Simulation database generation method and system based on CZT carrier transport characteristics and energy spectrum characteristics

Abstract

The invention provides a simulation database generation method and system based on CZT carrier transport characteristics and energy spectrum characteristics, and relates to the technical field of semiconductors, and the method comprises the steps: simulating energy deposition position distribution data of a radioactive source gamma ray in a CZT detector through GEANT4; constructing a detector electric field model according to the energy deposition position distribution data, and obtaining potential distribution on a carrier drift path; according to the energy deposition value and the energy required for generating a pair of carriers, combining a fano factor to generate an initial carrier number obeying Gaussian distribution; setting a carrier mobility lifetime product parameter, and iteratively calculating capture attenuation in a carrier drift process according to a preset step length; based on the Sockley-Ramo theorem, calculating the quantity of electric charges induced on the electrode during electron and hole drift step by step; repeatedly calculating a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; carrying out normalization processing after the simulation energy spectrum and the actually measured background energy spectrum are superposed, and optimizing mobility life product parameters through a least square method; and regenerating a complete database containing single nuclides and mixed nuclides by adopting the optimized parameters for machine learning training.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor technology, and in particular to a simulation database generation method and system based on CZT carrier transport characteristics and energy spectrum characteristics. BACKGROUND

[0002] The CZT spectrometer has high energy resolution and is suitable for nuclide identification. The current mainstream mixed nuclide and low-dose nuclide identification method is a neural network algorithm based on machine learning, such as FCN, CNN neural network, etc. Machine learning needs to provide a large number of test sets and training sets. Among the most concerned nuclides, except for a few nuclides such as Cs-137 and Co-60, which can use exempt sources to collect energy spectrum data, other nuclides do not have experimental conditions, so simulation research using GEANT4 is a feasible way to generate training energy spectrum.

[0003] However, the full energy peak of the conventional scintillator spectrum is a standard Gaussian broadening, which causes the simulated energy spectrum to lack the low-energy tailing feature existing in the actual measured data mismatch problem; during the drift process of the CZT semiconductor carrier, the electron and the hole are captured, and in particular, the capture probability of the hole is usually 2 orders of magnitude larger than that of the electron. This characteristic is reflected in the energy spectrum, that is, there is a serious low-energy tailing in the full energy peak, as shown in FIG. 1. Therefore, in order to improve the accuracy of the machine learning training energy spectrum, it is necessary to study the transport characteristics of the CZT semiconductor carrier, and to use the characteristics to reasonably calculate the energy spectrum response of the simulated energy spectrum by GEANT4, so as to obtain a simulated energy spectrum close to the actual energy spectrum. Figure 2 SUMMARY

[0004] The present application aims to provide an efficient speech recognition method, device, equipment and storage medium, to solve the problem of the full energy peak of the conventional scintillator spectrum being a standard Gaussian broadening, which causes the simulated energy spectrum to lack the low-energy tailing feature existing in the actual measured data mismatch.

[0005] ​To achieve the above object, the application provides the following technical scheme: a machine learning simulation database generation method based on CZT carrier transport characteristics and energy spectrum characteristics, steps comprising: simulating the energy deposition position distribution data of the gamma rays of a radioactive source in a CZT detector through GEANT4; constructing a detector electric field model according to the energy deposition position distribution data to obtain the potential distribution on the carrier drift path; generating the initial carrier number subject to Gaussian distribution according to the energy deposition value and the energy required to generate a pair of carriers, combined with the Fano factor; setting the carrier mobility-lifetime product parameter, and iteratively calculating the capture attenuation in the carrier drift process according to the preset step; step-by-step calculating the charge amount induced on the electrode when the electron and hole drift based on the Shockley-Ramo theorem; repeatedly calculating a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; normalizing the superposition of the simulation energy spectrum and the measured background energy spectrum, and optimizing the mobility-lifetime product parameter through the least square method; and regenerating the complete database containing single nuclides and mixed nuclides using the optimized parameter for machine learning training.

[0006] Optionally, the step of simulating the energy deposition position distribution of the gamma rays of a radioactive source in a CZT detector through GEANT4 specifically comprises: constructing a 10*10*5mm³ CZT detector three-dimensional geometric model; setting the radioactive source to be located 10mm above the detector, and the radioactive source types to include Cs-137 and Co-60; simulating the three-dimensional coordinate distribution of the energy deposition points generated by the interaction of gamma rays and the detector; recording the energy value and spatial position information of each deposition point to form the energy deposition position distribution.

[0007] Optionally, the step of constructing a detector electric field model to obtain the potential distribution on the carrier drift path specifically comprises: establishing a 10*10*5mm³ CZT detector three-dimensional geometric model in COMSOL; setting a 1mm-diameter point electrode on the bottom surface of the detector, applying a 900V high voltage, and grounding the remaining five surfaces; and calculating the three-dimensional electric field distribution inside the detector by using extremely refined grid division.

[0008] Optionally, the step of generating the initial carrier number subject to Gaussian distribution according to the energy deposition value and the energy required to generate a pair of carriers, combined with the Fano factor specifically comprises: determining the energy of a pair of electrons-holes generated by a gamma ray in CZT; calculating the theoretical carrier pair number according to the deposition energy value; introducing the Fano factor to establish a carrier number Gaussian distribution model; and randomly generating the initial carrier number subject to statistical fluctuation characteristics for each deposition event.

[0009] Optionally, the carrier mobility-lifetime product parameter comprises an electron mobility-lifetime product and a hole mobility-lifetime product, wherein the typical value of the electron mobility-lifetime product is set to range from 1*10 - cm² / V to 1*10- cm-2 / V, the typical value of the product of hole mobility lifetime is set in the range of 1x10 -5 cm-2 / V to 1x10 -4 cm-2 / V.

[0010] Optionally, the step of calculating the amount of induced charge on the electrodes based on the Shockley-Ramo theorem during the drift of electrons and holes specifically includes: within each drift step, calculating the amount of weight potential change caused by the movement of carriers; respectively accumulating the instantaneous charge induced by electrons and holes on the anode and cathode; terminating the calculation when the carriers reach the boundary of the detector; and accumulating the amount of induced charge of all drift steps as the final signal output of the case.

[0011] Optionally, the step of superimposing the simulated energy spectrum on the measured background energy spectrum and performing normalization processing, and optimizing the mobility lifetime product parameters by the least square method specifically includes: collecting the measured Cs-137 and Co-60 energy spectrum as the reference data; selecting different parameter values between 0.1 times and 10 times of the typical values of electron mobility lifetime product and hole mobility lifetime product; performing energy calibration and background superposition on the simulated energy spectrum generated by each group of parameters for normalization processing; determining the optimal parameter combination by minimizing the least square calculation value; and taking the optimal parameter combination as the parameter for subsequent simulation calculation.

[0012] Optionally, the step of regenerating a complete database containing single nuclides and mixed nuclides using the optimized parameters for machine learning training specifically includes: selecting radioactive sources of different nuclides and different positions to generate a single nuclide energy spectrum library; generating multiple Poisson distribution random samples for each nuclide; randomly selecting several nuclide energy spectra and superimposing them in different proportions; generating multiple training samples for each mixed combination; and establishing a complete database containing single nuclides and mixed nuclides.

[0013] In another aspect, the present application also provides a machine learning simulation database generation system based on CZT carrier transport characteristics and energy spectrum characteristics, comprising: an energy deposition distribution simulation module for simulating the energy deposition position distribution data of gamma rays of a radioactive source in a CZT detector by GEANT4; an electric field modeling module for constructing a detector electric field model according to the energy deposition position distribution data using COMSOL Multiphysics to obtain the potential distribution on the carrier drift path; a carrier generation module for generating the initial number of carriers subject to a Gaussian distribution according to the energy deposition value and the energy required to generate a pair of carriers in combination with the Fano factor; a capture attenuation calculation module for setting the carrier mobility lifetime product parameter and iteratively calculating the capture attenuation in the carrier drift process according to a preset step size; a charge quantity calculation module for calculating the charge quantity induced on the electrode when the electrons and holes drift step by step based on the Shockley-Ramo theorem; a database generation module for repeatedly calculating a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; a parameter optimization module for normalizing the superimposed simulation energy spectrum and the measured background energy spectrum, and optimizing the mobility lifetime product parameter by the least square method; and a complete database generation module for regenerating a complete database containing single nuclides and mixed nuclides using the optimized parameter for machine learning training.

[0014] In another aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the machine learning simulation database generation method based on CZT carrier transport characteristics and energy spectrum characteristics when executing the computer program.

[0015] Compared with the prior art, the present application has the following beneficial effects: The present application generates a training database using simulation data, overcomes the difficulty that many radioactive source data cannot be obtained by actual measurement, and enriches the types of radioactive source data in the nuclide library.

[0016] The present application combines electric field simulation data with simulated radiation data, utilizes carrier transport characteristics, and simulates the full energy peak low energy tailing characteristics of a cadmium zinc telluride detector, so that the energy spectrum in the simulation energy spectrum database is closer to the actual measured energy spectrum, and the effectiveness of the database is enhanced.

[0017] The present application dynamically optimizes the mobility lifetime product parameter by the least square method, so that the simulation energy spectrum and the measured energy spectrum are highly consistent in peak shape and tailing characteristics, and the simulation data can be used to generate a large-scale training set of scarce nuclides, which completely solves the core pain points of difficult acquisition of radioactive samples and limited coverage spectrum in traditional machine learning, and provides high-fidelity database support for nuclide identification in complex scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flow chart of the method steps of the present application.

[0019] Figure 2 A measured spectrum of the radioactive source of the present application.

[0020] Figure 3 A schematic diagram of the position distribution of the gamma-ray deposited energy of the present application.

[0021] Figure 4 A schematic diagram of the COMSOL detector model and potential simulation of the present application.

[0022] Figure 5 A schematic diagram of the electric field distribution of the detector of the present application.

[0023] Figure 6 A schematic diagram of the central axis potential of the detector of the present application.

[0024] Figure 7 A schematic diagram of the system structure of the present application.

[0025] In the figure: 10-energy deposition distribution simulation module, 20-electric field modeling module, 30-carrier generation module, 40-capture attenuation calculation module, 50-charge amount calculation module, 60-database generation module, 70-parameter optimization module, 80-complete database generation module. DETAILED DESCRIPTION

[0026] The scheme of the present application will be clearly and completely explained in combination with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0028] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" (or comprise), "comprises" (or comprises) and "comprising" (or comprising) when used in this specification, can mean "consist of." It is further understood that the terms "include," "includes" and / or "comprise," "comprises," and / or "comprising" when used in this specification, can mean "consist of." It is to be understood that where the term "connected" or "coupled" is used herein, it is understood that, while not expressly recited in most contexts, something "connected" or "coupled" to something else (e.g., component A coupled to component B) can be directly connected or coupled, or it can have one or more intervening components or elements interposed therebetween (e.g., component A coupled to component B through component C). It will be further understood that the terms "connected" or "coupled" as used herein can include wireless connection or wireless coupling. The term "and / or" as used herein encompasses all of the associated listed items individually and in all possible combinations.

[0029] Those skilled in the art will appreciate that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0030] It should be understood that the sequence numbers and sizes of the steps in the embodiments are not intended to mean the order of execution, and the execution order of the processes is determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0031] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0032] Reference will now be made to Figures 1-6The application discloses a machine learning simulation database generation method based on CZT carrier transport characteristics and energy spectrum characteristics.

[0033] Specifically, the interaction of gamma rays of radioactive sources such as Cs-137 and Co-60 in a CZT detector is simulated by GEANT4, and energy deposition position distribution data as shown in the accompanying drawings are recorded. Figure 3 GEANT4 is a Monte Carlo particle transport simulation toolkit developed by the European Center for Nuclear Research, and is a simulation platform in the field of nuclear and particle physics.

[0034] Specifically, COMSOL Multiphysics is a large-scale advanced numerical simulation software. It is widely used in scientific research and engineering calculation in various fields, and simulates various physical processes in the field of science and engineering. COMSOL Multiphysics is based on the finite element method, and solves partial differential equations or partial differential equation systems to realize simulation of real physical phenomena, and solves real-world physical phenomena by mathematical methods.

[0035] The application deeply integrates GEANT4 particle transport simulation, COMSOL electric field modeling and carrier transport theory, accurately reproduces the unique low-energy tailing effect of the CZT detector by establishing a complete physical chain including the carrier capture decay model and the Shockley-Ramo theorem; the least square method is used to dynamically optimize the mobility-lifetime product parameter, so that the simulation spectrum and the measured spectrum are highly consistent in peak shape and tailing characteristics; meanwhile, the simulation data can be used to generate a large number of training sets of rare nuclides, and the core pain points of difficult acquisition of radioactive samples and limited coverage spectrum in traditional machine learning are solved, thereby providing high-fidelity database support for nuclide identification in complex scenarios.

[0036] This application uses simulated data to generate a training database, overcoming the difficulty that many radioactive source data cannot be obtained through actual measurement, and enriching the types of radioactive source data in the nuclide library.

[0037] This application combines electric field simulation data with simulated radiation data and utilizes carrier transport characteristics to simulate characteristics such as the low-energy tailing of the full-energy peak of the cadmium zinc telluride detector, making the energy spectrum in the simulated energy spectrum database closer to the measured energy spectrum, thereby enhancing the effectiveness of the database.

[0038] This application's machine learning-based radionuclide identification algorithm significantly improves its ability to analyze mixed radionuclides and noisy data by automatically extracting implicit features from complex energy spectra, such as weak and overlapping peaks. Its robustness suppresses noise interference and adapts to performance differences between detectors. Combined with a lightweight model, it enables real-time analysis, meeting the needs of scenarios such as nuclear emergency response. The algorithm supports incremental learning and multimodal data fusion, rapidly adapting to new radionuclides or complex environments. It also reduces false alarm rates through probabilistic output and integrated learning strategies, demonstrating efficient and accurate intelligent decision-making in nuclear safety, medical diagnosis, and environmental monitoring, providing innovative solutions for nuclear technology.

[0039] In some embodiments, the step of simulating the energy deposition position distribution of the radiation source gamma rays in the CZT detector by GEANT4 specifically includes: constructing a 10×10×5 mm³ three-dimensional geometric model of the CZT detector; setting the radiation source 10 mm above the detector, and the radiation source types include Cs-137 and Co-60; simulating the three-dimensional coordinate distribution of energy deposition points generated by the interaction between the gamma rays and the detector; recording the energy value and spatial position information of each deposition point to form the energy deposition position distribution.

[0040] This application constructs a 10×10×5mm³ standard detector model and sets dual-source incidence conditions of Cs-137 and Co-60. Three-dimensional precise recording of the position and intensity of gamma-ray energy deposition is achieved in GEANT4, with a spatial resolution of up to submillimeter level. This not only provides physically realistic initial conditions for subsequent carrier transport calculations, but also ensures the applicability of the model across a wide energy range by fully covering typical energy segments, allowing the simulation results to be directly mapped to the spatial response characteristics of actual detectors.

[0041] In some embodiments, as Figure 4 and Figure 5As shown, the step of constructing the detector electric field model and obtaining the potential distribution on the carrier drift path specifically comprises: establishing a 10x10x5mm3 CZT detector three-dimensional geometric model in COMSOL; setting a 1mm diameter point electrode on the bottom surface of the detector, applying a 900V high voltage, and grounding the remaining five surfaces; and calculating the three-dimensional electric field distribution inside the detector by using extremely refined grid division. The point on the central axis, i.e., the point closest to the central axis, is selected to view the internal electric field distribution of the detector, as shown in FIG. 3. Figure 6 As shown, the potential distribution is extremely similar to that of an ideal hemispherical electrode.

[0042] Based on the COMSOL electric field modeling, the present application adopts a unique design of a 900V high voltage point electrode and extremely refined grid, forms a highly non-uniform weight electric field distribution inside the detector, and the potential gradient change is highly consistent with the charge collection characteristics of the real CZT detector. By extracting the continuous potential distribution data on the carrier drift path, key field strength parameters are provided for the induced charge calculation of the Shockley-Ramo theorem. This cross-physical field coupling simulation significantly improves the accuracy of the signal generation link, and especially realizes millimeter-level precision quantitative characterization of the charge collection efficiency in the strong electric field region near the point electrode.

[0043] In some embodiments, the step of generating the initial carrier number subject to Gaussian distribution according to the energy deposition value and the energy required to generate a pair of carriers specifically comprises: determining the energy required for a pair of electrons-holes generated by a gamma ray in the CZT; calculating the theoretical carrier pair number according to the deposition energy value; introducing a Fano factor to establish a Gaussian distribution model of the carrier number; and randomly generating the initial carrier number subject to statistical fluctuation characteristics for each deposition event.

[0044] Specifically, the energy required for a pair of electrons-holes generated by a gamma ray in the CZT is E pair = 4.6eV, and the calculation formula of the carrier pair number generated by a gamma ray with energy E γ is: , wherein: N is the carrier pair number, E γ is the energy of the gamma ray, E pair is the energy required for a pair of electrons-holes generated in the CZT.

[0045] For a semiconductor detector, the statistical fluctuation of the carrier is related to the Fano factor F: , wherein: ΔN is the statistical fluctuation of the carrier number, F is the Fano factor, the Fano factor of the CZT semiconductor is F = 0.89, which represents the dispersion degree of the statistical process, N is the carrier pair number.

[0046] For a semiconductor detector, according to the Shockley-Ramo theorem, the instantaneous induced current i on the electrode caused by a moving point charge q is proportional to the drift velocity v of the charge multiplied by the weighting field Ew: wherein: is the instantaneous induced current on the electrode, is the charge amount of the moving point charge, is the drift velocity of the charge, is the weighting field.

[0047] The theorem can be equivalently expressed as that the accumulated charge amount on the electrode is proportional to the change amount of the weighting potential φ in the process of the charge q moving from the starting point a to the ending point b.

[0048] wherein: is the accumulated induced charge amount on the electrode, is the charge amount of the moving point charge, and are the weighting potentials at the starting point a and the ending point b of the charge.

[0049] Based on the fixed ionization energy threshold of 4.6 eV / eV and the Gaussian fluctuation model of the Fano factor 0.89, the present application converts the γ-ray deposition energy into the initial carrier number with statistical fluctuation characteristics, the randomization processing of which not only reflects the inherent quantum efficiency fluctuation of the semiconductor detector, but also guarantees the energy linearity of the energy spectrum response through strict energy-carrier number linear relationship. The statistical modeling of microscale makes the simulation energy spectrum keep statistical consistency with the measured energy spectrum in details such as count fluctuation and peak-to-valley ratio.

[0050] In some embodiments, the carrier mobility-lifetime product parameter includes an electron mobility-lifetime product and a hole mobility-lifetime product, wherein a typical value of the electron mobility-lifetime product is set in a range of 1x10 - ³ cm² / V to 1x10 - ² cm² / V, and a typical value of the hole mobility-lifetime product is set in a range of 1x10 -5 cm² / V to 1x10 -4 cm² / V.

[0051] Specifically, during the drift process of the carriers, there is a capture phenomenon, and therefore an exponential decay processing is required for the number of carriers in the simulation process: wherein: is the remaining carrier number after the drift distance x, is the initial carrier number, x is the drift distance of the carrier, and μτ is the mobility-lifetime product of the carrier, the values of which are different for electrons and holes. For a CZT semiconductor, μe τ e ≈1×10 -3 cm 2 / V~1×10 -2 cm 2 / V,μ h τ h ≈1×10 -5 cm 2 / V, E is the local electric field strength.

[0052] This application sets the electron mobility lifetime product to 10 - ³~10 - ²cm² / V and the product of hole mobility lifetime 10 -5 ~10 - 4 This method accurately captures the intrinsic property of CZT semiconductors, where the hole capture rate is two orders of magnitude higher than that of electrons, within a differentiated parameter range of cm² / V. This physically based parameterization enables simulation results to spontaneously exhibit the typical characteristics of the real energy spectrum, where the electron peak is symmetrical and the hole peak is tailing. This provides a search space for subsequent parameter optimization that both meets theoretical expectations and retains room for adjustment.

[0053] In some embodiments, the step of calculating the amount of charge induced on the electrode when electrons and holes drift based on the Shockley-Ramo theorem specifically includes: calculating the change in weighted potential caused by carrier movement within each drift step; accumulating the instantaneous charge induced by electrons and holes on the anode and cathode respectively; terminating the calculation when the carrier reaches the detector boundary; and accumulating the induced charge of all drift steps as the final signal output of the case.

[0054] Specifically, first set μ e τ e =1×10 -3 cm 2 / V,μ h τ h =1×10 -5 cm 2 / V, which is the typical value of the parameter; according to the Cs-137 radiation source simulation of GEANT4, the starting position P0 of the deposited energy and the energy E are determined; with P0 as the benchmark, the grid data of the COMSOL simulation is searched to find the electric field strength and potential of the point closest to P0, which are approximated to the electric field and potential of P0; according to the energy E and the energy Epair required to generate a pair of carriers, as well as the Fano factor of CZT, the number of electron-hole pairs N0, that is, the initial carrier number, is randomly generated. , the initial carrier number Obey Gaussian distribution; , where: N0is the initial carrier number, N is the mean value, σ is the standard deviation.

[0055] The step of carrier drift is set to be 0.1 mm, and the next position P1is obtained according to the direction of the electric field intensity; the number of lost carriers in the process of one-step drift is calculated according to the exponential decay formula of carriers, and the remaining number of electron carriers and hole carriers is determined Nehand Nph, The calculation formula is: ; ; In the formula, Nehis the remaining number of electron carriers, Nphis the remaining number of hole carriers, N0is the initial carrier number, and x is the drift distance of the carrier, μτ is the product of the mobility and the lifetime of the carrier, and E is the local electric field intensity.

[0056] According to the Shockley-Ramo theorem, the induced charge amounts caused by the cathode and the anode in the process of one-step drift are calculated, and the calculation formula is: ; ; In the formulae: and Nehand Nph, Nehis the remaining number of electron carriers, Nphis the remaining number of hole carriers, Vw is the weight potential in the process of one-step drift. The above steps are repeated until the carrier reaches the boundary of the detector.

[0057] The total amount of induced charge of the electron and the hole is added to obtain the total collected charge amount; the above steps are repeated for not less than 10,000 GEANT4 simulation cases to obtain the Cs-137 source induced charge spectrum; the above steps are repeated according to the Co-60 radiation source simulation of GEANT4 to obtain the Co-60 source induced charge spectrum; the obtained induced charge spectrum is energy-scaled, and the background energy spectrum obtained by the real detector is superimposed thereon to obtain the simulation energy spectrum spec_cs_simu and spec_co_simu of Cs-137 and Co-60.

[0058] The application adopts a 0.1 mm step discrete drift calculation combined with the Shockley-Ramo theorem to realize the piecewise accumulation of the transient induced charge in the carrier transport process. The method accurately quantifies the signal loss caused by the carrier trapping effect by tracking the differential induced signals of the electrons and holes on the double electrodes in real time. The boundary arrival judgment mechanism and the continuous potential integral algorithm jointly guarantee the calculation efficiency and physical authenticity of the charge collection amount. The finally output pulse height spectrum completely retains the inherent energy resolution and nonlinear response characteristics of the detector.

[0059] In some embodiments, the step of superimposing the simulated energy spectrum and the measured background energy spectrum and then performing normalization processing, and optimizing the mobility lifetime product parameters by the least square method specifically comprises: collecting the measured Cs-137 and Co-60 energy spectra as reference data; selecting different parameter values between 0.1 times and 10 times of the typical values of the electron mobility lifetime product and the hole mobility lifetime product; performing energy calibration and background superposition on the simulated energy spectrum generated by each group of parameters to perform normalization processing; determining the optimal parameter combination by minimizing the least square calculation value; and taking the optimal parameter combination as the parameter for subsequent simulation calculation.

[0060] Specifically, the simulated energy spectrum and the real energy spectrum spec_cs_real and spec_co_simu are normalized, and the least square calculation is performed: ; In the formula: Cs-137 simulated energy spectrum, Cs-137 real energy spectrum, Co-60 simulated energy spectrum, Co-60 real energy spectrum, wherein spec_xx_xx(i) is the count of the i-th channel of the corresponding normalized energy spectrum, and thus the least square calculation value corresponding to the parameters μeτe and μhτh can be obtained.

[0061] Different parameter values are selected between 0.1 times and 10 times of the typical values of the parameters μeτe and μhτh, and the above steps are repeated; The least square calculation value The corresponding parameters μeτe and μhτh are the best parameters obtained by simulation. The carrier mobility lifetime product of the simulation program is set to the best μeτe and μhτh; The application takes the Cs-137 and Co-60 measured energy spectrum as the benchmark, and the method is optimized through scanning in the range of 0.1-10 times the parameters, which quickly locks the global optimal solution in the mobility life product parameter space to minimize χ², and the innovation is that the energy calibration, background superposition and normalization processing are embedded in the optimization process, so that the parameter calibration process simultaneously considers the matching requirements of multi-dimensional such as spectrum shape, absolute count rate and background noise, and finally obtains the parameter combination that makes the simulation spectrum reach the optimal fitting degree at the same time in the 662 keV and 1.33 MeV double characteristic peak area.

[0062] Optionally, the step of regenerating the complete database containing single nuclides and mixed nuclides using the optimized parameters for machine learning training specifically includes: selecting radioactive sources of different nuclides and different positions to generate a single nuclide spectrum library; generating multiple Poisson distribution random samples for each nuclide; randomly selecting several nuclide spectra and superimposing them in different proportions; generating multiple training samples for each mixed combination; and establishing a complete database containing single nuclides and mixed nuclides.

[0063] Specifically, radioactive sources of different nuclides and different positions are selected to generate different GEANT4 simulation spectrum databases; each set of data in the database is obtained according to the above steps to obtain different simulation spectra; according to the need to simulate the radioactivity intensity, not less than 10000 new simulation spectra are randomly generated according to the Poisson distribution and superimposed with the real background spectrum to form a single nuclide database; the simulation spectra of 2 different nuclides or 3 different nuclides are randomly selected and superimposed, and not less than 10000 random simulation spectra are generated for each combination according to the Poisson distribution to form a mixed nuclide database.

[0064] The database generated based on the optimized parameters introduces the count fluctuation consistent with the statistical law of coincidence radioactivity decay through the Poisson distribution, and the single nuclide library covers typical radioactive sources from low energy to high energy, and the mixed library simulates complex radiation field environment through random combination and proportion adjustment. This data architecture with physical accuracy and scene diversity makes the trained machine learning model have strong generalization ability for challenging tasks such as overlapping peak analysis and weak peak identification, and provides a plug-and-play solution for application scenarios with high real-time requirements such as nuclear emergency response.

[0065] Please refer to Figure 7In another aspect, the present application also provides a machine learning simulation database generation system based on CZT carrier transport characteristics and energy spectrum characteristics, comprising: an energy deposition distribution simulation module 10 for simulating the energy deposition position distribution data of gamma rays of a radioactive source in a CZT detector by GEANT4; an electric field modeling module 20 for constructing a detector electric field model according to the energy deposition position distribution data using COMSOL Multiphysics to obtain the potential distribution on the carrier drift path; a carrier generation module 30 for generating the initial number of carriers subject to Gaussian distribution according to the energy deposition value and the energy required to generate a pair of carriers in combination with the Fano factor; a capture attenuation calculation module 40 for setting the carrier mobility lifetime product parameter and iteratively calculating the capture attenuation in the carrier drift process according to a preset step size; a charge amount calculation module 50 for calculating the charge amount induced on the electrode when the electron and hole drift according to the Shockley-Ramo theorem; a database generation module 60 for repeatedly calculating a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; a parameter optimization module 70 for normalizing the simulation energy spectrum after superimposing the measured background energy spectrum, and optimizing the mobility lifetime product parameter by the least square method; and a complete database generation module 80 for regenerating a complete database containing single nuclides and mixed nuclides using the optimized parameter for machine learning training.

[0066] In another aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned machine learning simulation database generation method based on CZT carrier transport characteristics and energy spectrum characteristics when executing the computer program.

[0067] The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the whole or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0068] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of each method. Any reference to memory, storage, database or other medium used in each embodiment of the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0069] The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent transformation or direct or indirect application in the related technical field using the content of the present application specification and drawings is also included in the patent protection scope of the present application.

Claims

1. A method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics, characterized in that the steps include: The energy deposition position distribution data of the radiation source γ-ray in the CZT detector was simulated by GEANT4; Build a detector electric field model based on the energy deposition position distribution data to obtain the potential distribution on the carrier drift path; According to the energy deposition value and the energy required to generate a pair of carriers, the initial number of carriers obeying the Gaussian distribution is generated in combination with the Fano factor; Set the carrier mobility lifetime product parameter and iteratively calculate the capture decay during the carrier drift process according to the preset step size; Based on the Shockley-Ramo theorem, the amount of charge induced on the electrode when electrons and holes drift is calculated step by step; Repeat the calculation for a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; The simulated energy spectrum is superimposed with the measured background energy spectrum and then normalized, and the mobility lifetime product parameter is optimized by the least squares method; The optimized parameters are used to regenerate a complete database containing single and mixed nuclides for machine learning training.

2. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The step of simulating the energy deposition position distribution of the radiation source gamma ray in the CZT detector by GEANT4 specifically includes: Construct a 10×10×5 mm³ CZT detector 3D geometric model; The radioactive source is set 10 mm above the detector. The types of radioactive sources include Cs-137 and Co-60. Simulate the three-dimensional coordinate distribution of energy deposition points generated by the interaction between gamma rays and detectors; The energy value and spatial position information of each deposition point are recorded to form the energy deposition position distribution.

3. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The steps of constructing a detector electric field model and obtaining the potential distribution on the carrier drift path specifically include: A 10×10×5 mm³ CZT detector 3D geometric model was established in COMSOL. A point electrode with a diameter of 1 mm is set on the bottom surface of the detector, a high voltage of 900 V is applied, and the other five surfaces are grounded; An extremely fine grid division method is used to calculate the three-dimensional electric field distribution inside the detector.

4. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The step of generating an initial number of carriers that obeys a Gaussian distribution based on the energy deposition value and the energy required to generate a pair of carriers in combination with the Fano factor specifically includes: Determine the energy of gamma rays to create an electron-hole pair in CZT; Calculate the theoretical number of carrier pairs based on the deposition energy value; The Fano factor is introduced to establish a Gaussian distribution model of carrier number; The initial carrier number that conforms to the statistical fluctuation characteristics is randomly generated for each deposition event.

5. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The carrier mobility lifetime product parameter includes the electron mobility lifetime product and the hole mobility lifetime product, wherein the typical value setting range of the electron mobility lifetime product is 1×10 - ³cm² / V to 1×10 - ²cm² / V, the typical value setting range of the hole mobility lifetime product is 1×10 -5 cm² / V to 1×10 -4 cm² / V.

6. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The step of calculating the amount of charge induced on the electrode when electrons and holes drift based on the Shockley-Ramo theorem specifically includes: In each drift step, the change in weight potential caused by carrier movement is calculated; Accumulate the instantaneous charges induced by electrons and holes on the anode and cathode respectively; The calculation is terminated when the carrier reaches the detector boundary; The induced charge of all drift steps is accumulated as the final signal output of this case.

7. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 5, characterized in that: The steps of superimposing the simulated energy spectrum and the measured background energy spectrum, performing normalization processing, and optimizing the mobility lifetime product parameter by the least squares method specifically include: Collect measured Cs-137 and Co-60 energy spectra as baseline data; Select different parameter values ​​between 0.1 times and 10 times the typical values ​​of the electron mobility lifetime product and the hole mobility lifetime product; The simulated energy spectrum generated by each set of parameters is normalized by energy scaling and background superposition; Determine the optimal parameter combination by minimizing the least squares calculation value; The optimal parameter combination is used as the parameters for subsequent simulation calculations.

8. The method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics according to claim 1, characterized in that: The step of using the optimized parameters to regenerate a complete database containing single nuclides and mixed nuclides for machine learning training specifically includes: Select radioactive sources of different nuclides and locations to generate a single nuclide energy spectrum library; Generate multiple Poisson-distributed random samples for each nuclide; Randomly select several nuclide energy spectra and superimpose them in different proportions; Generate multiple training samples for each mixture combination; Build a complete database containing single and mixed nuclides.

9. A machine learning simulation database generation system based on CZT carrier transport characteristics and energy spectrum characteristics, characterized by: include: Energy deposition distribution simulation module, used to simulate the energy deposition position distribution data of the radioactive source gamma ray in the CZT detector through GEANT4; The electric field modeling module is used to build a detector electric field model using COMSOL Multiphysics based on the energy deposition position distribution data to obtain the potential distribution along the carrier drift path; A carrier generation module is used to generate an initial number of carriers that obeys a Gaussian distribution based on the energy deposition value and the energy required to generate a pair of carriers in combination with the Fano factor; The capture decay calculation module is used to set the carrier mobility lifetime product parameter and iteratively calculate the capture decay during the carrier drift process according to the preset step size; The charge calculation module is used to calculate the charge induced on the electrode when electrons and holes drift based on the Shockley-Ramo theorem. A database generation module is used to repeatedly calculate a preset number of simulation cases to generate a nuclide characteristic energy spectrum database; The parameter optimization module is used to normalize the simulated energy spectrum after superposition with the measured background energy spectrum, and optimize the mobility lifetime product parameter by the least squares method; The complete database generation module is used to regenerate a complete database containing single and mixed nuclides using optimized parameters for machine learning training.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for generating a machine learning simulation database based on CZT carrier transport characteristics and energy spectrum characteristics as described in any one of claims 1 to 8 are implemented.

Images