Tritiated water measurement method, device, equipment, medium and product

Abstract

The invention discloses a tritiated water measurement method, device, equipment, medium and product, and relates to the field of radiation detection, the method comprises the following steps: obtaining an actual nuclear pulse signal detected by a double-layer scintillator detector, and carrying out classification by using a trained convolutional neural network according to the actual nuclear pulse signal to obtain an actual energy deposition mode type; the actual energy deposition mode types include that the particles do not deposit energy in the scintillators, the particles only deposit energy in the second layer of scintillators, the particles only deposit energy in the first layer of scintillators and the particles have energy deposition in the two layers of scintillators, and the tritium beta rays are determined according to the actual energy deposition mode types. According to the invention, accurate measurement of tritium water can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of radiation detection, in particular to a tritium water measurement method, device, equipment, medium and product. BACKGROUND

[0002] In nuclear radiation detection, the waveform of the nuclear pulse signal contains a wealth of information, especially in the pulse shape, which reflects the position of energy deposition, particle species and the dynamic process of interaction with the material. In the prior art, the traditional pulse shape discrimination (PSD) method mainly relies on manually setting thresholds for the proportion of fast and slow components, the rising slope and the decay characteristics of the nuclear pulse signal to distinguish different types of particles (such as low-energy beta rays and environmental gamma rays). However, since the nuclear pulse signals generated by beta rays and gamma rays in a solid scintillation detector often have a large overlap, and the subtle differences in the signal are affected by multiple factors such as noise, uneven energy deposition and detector material characteristics, the traditional PSD method is difficult to accurately analyze the details of the signal, which further leads to inaccurate determination of particle species, and restricts the practical application of the detector in complex mixed radiation fields.

[0003] In addition, in tritium water detection, tritium mainly decays in the form of low-energy beta rays, and the beta particles emitted have low energy, while the environment is often accompanied by background gamma rays. These two particles will form different pulse waveforms when they deposit energy in the scintillator, but the difference is often very subtle. How to accurately extract and analyze these subtle waveform characteristics and then effectively distinguish beta rays from gamma rays has become a technical problem to be solved. SUMMARY

[0004] The purpose of the present application is to provide a tritium water measurement method, device, equipment, medium and product, which can realize accurate measurement of tritium water.

[0005] To achieve the above purpose, the present application provides the following solutions:

[0006] In a first aspect, the present application provides a tritium water measurement method, comprising:

[0007] obtaining an actual nuclear pulse signal detected by a double-layer scintillator detector;

[0008] classifying the actual nuclear pulse signal using a trained convolutional neural network to obtain an actual energy deposition mode type; the actual energy deposition mode type includes: no energy deposition of particles in the scintillator, energy deposition of particles only in the second layer scintillator, energy deposition of particles only in the first layer scintillator, and energy deposition of particles in both layers of the scintillator;

[0009] determining a tritium beta ray according to the actual energy deposition mode type.

[0010] In an embodiment, the trained convolutional neural network comprises, in sequence, an input layer, a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a global average pooling layer, a full connection layer, and an output layer.

[0011] In an embodiment, the first convolutional layer and the first max-pooling layer are connected through a ReLU activation function; the second convolutional layer and the second max-pooling layer are connected through a ReLU activation function; and the full connection layer and the output layer are connected through a Softmax activation function.

[0012] In an embodiment, the trained convolutional neural network is used to classify the actual nuclear pulse signal according to the actual nuclear pulse signal, to obtain the actual energy deposition pattern type, specifically comprising:

[0013] The trained convolutional neural network is used to extract features of the actual nuclear pulse signal, to obtain local features of the signal; the local features include a fast-slow component ratio, a rising edge characteristic of a pulse, a peak amplitude, and a decay characteristic;

[0014] The trained convolutional neural network is used to classify the actual nuclear pulse signal according to the local features, to obtain the actual energy deposition pattern type.

[0015] In an embodiment, the training process of the convolutional neural network specifically comprises:

[0016] The simulated nuclear pulse signal is used as the input of the one-dimensional convolutional neural network, and the simulated energy deposition pattern type is used as the output of the one-dimensional convolutional neural network; a cross-entropy loss function and a gradient descent method are used to optimize the hyperparameters of the one-dimensional convolutional neural network, to obtain the trained convolutional neural network.

[0017] In an embodiment, the tritium beta ray is determined according to the actual energy deposition pattern type, specifically comprising:

[0018] When the actual energy deposition pattern type is that a particle only deposits energy in the first layer of scintillator, it is determined that the actual nuclear pulse signal is a tritium beta ray.

[0019] In a second aspect, the present application provides a tritium water measuring device, comprising:

[0020] An acquisition module is configured to acquire an actual nuclear pulse signal detected by a double-layer scintillator detector;

[0021] The classification module is configured to classify the actual nuclear pulse signal by using the trained convolutional neural network to obtain an actual energy deposition mode type, wherein the actual energy deposition mode type includes: no energy deposition of a particle in a scintillator, energy deposition of the particle only in a second layer of the scintillator, energy deposition of the particle only in a first layer of the scintillator, and energy deposition of the particle in both layers of the scintillator.

[0022] The tritium beta ray determination module is configured to determine the tritium beta ray according to the actual energy deposition mode type.

[0023] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the tritium water measurement method.

[0024] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the tritium water measurement method.

[0025] In a fifth aspect, the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the tritium water measurement method.

[0026] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0027] The present application provides a tritium water measurement method, device, equipment, medium and product, which classifies the actual nuclear pulse signal by using the trained convolutional neural network to obtain an actual energy deposition mode type, wherein the actual energy deposition mode type includes: no energy deposition of a particle in a scintillator, energy deposition of the particle only in a second layer of the scintillator, energy deposition of the particle only in a first layer of the scintillator, and energy deposition of the particle in both layers of the scintillator; and determines the tritium beta ray according to the actual energy deposition mode type. The trained convolutional neural network classifies the pulse shape of the different energy deposition of the particle in the scintillator, accurately distinguishes the beta ray and the gamma ray, and thus improves the accurate measurement of the tritium beta ray. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0029] Figure 1 The principle diagram of the beta-gamma mixed field detection of the double-layer scintillator single tritium measurement probe;

[0030] Figure 2 Fig. 3 is a diagram of three possible energy deposition scenarios of particles in a double-layer scintillator detector;

[0031] Figure 3 Fig. 4 is a diagram of a 1D-CNN model;

[0032] Figure 4 Fig. 5 is a diagram of actual operation of a double-layer scintillator detector tritium measurement system;

[0033] Figure 5 Fig. 6 is a diagram of a NPU predicting results according to nuclear pulse signals;

[0034] Figure 6 Fig. 7 is a flowchart of a tritiated water measurement method;

[0035] Figure 7 Fig. 8 is a functional module diagram of a tritiated water measurement device according to an embodiment of the present application;

[0036] Figure 8 Fig. 9 is a structural diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0038] Currently, there are many problems in tritiated water detection:

[0039] Lack of multi-signal distinguishing ability: existing technologies usually have difficulty in distinguishing β-ray and γ-ray signals at the same time, especially in a complex radiation environment, which easily leads to signal aliasing and misjudgment.

[0040] Insufficient capture of low-energy β-ray signals: since the β particles emitted by tritium have low energy, the energy signals deposited in the detector are weak, and existing methods have difficulty in accurately capturing and distinguishing this weak signal, thereby affecting the accurate measurement of the β energy spectrum.

[0041] Inaccurate pulse shape analysis: traditional pulse shape discrimination algorithms mainly rely on preset thresholds to distinguish the fast and slow components of signals, and are difficult to fully extract the subtle features contained in the nuclear pulse signals, resulting in inaccurate identification of different particle types (such as “10”, “01”, and “11” scenarios).

[0042] Inadequate real-time data processing capability: the existing nuclear pulse signal processing system cannot achieve efficient and real-time processing of a large amount of data in the data acquisition and processing process, which limits the ability of online monitoring and rapid response, due to the limitation of hardware response time and signal processing algorithm.

[0043] To solve the above problems, the present application provides a tritium water measurement method.

[0044] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0045] In an exemplary embodiment, as shown in Figure 6 A tritium water measurement method is provided, which is executed by a computer device, specifically by a terminal or a server, or by both a terminal and a server. In the present embodiment, the method is applied to a server, which includes the following steps.

[0046] Step 601: obtaining an actual nuclear pulse signal detected by a double-layer scintillator detector.

[0047] Step 602: classifying the actual nuclear pulse signal by using a trained convolutional neural network to obtain an actual energy deposition mode type; the actual energy deposition mode type includes: no energy deposition of particles in the scintillator, energy deposition of particles only in the second layer scintillator, energy deposition of particles only in the first layer scintillator, and energy deposition of particles in both layers of the scintillator.

[0048] Step 603: determining tritium beta rays according to the actual energy deposition mode type.

[0049] By classifying the pulse shapes of different energy depositions of particles in the scintillator by using the trained convolutional neural network, beta rays and gamma rays can be accurately distinguished, thereby improving the accurate measurement of tritium beta rays.

[0050] In an exemplary embodiment, the trained convolutional neural network includes an input layer, a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a global average pooling layer, a fully connected layer, and an output layer connected in sequence.

[0051] In practical applications, the first convolutional layer and the first max-pooling layer are connected through a ReLU activation function; the second convolutional layer and the second max-pooling layer are connected through a ReLU activation function; and the fully connected layer and the output layer are connected through a Softmax activation function. In the present application, since the processing object of the convolutional neural network is a nuclear pulse signal, the data volume is small, and in order to improve the calculation efficiency, the convolutional neural network is lightened by using random inactivation to prevent overfitting.

[0052] In an exemplary embodiment, the actual energy deposition pattern type is determined by classifying the actual nuclear pulse signal using the trained convolutional neural network, specifically comprising:

[0053] The local features of the signal are extracted from the actual nuclear pulse signal using the trained convolutional neural network, including the fast-slow component ratio, the rising edge characteristics of the pulse, the peak amplitude, and the decay characteristics, wherein the fast-slow component ratio, the rising edge characteristics of the pulse, the peak amplitude, and the decay characteristics are used to describe the shape of the nuclear pulse.

[0054] The actual energy deposition pattern type is determined by classifying the actual nuclear pulse signal using the trained convolutional neural network according to the local features. Specifically, the local features include but are not limited to: 1. The rising edge rate of the pulse: used to reflect the time response characteristics of the light output of the scintillator excited by particles, wherein the rising edge of the gamma event is steep and the response is fast; while the rising edge of the beta event is relatively flat; 2. The peak amplitude: the gamma particles usually correspond to a higher pulse peak due to the energy deposition; while the peak of the beta particles is relatively low due to the continuous deposition along the track; 3. The fast-slow component ratio: the information reflecting the decay characteristics of the scintillator is obtained by calculating the integral ratio of the pulse front and the tail, wherein the fast component ratio of the gamma event is higher; 4. The decay trend and the tail broadening: the full width at half maximum (FWHM) and the long tail dragging characteristics of the pulse, used to further distinguish the gamma and beta signals. The above features are input into the trained convolutional neural network model, the model realizes deep feature extraction of the time domain waveform through multi-layer convolution structure, and outputs the class label of the gamma or beta event through the fully connected layer, to obtain the actual energy deposition pattern type.

[0055] In an exemplary embodiment, the training process of the convolutional neural network specifically comprises:

[0056] The simulated nuclear pulse signal is used as the input of the one-dimensional convolutional neural network, and the simulated energy deposition pattern type is used as the output of the one-dimensional convolutional neural network, the hyperparameters of the one-dimensional convolutional neural network are optimized using the cross-entropy loss function and the gradient descent method, to obtain the trained convolutional neural network.

[0057] In an exemplary embodiment, the tritium beta ray is determined according to the actual energy deposition pattern type, specifically comprising:

[0058] When the actual energy deposition pattern type is that the particle only deposits energy in the first layer of scintillator, it is determined that the actual nuclear pulse signal is a tritium beta ray.

[0059] As Figure 1As shown, the present application is based on a dual-layer scintillator to detect low-energy beta rays in tritiated water. When a beta particle (electron) passes through the first layer of scintillator, it interacts with the atoms in the material, exciting the material molecules, thus producing scintillation light. By detecting the photon signals emitted by the first layer of scintillator, the energy and position of the beta particle can be obtained. Gamma rays interact with the second layer of scintillator mainly through photoelectric effect, Compton scattering and pair production processes. When gamma rays interact with the second layer of scintillator, the resulting scintillation light is recorded by a photomultiplier tube (PMT) or a silicon photomultiplier (SiPM), thus obtaining the relevant information of the gamma rays. Since beta particles and gamma rays have different energies and interaction mechanisms, the dual-layer scintillator can distinguish between signals from beta particles and gamma rays by distinguishing their light yields, decay times, etc. in different layers.

[0060] 1. Particle discrimination principle

[0061] If an incident particle triggers the detector to produce an output signal, it must be one of the three cases of energy deposition in the scintillation crystal. Figure 2 As shown, the four cases are used to determine the type of incident particle or its interaction based on the energy deposition pattern in different scintillator layers. The case number is coded in binary, with "1" (or "0") indicating that the particle has (or has not) energy deposition in a certain layer of scintillator. For example, case "10" (case number is abbreviated for brevity) indicates that the particle only deposits energy in the first layer of scintillator BC-444, and case "11" indicates that the particle deposits energy in all two layers of scintillator. The "00" form indicates that the particle does not deposit in both layers of scintillator. Through these four cases, the system can determine the type of incident particle or its interaction based on the energy deposition pattern in different scintillator layers.

[0062] Based on the above case number, the basic principle of particle discrimination is to distinguish different types of particles (such as beta particles, gamma rays, etc.) by their energy deposition patterns in different layers of scintillator. The specific analysis method is as follows:

[0063] Case "00": The particle does not deposit energy in any scintillator. This usually indicates that the particle does not interact with the detector, or the particle passes through the gap of the detector and does not cause a trigger signal. This is usually used for background noise or low-energy particle screening.

[0064] Case "01": The particle only deposits energy in the second layer (such as CsI(Na)) scintillator. This is usually a characteristic of gamma rays, as gamma rays interact strongly with inorganic scintillators, especially through photoelectric effect or Compton scattering to produce scintillation light that is mainly captured by the second layer of scintillator. The first layer of scintillator has a weak response to gamma rays.

[0065] Case "10": The particle only deposits energy in the first layer (e.g. BC-444) scintillator. This case is common for beta particle detection, because beta particles have strong interaction with plastic scintillator (e.g. BC-444) and can produce a large amount of scintillation light in this material, while they have weak interaction with inorganic scintillator, so they rarely deposit in the second layer scintillator.

[0066] Case "11": The particle deposits energy in both layers of scintillators. This kind of event usually comes from high energy beta or gamma particles, the former can penetrate the outer layer and enter the inner layer, the latter can Compton scatter in the outer layer and continue to deposit energy in the inner layer. It can also be caused by two particles from a beta-gamma composite source interacting with the two layers respectively. This case has a certain degree of mixing, which needs to be further distinguished by pulse waveform or energy ratio.

[0067] 2. Pulse shape discrimination algorithm

[0068] Pulse shape discrimination (PSD) algorithm is a signal processing method used to distinguish different types of radiation particles or events, which is widely used in signal discrimination of scintillator detectors. Its principle is based on the different time characteristics of the light signals produced by different particles interacting with the scintillator, as shown in Table 1. By analyzing these characteristics, different radiation events can be distinguished.

[0069] Table 1: Time characteristic parameters of different scintillator materials

[0070]

[0071] The interaction of different types of particles (such as beta particles and gamma rays) with the scintillator will result in different time characteristics of light output, i.e. the process of scintillator emitting light under particle radiation involves the transition of electronic states inside the material. According to the different lifetimes of excited states, the output of scintillation light can be divided into fast and slow components:

[0072] The fast component mainly corresponds to the rapid transition of high-energy excited state particles in the scintillator. These high-energy excited state particles complete energy release and emit photons in a short time through strong coupling with the surrounding medium. The typical time constant of the fast component is usually several nanoseconds to tens of nanoseconds. For example, the fast component time constant of NaI(Tl) is about 230 ns, and the fast component time constant of BC-444 is about 1.4 ns. The light output of the fast component dominates in the initial stage of the signal and forms the rapid rising part of the light signal.

[0073] The slow component corresponds to the decay of the excited state particles with longer lifetime, usually involving the spontaneous transition of metastable particles. The existence of metastable state makes the decay time constant of the slow component longer, generally in tens to hundreds of nanoseconds, even up to microseconds. For example: the time constant of the slow component of BaF2 is about 620 ns, and the time constant of the slow component of CsI(Na) is about 630 ns. The light output of the slow component is weaker than that of the fast component, but it shows a gradually decaying part in the tail of the signal.

[0074] The separation of the fast and slow components makes the scintillation signal exhibit typical double exponential decay characteristics. By analyzing the proportion of the fast and slow components in the signal, the information of the interaction between the particles and the scintillator can be effectively obtained.

[0075] In addition, there are also differences in the time characteristics of different particle excitations. First, the interaction mechanism is that β particles mainly interact with the scintillator through ionization and excitation. Since β particles are charged particles, they directly interact with atomic electrons through Coulomb force when passing through matter, resulting in a high probability of electron excitation and ionization. Second, the fast component is dominant: the excited state produced by β particles is mainly fast transition, and the released energy is quickly converted into fast component photons. This makes the proportion of the fast component in the scintillation light signal induced by β particles significantly higher than that of the slow component. Third, the scintillation signal induced by β particles usually has a steep rising edge and a short decay time. By analyzing the time distribution of the light signal, β particles and γ rays can be distinguished.

[0076] Based on the physical mechanism that in a double-layer scintillator detector, the nuclear pulse signals produced by different types of rays (such as β particles and γ rays) in the scintillator have their own shape characteristics, the shape characteristics of the nuclear pulse signal can be used to effectively distinguish the type of incident particles. This application proposes a method of nuclear pulse signal feature learning and particle type discrimination based on artificial neural network technology, which selects a one-dimensional convolutional neural network (1D-CNN) as the core model for signal discrimination. The nuclear pulse signal data can be regarded as a one-dimensional time series input into the 1D-CNN, and the model structure is as follows: Figure 3The collected pulse signal and its corresponding particle type label are respectively constructed as input matrix X and label vector Y at the data level, and model training is performed through functions such as "trainNetwork(X, Y, layers, options)" in the MATLAB deep learning toolbox. Among them, layers are used to define the hierarchical structure of the network, including the input layer, one-dimensional convolution layer, activation function layer (such as ReLU), pooling layer, fully connected layer and Softmax output layer, etc.; options set the training parameters such as the number of training rounds, learning rate, optimization algorithm, etc. Through this process, the neural network continuously updates the convolution kernel and fully connected layer weight in the training stage, so as to learn the different pulse shape characteristics of beta particles and gamma rays formed in the double-layer scintillator, such as fast and slow component ratio, rise time and pulse shape integral ratio, etc. For the double-layer scintillator detector designed in this application, the neural network can accurately distinguish three typical energy deposition situations ("10", "01" and "11") after fully extracting the details of the nuclear pulse time sequence, and save the mapping relationship in the form of weight parameters, realizing real-time classification and efficient measurement of tritium beta particles and environmental gamma rays under mixed field conditions. Since the number of network layers and the size of weight parameters can be flexibly set according to the detection requirements, combined with appropriate data preprocessing and hyperparameter adjustment, the model can significantly improve the accuracy and sensitivity of particle discrimination while ensuring real-time, providing effective technical support for tritium detection and other low-energy beta / gamma mixed field measurements.

[0077] The specific implementation method is as follows: first, collect or simulate nuclear pulse signal data, which includes nuclear pulse signal shapes and corresponding particle type labels under various situations (such as "10", "01", "11"), and normalize the data. Then, a one-dimensional convolutional neural network (1D-CNN) is selected as the core model, which includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer. The input layer receives nuclear pulse time sequence data of fixed length, the convolutional layer and the pooling layer extract key characteristics such as fast and slow component ratio, rising edge and peak amplitude, the fully connected layer maps the extracted features to the classification output of particle type or situation, and the output layer outputs the probability distribution of each situation using the Softmax function. The convolutional layer includes a first convolutional layer and a second convolutional layer, and the pooling layer includes a first maximum pooling layer, a second maximum pooling layer and a global average pooling layer; in the model training process, the model parameters are optimized through the cross-entropy loss function, the gradient descent is performed combined with the Adam optimizer, the training set, the validation set and the test set are used to optimize the model weight, adjust the hyperparameters and evaluate the model performance, and finally the mapping relationship between the nuclear pulse signal shape and the particle type is used to realize accurate classification.

[0078] The specific implementation method is as follows:

[0079] 1. Data acquisition and preprocessing

[0080] Firstly, nuclear pulse signal data is collected or simulated, which is derived from experimental measurements of double-layer scintillator detectors or GEANT4-based Monte Carlo simulations. The data includes the shapes of nuclear pulse signals under various conditions (such as "10", "01", "11"), and the corresponding particle type (beta particle or gamma ray) of each signal is labeled as a tag to form a complete data set. To improve the generalization ability of the model, the signal data is normalized to compress the amplitude values of all signals into the same range (e.g., [0, 1]). In addition, the signal time series data is cropped or padded to a fixed length to ensure that it can be input into the neural network.

[0081] 2. Model selection and structure design

[0082] A one-dimensional convolutional neural network (1D-CNN) is selected as the core model, and the network structure includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer.

[0083] The input layer receives fixed-length nuclear pulse signal time series data as input, and each signal is discrete waveform data processed by the analog front-end circuit and normalized. These data reflect key information such as the proportion of fast and slow components, the rising edge characteristics, and the peak amplitude. The length of the input data is determined by the signal sampling frequency and the pulse duration.

[0084] The convolutional layer extracts local features of the signal through one-dimensional convolution operations. Multiple convolution kernels (such as 32, 64, 128, etc.) are set, and each convolution kernel can capture features within different time windows, such as the proportion of fast and slow components, the rising edge characteristics of the pulse, the peak amplitude, and the decay characteristics. These features are important for distinguishing between beta particles and gamma rays, as well as determining the conditions "10", "01", and "11".

[0085] The pooling layer uses Max Pooling or Average Pooling operations to reduce the dimensionality of the high-dimensional features extracted by the convolutional layer. This process not only reduces the computational load of the model but also enhances the robustness of the features, avoiding the influence of local noise on the performance of the model.

[0086] The fully connected layer maps the features output by the pooling layer to a low-dimensional space, further compressing the features and performing nonlinear combinations. Multiple fully connected layers can be used to enhance the expressive power of the model, ensuring that the model can learn the complex mapping relationship between nuclear pulse signals and particle types, and condition categories.

[0087] The output layer uses a Softmax activation function to map the output of the fully connected layer to the probability distribution of each condition ("10", "01", "11"). The model determines the classification result based on the maximum probability value.

[0088] The training process of the model focuses on the preparation of the data set. By combining experimental measurements with GEANT4 simulations, a nuclear pulse signal data set is constructed, covering the "10", "01", and "11" energy deposition scenarios. In the experimental phase, the detector system is used to collect actual nuclear pulse signals, and the corresponding signal data is obtained by simulating different types and energies of incident radiation. To ensure the comprehensiveness of the data, multiple collimators and detector combinations are used to collect nuclear pulse signals from different spatial positions, ensuring balanced distribution of data for each scenario. Subsequently, normalization processing is performed on the signals to eliminate the effects of amplitude and baseline offset, laying a unified standard for the input of the neural network.

[0089] After data processing is completed, the data set is divided into training, validation, and test sets to achieve staged optimization and evaluation of the model. The training set is used to optimize the weight and bias parameters of the neural network, the validation set is used to monitor model performance during training and adjust hyperparameters such as learning rate, convolution kernel size, and network depth, and the test set is used to independently evaluate the generalization ability of the model. During model training, the input layer receives standardized one-dimensional nuclear pulse signal data, the convolution layer captures local features in the signal through convolution kernels, such as the proportion of fast and slow components, the steepness of the rising edge, and the peak amplitude, which directly reflect the particle type and energy deposition scenario. The pooling layer enhances the robustness of the features and reduces redundancy through dimension reduction, and the fully connected layer maps the pooled features to a low-dimensional space. The output layer uses the Softmax function to convert the network output features into a probability distribution of the three scenarios.

[0090] During optimization, the cross-entropy loss function is used to measure the error between the model output and the true label, and the Adam optimizer is used for gradient descent to update the weights. During training, hyperparameters are adjusted gradually, and the accuracy and loss value of the model are monitored in real-time through the validation set to avoid overfitting or underfitting. After training is complete, the test set data is input into the model to evaluate the classification accuracy, confusion matrix, and F1 score, etc. to verify the generalization ability of the model on unseen data.

[0091] Finally, the trained model is deployed to the embedded NPU of the detector system for real-time processing of collected nuclear pulse signals. In actual operation, the NPU uses the trained model to classify real-time signals into "10", "01", or "11" scenarios, and transmits the classification results to the host computer through the data interface for display and storage. Meanwhile, the host computer can also statistically analyze the probability distribution of particle energy deposition scenarios based on multiple classification results, providing support for high-precision tritium detection of the detector.

[0092] In actual operation, the well-trained convolutional neural network is deployed on a nuclear pulse signal processor, which can adopt different hardware implementation schemes such as GPU, FPGA and NPU to realize real-time processing of nuclear pulse signals. As shown in Figure 4 , the nuclear pulse signal processor receives the digitized signal from the detector, extracts the pulse shape of the signal, and then analyzes the input data using the pre-trained neural network model, as shown in Figure 5 , the neural network model is used to analyze the input data, and the output of the neural network model is a probability distribution of particle type labels, as shown in Figure 5 (a) is the particle type label of beta particles, Figure 5 (b) is the particle type label of gamma particles, Figure 5 (c) is the particle type label of discarded particles. The NPU predicts the particle type label according to the nuclear pulse signal, which reflects the recognition result of different particles (such as beta particles and gamma rays) inside the detector. For the case "00", it is directly discarded. At the same time, by calculating the probability distribution of the output, the processor can evaluate the confidence of the classification result, thereby further improving the reliability of the detection. This scheme realizes efficient and real-time nuclear pulse signal analysis, and provides reliable technical support for online monitoring of mixed field radiation.

[0093] The present application utilizes the separation ability of the double-layer scintillator for low-energy beta rays and background gamma rays, and combines the analysis of the detector output signal to accurately measure the tritium concentration of the tritium water sample. A one-dimensional convolutional neural network (1D-CNN) is used as the core model, which uses the time series data of the nuclear pulse signal to extract features such as the ratio of fast and slow components, the rising edge and the peak value through the network, and realizes efficient classification of beta particles and gamma rays, which is suitable for application in complex radiation fields.

[0094] Based on the same inventive concept, the embodiments of the present application also provide a tritium water measuring device for implementing the above-mentioned tritium water measuring method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more tritium water measuring device embodiments provided below can refer to the limitations of the tritium water measuring method described above, which will not be repeated here.

[0095] In one exemplary embodiment, as shown in Figure 7 , a tritium water measuring device is provided, comprising:

[0096] The acquisition module 701 is configured to acquire an actual nuclear pulse signal detected by a double-layer scintillator detector;

[0097] The classification module 702 is configured to classify the actual nuclear pulse signal by using the trained convolutional neural network to obtain an actual energy deposition mode type. The actual energy deposition mode type includes: no energy deposition of a particle in a scintillator, energy deposition of a particle only in a second layer of scintillator, energy deposition of a particle only in a first layer of scintillator, and energy deposition of a particle in both layers of scintillator.

[0098] The tritium beta ray determination module 703 is configured to determine a tritium beta ray according to the actual energy deposition mode type.

[0099] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal. An internal structure diagram of the computer device can be as shown in FIG. 8. Figure 8 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store tritiated water measurement data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a tritiated water measurement method.

[0100] Those skilled in the art can understand that Figure 8 The structure shown in FIG. 8 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-mentioned method embodiments.

[0101] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned method embodiments.

[0102] In an exemplary embodiment, a computer program product is provided, which includes a computer program. The computer program is executed by a processor to implement the above-mentioned method embodiments.

[0103] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0104] In the present application, all actions of obtaining signals, information or data are carried out in compliance with the data protection regulations and policies of the country where the device is located, and with the authorization given by the owner of the corresponding device.

[0105] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0106] The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on blockchain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0107] Any technical features in the above embodiments can be combined, and for the sake of brevity, not all possible combinations are described above, however, it should be understood that the application encompasses all possible combinations of the technical features described above.

[0108] The principles and implementation manners of the present application are described herein by using specific examples, and the above embodiments are only used to help understand the method of the present application and its core idea; meanwhile, according to the idea of the present application, the specific implementation manners and application scopes will be changed by those skilled in the art. In conclusion, the content of the present specification should not be understood as a limitation of the present application.

Claims

1. A method of measuring tritiated water, characterized by, The tritium water measurement method includes: Acquire the actual nuclear pulse signal detected by the double-layer scintillator detector; The actual energy deposition pattern type is obtained by classifying the actual nuclear pulse signal using a trained convolutional neural network. The actual energy deposition pattern type includes: no energy deposition in the scintillator, energy deposition only in the second scintillator, energy deposition only in the first scintillator, and energy deposition in both scintillators. The tritium β-rays are determined based on the actual energy deposition mode type.

2. The method of claim 1, wherein, The trained convolutional neural network includes an input layer, a first convolutional layer, a first max pooling layer, a second convolutional layer, a second max pooling layer, a global average pooling layer, a fully connected layer, and an output layer connected in sequence.

3. The method of claim 2, wherein, The first convolutional layer is connected to the first max pooling layer via a ReLU activation function; the second convolutional layer is connected to the second max pooling layer via a ReLU activation function; and the fully connected layer is connected to the output layer via a Softmax activation function.

4. The method of claim 1, wherein, Based on the actual nuclear pulse signals, a trained convolutional neural network is used to classify the actual energy deposition mode types, which specifically include: The actual nuclear pulse signal is used to extract features using a trained convolutional neural network to obtain local features of the signal; the local features include the ratio of fast and slow components, the rise time characteristics of the pulse, the peak amplitude, and the attenuation characteristics. The actual nuclear pulse signal is classified based on the local features using a trained convolutional neural network to obtain the actual energy deposition mode type.

5. The method of claim 1, wherein, The training process of a convolutional neural network specifically includes: Using simulated nuclear pulse signals as input to a one-dimensional convolutional neural network (CNN) and simulated energy deposition mode types as output, the hyperparameters of the CNN are optimized using the cross-entropy loss function and gradient descent method to obtain a trained CNN.

6. The method of claim 1, wherein, The tritium β-rays are determined based on the actual energy deposition mode type, specifically including: When the actual energy deposition mode is such that the particle deposits energy only in the first scintillator, the actual nuclear pulse signal is determined to be tritium beta rays.

7. A tritiated water measuring device characterized by comprising: The tritium water measuring device includes: The acquisition module is used to acquire the actual nuclear pulse signal detected by the dual-layer scintillator detector; The classification module is used to classify the actual nuclear pulse signal using a trained convolutional neural network to obtain the actual energy deposition pattern type; the actual energy deposition pattern type includes: the particle does not deposit energy in the scintillator, the particle only deposits energy in the second scintillator, the particle only deposits energy in the first scintillator, and the particle deposits energy in both scintillators. The tritium beta ray determination module is used to determine tritium beta rays based on the actual energy deposition mode type.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the tritium water measurement method according to any one of claims 1-6.

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

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the tritium water measurement method according to any one of claims 1-6.

Images