Nuclide activity measurement method based on neural network

Abstract

The invention provides a nuclide activity measurement method based on a neural network, and relates to the technical field of nuclide activity measurement, and the method comprises the steps: constructing a coincidence measurement system, obtaining a plurality of groups of counting parameters of nuclides under different beta detection efficiencies through the system, and carrying out the correction processing of the counting parameters, and inputting the corrected counting parameter into a one-dimensional convolutional neural network model, and obtaining the nuclide real activity based on a condition that model output extrapolation is performed until beta detection efficiency is 100%. According to the method, the neural network is adopted to replace a traditional extrapolation fitting method, the nonlinear relation between the nuclide counting rate and the activity is automatically learned, the accuracy and robustness of activity measurement are remarkably improved, the method does not depend on an explicit function form, abnormal data can be effectively handled, the accuracy of beta, gamma and coincidence counting rates is ensured, and the method is suitable for popularization and application. The interference of environmental background and cosmic rays is effectively inhibited, the signal-to-noise ratio is increased, the measurement efficiency is improved, the operation is simpler and more convenient, and the method is suitable for various nuclear technology application scenes.

Description

Technical Field

[0001] This invention relates to the field of radionuclide activity measurement technology, and in particular to a radionuclide activity measurement method based on neural networks. Background Technology

[0002] Nuclide activity measurement is a fundamental and crucial research area in nuclear science and technology, possessing significant theoretical importance and broad application value. Nuclide activity reflects the number of decays per unit time of a radioactive substance, directly impacting the accurate calibration of radioactive source intensity, radiation dose assessment, nuclear safety monitoring, and environmental radiation control—among other important applications. Among existing activity measurement methods, liquid scintillation counting combined with TDCR (Triple-to-Double Coincidence Ratio) technology has become a widely used and important approach. Traditionally, when measuring the activity of nuclides such as P32 using the TDCR method, a chemiluminescence inhibitor (such as acetic acid CH3COOH) is added to artificially reduce the detection efficiency of the β channel, thereby extrapolating the activity value corresponding to a detection efficiency of 100%. This method uses a fitting formula to fit the data; the presence of outlier data points leads to poor fitting results, ultimately resulting in inaccurate activity calculations. In addition, another method combining TDCR technology with external standard source methods for measuring nuclides such as C14, while improving the accuracy of activity measurement in specific applications, can only measure pure β sources and cannot measure β-γ sources. Therefore, there is an urgent need for a new method that is simple in process, low in cost, adaptable to multiple nuclide types, and especially capable of rapid and high-precision measurement of β-γ nuclides.

[0003] 4πβ(LS)-γ coincidence measurement is a technique that combines a β-channel detector (a combination of a liquid scintillator and a photomultiplier tube) and a γ-channel detector (NaI crystal detector). This method identifies valid radioactive decay events by simultaneously detecting β particles and γ photons within a set coincidence time window using two independent detection channels. Coincidence measurement has inherent advantages in effectively suppressing external background radiation interference and improving the signal-to-noise ratio. With the development of computer technology and the rapid advancement of artificial intelligence, using AI to perform offline analysis of the measured data to determine nuclide activity will be a future direction for nuclear technology development.

[0004] Therefore, it is essential to design a method for measuring nuclide activity based on neural networks. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a method for measuring nuclide activity based on neural networks.

[0006] To achieve the above objectives, the present invention provides the following solution: This invention provides a method for measuring nuclide activity based on a neural network, comprising: Step 1: Construct a measurement system that meets the requirements; Step 2: Obtain multiple sets of counting parameters for nuclides at different β detection efficiencies using the system; Step 3: Correct the counting parameters; Step 4: Input the corrected counting parameters into the one-dimensional convolutional neural network model; Step 5: Extrapolate the model output to the condition that the β detection efficiency is 100% to obtain the true activity of the nuclide.

[0007] Preferably, in step 1, constructing the conformity measurement system specifically involves: The coincidence measurement system includes a main detector system, an anti-coincidence detection system, a digital data acquisition and processing system, and a software analysis platform.

[0008] Preferably, the main detector system includes a liquid scintillator, a photomultiplier tube, and a NaI crystal detector. The NaI crystal detector is used for detecting gamma-channel signals, and the liquid scintillator and the photomultiplier tube together constitute a beta-channel detector for recording light signals generated by beta particle excitation.

[0009] Preferably, the anti-coincidence detection system includes: The system includes a lead chamber and anti-coincidence detectors. The main detector system is installed inside the lead chamber, and multiple anti-coincidence detectors are installed around the lead chamber. The anti-coincidence detectors are plastic scintillators. The anti-coincidence detection system is used to identify and eliminate external background interference signals.

[0010] Preferably, the digital data acquisition and processing system is a high-performance multichannel analyzer or a programmable data acquisition system, used to receive the detector's raw signals and complete β-γ coincidence determination, anti-coincidence event identification, energy discrimination, and time stamping.

[0011] Preferably, the counting parameters include coincidence count rate, beta count rate, gamma count rate, anti-coincidence count rate, and nuclide activity under these conditions.

[0012] Preferably, in step 3, the counting parameters are corrected, specifically as follows: The counting parameters are subjected to dead time correction, random coincidence correction and background suppression.

[0013] Preferably, in step 4, the corrected counting parameters are input into the one-dimensional convolutional neural network model, specifically as follows: The β count rate, γ count rate, and coincidence count rate in the corrected counting parameters are used as input features. These are then fed into a 3D convolutional neural network model, where features are extracted through a multi-layer convolutional structure.

[0014] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention provides a method for measuring nuclide activity based on a neural network. The method includes constructing a coincidence measurement system, obtaining multiple sets of counting parameters of the nuclide under different β detection efficiencies through the system, correcting the counting parameters, inputting the corrected counting parameters into a one-dimensional convolutional neural network model, and extrapolating the model output to the condition that the β detection efficiency is 100% to obtain the true activity of the nuclide. This invention employs a neural network to replace the traditional extrapolation fitting method, automatically learning the nonlinear relationship between nuclide count rate and activity, significantly improving the accuracy and robustness of activity measurement. Compared to traditional methods, this method does not rely on explicit functional forms and can effectively handle anomalous data. The accompanying TDCR triple coincidence structure and dead time and accidental coincidence correction mechanism ensure the accuracy of β, γ, and coincidence count rates. Through lead shielding and an anti-coincidence detector system, background noise and cosmic ray interference are effectively suppressed, improving the signal-to-noise ratio. The method is compatible with β and β-γ composite sources, applicable to a wider range of nuclide types. Combined with the GammaAnt software platform, real-time data acquisition, analysis, and activity prediction can be achieved, featuring high automation and intelligence, higher measurement efficiency, and simpler operation, making it suitable for various nuclear technology application scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 To conform to the schematic diagram of the measurement system structure; Figure 3 To conform to the schematic diagram of the working principle of the measurement system; Figure 4 This is a schematic diagram of a one-dimensional convolutional neural network model.

[0017] Figure reference numerals: 1. Photomultiplier tube; 2. Plastic scintillator; 3. Liquid scintillator; 4. Lead chamber; 5. NaI crystal detector. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The purpose of this invention is to provide a neural network-based method for measuring nuclide activity. This method replaces traditional extrapolation and fitting methods with a neural network, automatically learning the nonlinear relationship between nuclide count rate and activity, significantly improving the accuracy and robustness of activity measurement. Compared to traditional methods, this method does not rely on explicit functional forms and can effectively handle anomalous data. The accompanying TDCR triple coincidence structure and dead-time and accidental coincidence correction mechanisms ensure the accuracy of β, γ, and coincidence count rates. Lead shielding and an anti-coincidence detector system effectively suppress background noise and cosmic ray interference, improving the signal-to-noise ratio. The method is compatible with both β and β-γ composite sources, making it applicable to a wider range of nuclide types. Combined with the GammaAnt software platform, it enables real-time data acquisition, analysis, and activity prediction, exhibiting high automation and intelligence, higher measurement efficiency, and simpler operation, making it suitable for various nuclear technology applications.

[0020] 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.

[0021] like Figure 1 As shown, this invention provides a method for measuring nuclide activity based on a neural network, comprising: Step 1: Construct a measurement system that meets the requirements; Step 2: Obtain multiple sets of counting parameters for nuclides at different β detection efficiencies using the system; Step 3: Correct the counting parameters; Step 4: Input the corrected counting parameters into the one-dimensional convolutional neural network model; Step 5: Extrapolate the model output to the condition that the β detection efficiency is 100% to obtain the true activity of the nuclide.

[0022] In step 1, the conformity measurement system is constructed, specifically as follows: The coincidence measurement system includes a main detector system, an anti-coincidence detection system, a digital data acquisition and processing system, and a software analysis platform. Each system has a clearly defined function and works closely together. By optimizing detection efficiency, enhancing anti-interference capabilities, improving time resolution, and increasing automation levels, the system aims to build a modern radiometric measurement platform with high sensitivity, high reliability, and intelligent features, suitable for multiple fields such as nuclear science research, environmental monitoring, and nuclear safety regulation.

[0023] The main detector system includes a liquid scintillator 3, a photomultiplier tube 1, and a NaI crystal detector 5. The NaI crystal detector 5 is used for detecting gamma-ray signals. The liquid scintillator 3 and the photomultiplier tube 1 together form a beta-ray detector, which is used to record the light signals generated by beta particle excitation. The liquid scintillator 3 (LS) is configured as the detection medium for beta particles and coupled with the high-performance photomultiplier tube 1 (PMT) to collect the scintillation signals generated by beta events. At the same time, two NaI crystal detectors 5 are configured for energy deposition detection of gamma rays. This system constitutes a beta channel and a gamma channel. By optimizing the spatial structure, it achieves 4π full solid angle coverage of beta rays from the radiation source, improving detection efficiency and statistical accuracy.

[0024] The anti-coincidence detection system includes a lead chamber 4 and anti-coincidence detectors. The main detector system is located inside the lead chamber 4, and multiple anti-coincidence detectors are arranged around the lead chamber 4. The anti-coincidence detectors are plastic scintillators 2. The anti-coincidence detection system is used to identify and eliminate external background interference signals. A high-density shield made of lead material is constructed as a basic background suppression structure to effectively reduce the background gamma rays and cosmic rays in the environment. When external particles simultaneously generate signals in the anti-coincidence detectors and the main detector, the event is marked as interference by digital logic and eliminated, thus achieving active background suppression.

[0025] The digital data acquisition and processing system is a high-performance multichannel analyzer or a programmable data acquisition system, used to receive raw signals from the detectors and complete β-γ coincidence determination, anti-coincidence event identification, energy discrimination, and time stamping. The system incorporates a high-performance multichannel analyzer or programmable data acquisition system with multi-channel input capability, a time-to-digital converter (TDC) module, a high-speed analog-to-digital converter (ADC) module, and a logic triggering judgment module. This system receives raw signals from all detectors, completes β-γ coincidence determination, anti-coincidence event identification, energy discrimination, and time stamping, and allows for custom settings of key parameters such as coincidence time windows and energy thresholds, achieving digital, intelligent, and centralized signal processing.

[0026] The software analysis platform is the self-developed software system GammaAnt, which is used to display coincidence event counts in real time, draw coincidence spectra, generate energy-time two-dimensional correlation diagrams, and support the storage, playback, export and reprocessing of measurement data. The system should have an activity calculation module, energy spectrum fitting function and data visualization tools to meet the needs of multi-angle analysis and comparison of measurement results.

[0027] The counting parameters include coincidence count rate, beta count rate, gamma count rate, anticoincidence count rate, and nuclide activity under these conditions.

[0028] This invention provides an embodiment, the structural schematic of which is shown below. Figure 2 As shown, the system comprises a liquid scintillator 3 (LS), a photomultiplier tube 1 (PMT), a lead chamber 4, a NaI crystal detector 5, an anti-coincidence detector, and a multichannel analysis system. The liquid scintillator 3 and the NaI crystal detector 5 constitute the main detector, used to detect β-particle and γ-ray signals respectively. The plastic scintillator 2 acts as an anti-coincidence detector, used to identify and eliminate external background interference signals such as cosmic rays. The main detector is housed inside the lead chamber 4, which effectively shields the ambient background radiation, further reducing the background level and improving measurement sensitivity. In the main detector system, the NaI crystal detector 5 is specifically used for detecting γ-ray signals. The liquid scintillator 3 and the photomultiplier tube 1 together form a β-ray detector, used to record the light signals generated by β-particle excitation. Through the coordinated operation of the main detector and the anti-coincidence detector, combined with the multichannel analysis system for energy analysis and coincidence discrimination of the detected signals, efficient and accurate measurement of nuclide activity can be achieved, effectively suppressing background interference and improving the overall measurement accuracy and reliability. Its working principle is as follows Figure 3 As shown, when the β-γ source decays, the β rays are absorbed by the liquid scintillator 3 (LS) and converted into Cherenkov photons. These photons are then received by the photomultiplier tube 1 (PMT) and generate a signal, which is input to the TDCR submodule in the multichannel analyzer via the PMT. When the input signals from three PMTs arrive within the coincidence time window, these signals are considered a triple coincidence event or a double coincidence event. If only two PMT signals arrive at the TDCR submodule within the resolution time, it is considered a double coincidence event. If only one PMT signal arrives at the TDCR submodule, this signal is considered invalid. It is important to note that both double and triple coincidence events are considered valid events, indicating that the detector has successfully recorded the coincidence event. When the gamma rays are collected by the NaI crystal detector 5 and processed by the delay module, a γ-channel signal is output. If the gamma-channel signal and the beta-channel signal overlap within the resolution time, it is considered a valid coincidence event. The corrected coincidence event count rate, beta-channel count rate, gamma-channel count rate, and anti-coincidence count rate are then transmitted to the computer. The computer further processes this data using dedicated GammaAnt software to calculate and obtain information such as the coincidence count rate, beta-channel count rate, gamma-channel count rate, anti-coincidence count rate, and nuclide activity under these conditions.

[0029] In step 3, the counting parameters are corrected, specifically as follows: The counting parameters are subjected to dead time correction, random coincidence correction and background suppression.

[0030] In step 4, the corrected counting parameters are input into the one-dimensional convolutional neural network model, specifically as follows: The β count rate, γ count rate, and coincidence count rate in the corrected counting parameters are used as input features. These are then fed into a 3D convolutional neural network model, where features are extracted through a multi-layer convolutional structure.

[0031] The present invention will now describe in detail the method for calculating nuclide activity based on neural network technology: By artificially adjusting the detection efficiency of the β channel, the detection system obtains multiple sets of counting parameters (such as β count rate, γ count rate, coincidence count rate, etc.) under different efficiency conditions, thereby establishing a mapping relationship between the measured parameters and the nuclide activity. Unlike traditional methods that rely on explicit functional forms for linear or nonlinear fitting, this invention utilizes neural networks, especially one-dimensional convolutional neural networks (1D-CNN), using measurement data such as the count rate of each channel as input features to train the model to automatically learn the implicit relationship between the measured parameters and activity. The neural network essentially achieves high-order nonlinear fitting of the extrapolation curve. This process is equivalent to using a neural network to replace the traditional fitting formula for extrapolation, and finally extrapolating the model output to the ideal condition of 100% β detection efficiency, thus obtaining the true activity of the nuclide. This method retains the physical logic of extrapolation methods—"indirect measurement through efficiency changes"—while overcoming the limitations of traditional function fitting in handling abnormal data and nonlinear features, achieving more stable and accurate activity extrapolation. The model structure is as follows: Figure 4 As shown, by training a neural network model, the model learns the activity features under different β-probe efficiencies. The model structure is as follows. Figure 4 As shown, the neural network is trained on multiple sets of input measurement parameters to automatically learn the variation law of nuclide activity under different β detection efficiencies, thereby establishing a nonlinear mapping relationship between measurement parameters and activity. This model can extract activity feature information contained in the count rate of each channel and gradually approximate the fitting curve in the traditional extrapolation method through a deep network structure. After training, the neural network has the ability to accurately predict activity based on actual measurement parameters, achieving effective extrapolation of nuclide activity when the β detection efficiency is 100%.

[0032] In the model construction process of this method, the most critical step lies in high-quality, high-precision data collection and processing. Only by ensuring the accuracy and reliability of the input data can a solid foundation be laid for the training of the neural network model, thereby achieving accurate prediction of nuclide activity. To obtain accurate β-channel counts, this method introduces TDCR (Triple-to-Double Coincidence). The Ratio technique, by setting up a triple coincidence detection system composed of three high-performance photomultiplier tubes (PMTs), not only significantly improves the detection efficiency of β events, but also effectively suppresses false counts caused by non-real events such as PMT dim light and electrical noise, thus optimizing the quality of β count data from the source. Subsequently, in response to systematic loss events caused by detector response time, this invention performs dead-time correction on the β channel count rate, further improving its data accuracy. The γ channel uses a sodium iodide (NaI) scintillation detector with excellent energy resolution and high γ absorption capability to ensure high-fidelity acquisition of γ-ray energy spectrum information. At the same time, dead-time correction is also implemented on the γ count data to eliminate statistical bias caused by high count rate or signal stacking. In the processing of coincidence channels, the method introduces a dual mechanism of accidental coincidence correction and dead-time compensation, which effectively eliminates false coincidences caused by the superposition of high-frequency background events and improves the accuracy and representativeness of the coincidence channel count rate. The entire measurement system operates in a low-background environment, with the experimental setup housed in a heavy-duty lead-shielded chamber, significantly reducing interference from ambient gamma rays and cosmic rays. Furthermore, an anti-coincidence detector system composed of a high-performance plastic scintillator 2 is installed outside the lead chamber 4 to further identify and eliminate irrelevant particle interference events from the outside, greatly improving data clarity and the signal-to-noise ratio of the experiment. Finally, the system-corrected β-channel count rate, γ-channel count rate, and coincidence channel count rate are used for preliminary activity calculations and input as feature data into a constructed one-dimensional convolutional neural network model. Through deep learning of these physical measurement data, the neural network automatically extracts the intrinsic relationship between count rate changes and nuclide activity. In this process, it can also identify and adjust the influence of anomalous data, thereby constructing a fitting curve with extrapolation capabilities. This enables accurate extrapolation and prediction of nuclide activity under ideal detection conditions (β-detection efficiency of 100%), significantly improving the automation level of the measurement system and the overall accuracy and robustness of activity measurement.

[0033] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0034] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A neural network based measurement method of a nuclide activity, characterized by, The application relates to a method for measuring the activity of a radionuclide, comprising the following steps: Step 1: constructing a coincidence measurement system; Step 2: obtaining multiple sets of counting parameters of the radionuclide under different beta detection efficiencies through the system; Step 3: correcting the counting parameters; Step 4: inputting the corrected counting parameters into a one-dimensional convolutional neural network model; Step 5: based on the model, extrapolating to the condition that the beta detection efficiency is 100%, and obtaining the real activity of the radionuclide.

2. The method of claim 1, wherein, In step 1, the coincidence measurement system is constructed, and specifically comprises a main detector system, an anticoincidence detection system, a digital data acquisition and processing system and a software analysis platform. The main detector system comprises a liquid scintillator, a photomultiplier tube and a NaI crystal detector, the NaI crystal detector is used for detecting a gamma channel signal, the liquid scintillator and the photomultiplier tube jointly constitute a beta channel detector, and the beta channel detector is used for recording a light signal generated by excitation of beta particles.

3. The method of claim 2, wherein, The anticoincidence detection system comprises a lead chamber and anticoincidence detectors, the main detector system is arranged in the lead chamber, a plurality of anticoincidence detectors are arranged around the lead chamber, the anticoincidence detectors are plastic scintillators, and the anticoincidence detection system is used for identifying and removing external background interference signals.

4. The method of claim 3, wherein, The digital data acquisition and processing system is a high-performance multichannel analyzer or a programmable data acquisition system, which is used for receiving original signals of the detectors, completing beta-gamma coincidence determination, anticoincidence event identification, energy discrimination and time marking.

5. The method of claim 4, wherein, The counting parameters comprise coincidence channel counting rate, beta channel counting rate, gamma channel counting rate, anticoincidence counting rate and radionuclide activity under the conditions.

6. The method of claim 5, wherein, In step 3, the counting parameters are corrected, and specifically dead time correction, accidental coincidence correction and background suppression treatment are performed on the counting parameters.

7. The method of claim 6, wherein, In step 4, the corrected counting parameters are input into a one-dimensional convolutional neural network model, and specifically the beta counting rate, the gamma counting rate and the coincidence counting rate in the corrected counting parameters are input into the one-dimensional convolutional neural network model, and features are extracted through a multilayer convolutional structure. ​ 8. The method of claim 7, wherein, ​ ​

Images