrum of the target neutrons to obtain the error synthesis term, synthesize the error synthesis term into the total net count uncertainty, generate the net count confidence interval based on the net count of the target neutrons and the total net count uncertainty, and output it. The closed-loop feedback update module 305 is configured to calculate the difference between the normalized measured count spectrum and the sum of each basic background spectrum template after weighting according to each current deduction coefficient, to obtain residual characteristics; to statistically analyze the fluctuation of the net target neutron count within the historical time window, to obtain variation characteristics; to evaluate whether background drift has occurred based on the residual characteristics and variation characteristics; and when background drift is determined to have occurred, to update the multi-source background template library in conjunction with the detector operating data.

[0070] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; The processor, when executing a program stored in memory, implements the neutron net counting method based on adaptive subtraction of background counting spectrum of the present invention.

[0071] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include RAM, or it can include NVM, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0072] The processors mentioned above can be general-purpose processors, including CPUs, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0073] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the neutron net counting method based on adaptive subtraction of background counting spectrum according to embodiments of the present invention.

[0074] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, electronic devices, apparatuses, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] The above are merely specific embodiments of the present invention. Any equivalent substitutions or modifications made by those skilled in the art within the scope of this invention should fall within the protection scope of the present invention. The protection scope of the present invention is defined by the claims.

Claims

1. A neutron net counting method based on adaptive subtraction of background counting spectrum, characterized in that, Includes the following steps: Collect the neutron count spectrum data to be tested and the detector operating condition data, and use the detector operating condition data to preprocess the neutron count spectrum data to be tested to obtain the normalized measured count spectrum. Obtain the historical background counting spectrum, divide the historical background counting spectrum into background components from multiple different sources, and construct an independent basic background spectrum template for each background component to form a multi-source background template library; Calculate the spectral similarity between the normalized measured count spectrum and each basic background spectral template in the multi-source background template library, and determine the current subtraction coefficient of each basic background spectral template based on the spectral similarity. The target neutron net count spectrum is obtained by subtracting each of the basic background spectrum templates weighted according to each of the current deduction coefficients from the normalized measured count spectrum. The net count spectrum of the target neutrons is statistically analyzed to obtain the net count of the target neutrons, and the error in the process of obtaining the net count spectrum of the target neutrons is evaluated to obtain the error synthesis term. The error synthesis term is synthesized into the total net count uncertainty. The net count confidence interval is generated and output based on the net count of the target neutrons and the total net count uncertainty. The residual characteristics are obtained by calculating the difference between the normalized measured count spectrum and the sum of the basic background spectrum templates after weighting according to each of the current deduction coefficients; The fluctuation of the net count of the target neutrons within the statistical historical time window is used to obtain the variation characteristics; the background drift is evaluated based on the residual characteristics and the variation characteristics; when the background drift is determined to have occurred, the multi-source background template library is updated in conjunction with the detector operating data.

2. The method as described in claim 1, characterized in that, The detector operating data specifically includes detector operating temperature, signal amplification gain status, and system dead time. The step of preprocessing the measured neutron count spectrum data using the detector's operating condition data to obtain the normalized measured count spectrum specifically includes: The energy channel alignment calibration of the neutron count spectrum data under test is performed using the detector operating temperature and the signal amplification gain state. The system dead time is used to compensate for the count loss of the neutron count spectrum data after channel alignment calibration, and smoothing filtering is performed to output the normalized measured count spectrum.

3. The method as described in claim 1, characterized in that, The step of dividing the historical background counting spectrum into background components from multiple different sources and constructing an independent basic background spectrum template for each background component to form a multi-source background template library specifically includes: The historical background counting spectrum is separated into environmental radiation background component, detector intrinsic background component, overlapping neutron background component, and electronic noise background component. For each background component, its attenuation characteristics or random fluctuation characteristics are extracted. A clustering algorithm is used to establish a separate energy spectrum profile feature matrix for the background component with corresponding characteristics. Each energy spectrum profile feature matrix is ​​stored as an independent basic background spectrum template in the multi-source background template library.

4. The method as described in claim 1, characterized in that, The calculation of the spectral similarity between the normalized measured count spectrum and each basic background spectral template in the multi-source background template library, and the determination of the current subtraction coefficient for each basic background spectral template based on the spectral similarity, specifically includes: The normalized measured count spectrum is multiplied by the inner product of each basic background spectrum template in the multi-source background template library, and the corresponding spectrum shape similarity is quantized and output. The spectral similarity is combined with the dynamically allocated weights generated for the normalized measured count spectrum to perform normalization mapping, generating the proportional weights of each of the basic background spectrum templates, and the proportional weights are used as the current deduction coefficients of each of the basic background spectrum templates. The step of subtracting each of the basic background spectrum templates weighted according to each of the current deduction coefficients from the normalized measured count spectrum to obtain the target neutron net count spectrum specifically includes: The energy peak full width at half maximum (FWHM) and the statistical fluctuation variance of the counts are analyzed in the normalized measured count spectrum. The energy deduction window is automatically constrained and calculated based on the energy peak full width at half maximum (FWHM) and the statistical fluctuation variance of the counts. Within the energy subtraction window, the target neutron net count spectrum is output by subtracting each of the basic background spectrum templates weighted according to each of the current subtraction coefficients from the normalized measured count spectrum.

5. The method as described in claim 4, characterized in that, The dynamically assigned weights are obtained through a pre-trained multi-layer feedforward neural network model. The specific acquisition process is as follows: Extract the feature tensor of the normalized measured count spectrum, the environmental feature tensor of the historical background count spectrum, and the detector operating condition data, and concatenate them into a multi-dimensional composite input feature vector. The multidimensional composite input feature vector is fed into the multilayer feedforward neural network model, and nonlinear correction feature extraction is performed through the hidden network layer; The output layer of the multi-layer feedforward neural network model outputs the dynamically assigned weights for each basic background spectrum template in the multi-source background template library.

6. The method as described in claim 1, characterized in that, The evaluation process for obtaining the net count spectrum of the target neutrons yields an error synthesis term, which is then combined into a net total count uncertainty. Specifically, this includes: The statistical variance of the net count of the target neutrons is calculated using the characteristic laws of the Poisson distribution, and the counting statistical error term is obtained; Based on the distribution deviation of each of the current deduction coefficients, calculate the background deduction error term introduced by the deviation of the background model deduction ratio; The root mean square difference between the normalized measured count spectrum and the sum of the basic background spectrum templates after weighting according to each of the current deduction coefficients is calculated to obtain the background model matching error term; The counting statistical error term, the background subtraction error term, and the background model matching error term are used as the error synthesis term to synthesize the net total counting uncertainty.

7. The method as described in claim 1, characterized in that, The residual characteristics are obtained by calculating the difference between the normalized measured count spectrum and the sum of the basic background spectrum templates after weighting according to each of the current deduction coefficients; The fluctuation of the net target neutron count within a historical time window is statistically analyzed to obtain variation characteristics. Based on the residual characteristics and the variation characteristics, it is assessed whether background drift has occurred. When background drift is determined to have occurred, the multi-source background template library is updated in conjunction with the detector operating data, specifically including: The normalized measured count spectrum is subtracted from the sum of the basic background spectrum templates after being weighted according to each of the current deduction coefficients to form a reconstructed residual sequence on the discrete energy channel, and the reconstructed residual sequence is used as the residual feature; The ratio of the standard deviation to the mean of the net count of the target neutrons within multiple consecutive historical time windows is used to obtain the variation index, and the variation index is used as the variation feature. When the dispersion of the reconstructed residual sequence exceeds the first drift detection threshold, or the variation index exceeds the second drift detection threshold, background drift is determined to have occurred. The multi-source background template library is retrieved, and a backup background template that matches the determined background drift type and the detector operating data is used for replacement, thereby updating the multi-source background template library.

8. The method as described in claim 7, characterized in that, After the backup background template matching the retrieved and determined background drift type and the detector operating data is replaced, the process further includes: If it is determined that background drift has occurred and there is no matching backup background template in the multi-source background template library, extract multiple normalized measured count spectra obtained continuously under the background drift state to construct a feature matrix; The feature matrix is ​​blindly decomposed using a nonnegative matrix factorization algorithm to extract independent abnormal background basis vectors. The independent abnormal background basis vectors are encapsulated to generate a dynamically supplemented basic background spectrum template, and the corresponding detector operating data are labeled. The dynamically supplemented basic background spectrum template is then compiled into the multi-source background template library.

9. The method as described in claim 1, characterized in that, The acquisition of the neutron count spectrum data to be tested specifically includes: Deploy at least three sliding data acquisition windows with different integration time durations concurrently in the system; The captured neutron pulse signals are synchronously accumulated and recorded using each of the integral time sliding acquisition windows to form multi-scale parallel corresponding neutron count spectrum data to be measured. This is to identify transient interference events based on the triggering characteristics under the integral time sliding acquisition windows of different durations, and to block the update of the multi-source background template library by the transient interference events.

10. A neutron net counting system based on adaptive subtraction of background counting spectrum, characterized in that, The system comprises a plurality of processing modules for performing the method as described in any one of claims 1 to 9, the system comprising: The preprocessing module is configured to collect the neutron count spectrum data to be tested and the detector operating condition data, and use the detector operating condition data to preprocess the neutron count spectrum data to be tested to obtain the normalized measured count spectrum. The background multi-source modeling module is configured to acquire historical background counting spectra, divide the historical background counting spectra into background components from multiple different sources, and construct an independent basic background spectrum template for each background component to form a multi-source background template library. The adaptive subtraction module is configured to calculate the spectral similarity between the normalized measured count spectrum and each basic background spectrum template in the multi-source background template library, determine the current subtraction coefficient of each basic background spectrum template based on the spectral similarity, and subtract each basic background spectrum template weighted according to each current subtraction coefficient from the normalized measured count spectrum to obtain the target neutron net count spectrum. The error assessment and output module is configured to perform statistics on the target neutron net count spectrum to obtain the target neutron net count, assess the error in the process of obtaining the target neutron net count spectrum to obtain an error synthesis term, synthesize the error synthesis term into a net count total uncertainty, generate a net count confidence interval based on the target neutron net count and the net count total uncertainty, and output it. The closed-loop feedback update module is configured to calculate the difference between the normalized measured count spectrum and the sum of each of the basic background spectrum templates after weighting according to each of the current deduction coefficients, to obtain residual characteristics; to statistically analyze the fluctuation of the net count of the target neutrons within the historical time window, to obtain variation characteristics; to evaluate whether background drift has occurred based on the residual characteristics and the variation characteristics; and when background drift is determined to have occurred, to update the multi-source background template library in conjunction with the detector operating data.

Images