CN-122019989-A - Nuclear radiation energy spectrum double-domain collaborative optimization method and related device

CN122019989ACN 122019989 ACN122019989 ACN 122019989ACN-122019989-A

Abstract

The invention discloses a nuclear radiation energy spectrum double-domain collaborative optimization method and a related device, belonging to the technical field of nuclear radiation measurement, wherein the method comprises the steps of obtaining an original distorted pulse sequence generated by nuclear radiation measurement; the method comprises the steps of constructing a dual-domain collaborative energy spectrum optimization model, wherein the dual-domain collaborative energy spectrum optimization model comprises a pulse domain restoration model and an energy spectrum domain reconstruction model which are arranged in parallel, the pulse domain restoration model is used for carrying out pulse-by-pulse restoration on the amplitude and the arrival time of an original distorted pulse sequence, outputting a restored pulse sequence, and the energy spectrum reconstruction model is used for converting the restored pulse sequence into an energy spectrum, and in the training process of the dual-domain collaborative energy spectrum optimization model, the pulse domain restoration model and the energy spectrum reconstruction model share the counter-propagation gradient of a joint loss function, so that the bidirectional closed-loop collaborative optimization of the pulse restoration model and the energy spectrum reconstruction model can be realized, the problem of error gradual amplification caused by a traditional serial processing mode is avoided, and the authenticity and the accuracy of an output energy spectrum are remarkably improved.

Inventors

TANG LIN
SHI KAIBO
YU YUE
ZHOU NAN
WANG WENHAO
Xie Yuanlun
TENG YUNLONG
QI JUN

Assignees

成都大学

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. The nuclear radiation energy spectrum double-domain collaborative optimization method is characterized by comprising the following steps of: acquiring an original distorted pulse sequence generated by nuclear radiation measurement; The method comprises the steps of constructing a double-domain collaborative energy spectrum optimization model, wherein the double-domain collaborative energy spectrum optimization model comprises a pulse domain restoration model and an energy spectrum domain reconstruction model which are arranged in parallel, the pulse domain restoration model is used for carrying out pulse-by-pulse restoration on the amplitude and the arrival time of an original distorted pulse sequence, outputting a restored pulse sequence, and the energy spectrum domain reconstruction model is used for converting the restored pulse sequence into an energy spectrum; in the training process of the two-domain collaborative energy spectrum optimization model, the pulse domain repair model and the energy spectrum domain reconstruction model share the counter-propagation gradient of the joint loss function.
2. The nuclear radiation energy spectrum double-domain collaborative optimization method according to claim 1, wherein the pulse domain repair model adopts a time sequence feature extraction network, and is any one of a lightweight UNet and a long-term and short-term memory neural network; the energy spectrum domain reconstruction model adopts an energy spectrum characteristic reconstruction network, and is any one of a frequency spectrum transducer, a depth residual error network or a multi-layer perceptron; the energy spectrum characteristic reconstruction model is embedded with a learnable nuclide base function library, the output energy spectrum of the energy spectrum characteristic reconstruction model is matched with the characteristic peak shape of the nuclide base function library through a soft alignment mechanism, and the calibrated energy spectrum is used as a supervision signal, so that the relative entropy of the output energy spectrum and the calibrated energy spectrum is smaller than a preset threshold value.
3. The method of claim 1, wherein a physical constraint term is embedded into the joint loss function during training of the two-domain collaborative spectrum optimization model, the physical constraint term including at least one of an energy conservation term, a peak position consistency term, and a noise floor term.
4. The nuclear radiation energy spectrum double-domain collaborative optimization method according to claim 3, wherein the energy conservation term is used for guaranteeing that the integral count of a repaired pulse sequence output by a pulse domain repair model is equal to the integral count of an original distorted pulse sequence, the peak position consistency term is used for restraining peak position deviation caused by temperature drift by taking an internal standard element characteristic peak as an anchor point, and the noise lower limit term is used for introducing a Fano factor and electronic noise power spectrum constraint.
5. The nuclear radiation energy spectrum dual-domain co-optimization method according to claim 1, wherein the method further comprises constructing a distorted pulse dataset: generating an initial distortion pulse data set covering different nuclear radiation measurement scenes, different energy intervals and counting rates through software simulation; Collecting actual measurement pulses in a standard radiation field, establishing a Bayesian inversion mapping relation between physical parameters to be corrected and actual measurement pulse output in a physical model adopted by software simulation, correcting the physical parameters through the mapping relation, performing SHAP value sensitivity analysis on the physical model, correcting charge sharing coefficients and polarization correction factors in the physical model, enabling the peak position difference between initial distortion pulses and actual measurement pulses to be smaller than a preset channel number, and realizing the calibration treatment on the initial distortion pulses; Distributing the pulse generation script after the calibration processing to a plurality of nodes through the federal learning framework, locally running the pulse generation script by each node to generate distortion pulses conforming to the characteristics of the local scene, training the local model based on the local distortion pulse data, uploading gradient statistics after the differential privacy processing, and realizing federal expansion of an initial distortion pulse data set after the calibration processing.
6. The method of claim 1, further comprising the step of small sample cross-domain robust training: Constructing a joint physical engine with multiple distortion mechanisms, coupling a circuit-level noise model, and generating multiple virtual distortion pulses; Constructing a twin network architecture of a main network and a twin network, inputting a real pulse sample by the main network, inputting a corresponding virtual distorted pulse sample by the twin network, reducing the distribution distance of the real pulse sample and the virtual distorted pulse sample in a latent space by contrast learning, and introducing physical regularization item constraint to generate the rationality of the sample; Designing a three-dimensional parameter space hierarchical sampling strategy for temperature, dose and device aging, mapping a real pulse sample to a plurality of domain distribution anchor points, wherein each anchor point represents different temperature, dose and aging condition combinations, and introducing a maximum mean difference loss function to carry out domain self-adaptive alignment; And adopting a federal transfer learning framework, wherein each node stores original data and only exchanges gradient statistics subjected to differential privacy treatment.
7. The method of claim 1, further comprising the step of deploying a dual-domain collaborative energy spectrum optimization model: The end side adopts a silicon drift detector array to cooperate with a time-to-digital converter to realize lossless digital processing of an original distorted pulse sequence; The side takes an FPGA and an on-chip microprocessor as cores, a hardware IP core of a dual-domain collaborative energy spectrum optimization model is deployed, the FPGA is used for realizing parallel computation of a pulse domain repair model and an energy spectrum domain reconstruction model, and the on-chip microprocessor updates parameters of the dual-domain collaborative energy spectrum optimization model according to the issued federal aggregation gradient; And the cloud side and the side are kept synchronous, gradient statistics subjected to differential privacy treatment are received, and the side double-domain collaborative energy spectrum optimization model parameters are updated through a federal learning protocol.
8. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the nuclear radiation energy spectrum two-domain co-optimization method of any one of claims 1-7.
9. A storage medium having stored thereon computer instructions which, when run, perform the steps of the nuclear radiation energy spectrum dual domain co-optimization method of any one of claims 1-7.
10. A terminal comprising a memory and a processor, said memory having stored thereon computer instructions executable on said processor, characterized in that said processor executes the steps of the dual domain co-optimization method of nuclear radiation energy spectrum of any of claims 1-7 when said computer instructions are executed.

Description

Nuclear radiation energy spectrum double-domain collaborative optimization method and related device Technical Field The invention relates to the technical field of nuclear radiation measurement, in particular to a nuclear radiation energy spectrum double-domain collaborative optimization method and a related device. Background In the nuclear radiation measurement process, the detector converts incident rays into electric pulse signals, and the pulse amplitude is analyzed to generate an energy spectrogram, so that the qualitative and quantitative analysis of nuclides is realized. However, in an actual measurement scenario, the original pulse output by the detector often deviates from the ideal form due to physical limits, environmental disturbances, etc., forming a distorted pulse. In the prior art, when the problems of distorted pulses and energy spectrum distortion are solved, pulse repair is generally carried out firstly, and then energy spectrum is generated based on the repaired pulses. The processing mode can directly transfer and amplify the error of the pulse repairing stage to the energy spectrum generating stage, and the whole information of the energy spectrum domain cannot be fed back to guide the pulse repairing process. Such serial processing mechanisms can lead to progressive amplification of errors, making the quality of the final spectrum difficult to guarantee, especially when dealing with complex distortions or multiple distortion stacks. For example, pulse pile-up induced amplitude distortion, if not repaired, can form false peaks in the energy spectrum or cause peak position shifts. In addition, in the prior art, the pulse repairing and the energy spectrum generation are mutually split, and the whole information of the energy spectrum domain is difficult to constrain and correct the pulse repairing process. Even if there is a slight deviation in the pulse repair phase, it may be amplified during the subsequent energy spectrum generation process, resulting in a significant difference between the final energy spectrum and the real situation. Therefore, how to realize the collaborative optimization of pulse repair and energy spectrum reconstruction and avoid error accumulation and amplification is a technical problem to be solved in the field. Disclosure of Invention The invention aims to solve the problems in the prior art and provides a nuclear radiation energy spectrum double-domain collaborative optimization method and a related device. The invention aims at realizing the technical scheme that the nuclear radiation energy spectrum double-domain collaborative optimization method comprises the following steps: acquiring an original distorted pulse sequence generated by nuclear radiation measurement; The method comprises the steps of constructing a double-domain collaborative energy spectrum optimization model, wherein the double-domain collaborative energy spectrum optimization model comprises a pulse domain restoration model and an energy spectrum domain reconstruction model which are arranged in parallel, the pulse domain restoration model is used for carrying out pulse-by-pulse restoration on the amplitude and the arrival time of an original distorted pulse sequence, outputting a restored pulse sequence, and the energy spectrum domain reconstruction model is used for converting the restored pulse sequence into an energy spectrum; in the training process of the two-domain collaborative energy spectrum optimization model, the pulse domain repair model and the energy spectrum domain reconstruction model share the counter-propagation gradient of the joint loss function. In an example, the pulse domain repair model adopts a time sequence feature extraction network, which is any one of a lightweight UNet and a long-short-term memory neural network; the energy spectrum domain reconstruction model adopts an energy spectrum characteristic reconstruction network, and is any one of a frequency spectrum transducer, a depth residual error network or a multi-layer perceptron; the energy spectrum characteristic reconstruction model is embedded with a learnable nuclide base function library, the output energy spectrum of the energy spectrum characteristic reconstruction model is matched with the characteristic peak shape of the nuclide base function library through a soft alignment mechanism, and the calibrated energy spectrum is used as a supervision signal, so that the relative entropy of the output energy spectrum and the calibrated energy spectrum is smaller than a preset threshold value. In an example, during training of the two-domain collaborative energy spectrum optimization model, a physical constraint term is embedded into the joint loss function, the physical constraint term including at least one of an energy conservation term, a peak position consistency term, and a noise floor term. In an example, the energy conservation term is used for ensuring that the integral count of the repaired