Gamma dose rate monitoring method and device based on eight paths of SiPINs

Abstract

The invention relates to the field of gamma dose rate monitoring, and provides a gamma dose rate monitoring method and device based on eight SiPINs, and the method comprises the steps: collecting pulse signals outputted by eight SiPIN detectors, inputting the pulse signals into a self-adaptive time domain convolutional neural network of an index attenuation response function of the SiPIN detectors, and obtaining a stacked pulse feature vector; processing the stacked pulse feature vector by adopting a multi-path synchronous constraint loss function to obtain pulse arrival time, pulse amplitude and confidence coefficient parameters; constructing node features of the graph neural network based on the pulse arrival time, the pulse amplitude and the confidence coefficient parameter, and performing fusion calculation processing on the node features through the graph neural network to obtain a gamma dose rate measurement value; and comparing the gamma dose rate measurement value with a preset gamma dose rate threshold value, generating a corresponding monitoring state identifier according to a comparison result, and obtaining a gamma dose rate monitoring result. According to the invention, high-sensitivity real-time monitoring, dynamic pulse accumulation processing and grading state early warning of the gamma dose rate are realized.

Description

Technical Field

[0001] This invention relates to the field of gamma dose rate monitoring, and more particularly to a gamma dose rate monitoring method and apparatus based on 8-channel SiPIN. Background Technology

[0002] Gamma-ray dose rate monitoring refers to a technical system for real-time detection and assessment of gamma-ray radiation intensity in the environment. Gamma rays, as high-energy electromagnetic radiation, are widely present in nuclear power plants, medical facilities, industrial inspection sites, and other locations. They possess strong penetrating power and can cause varying degrees of damage to human cells and tissues. High-precision gamma-ray dose rate monitoring systems can promptly detect radiation anomalies, assess radiation risks, and provide data support for radiation protection. They play a crucial role in nuclear safety regulation, environmental protection, and public health safeguards, offering stronger early warning capabilities, higher measurement accuracy, and a wider range of applications. However, gamma-ray dose rate monitoring also faces technical challenges, the key being how to achieve high-sensitivity detection in complex radiation environments, suppress background noise interference, and ensure long-term stable operation.

[0003] In existing technologies, gamma dose rate monitoring mainly employs gas detectors (such as ionization chambers) and semiconductor detectors for radiation detection, single detectors or arrays of a small number of detectors for signal acquisition, and traditional linear processing algorithms for data analysis, thus achieving basic gamma ray detection and dose rate calculation functions. However, gas detectors are bulky and require high airtightness, semiconductor detectors are extremely expensive, and single detectors suffer from small detection area and low signal-to-noise ratio. Traditional processing algorithms struggle to effectively distinguish gamma ray pulse signals from background noise and lack multi-channel signal fusion mechanisms and adaptive noise suppression capabilities. This results in poor performance of existing systems under low dose rate environments, strong noise interference, and long-term continuous monitoring conditions, particularly when facing complex factors such as ambient background radiation, electronic device noise, and temperature drift, making it difficult to achieve high-sensitivity, low-cost, and accurate gamma dose rate monitoring. Summary of the Invention

[0004] In view of this, the present invention proposes a gamma dose rate monitoring method and device based on 8-channel SiPIN, which solves the problem that the existing technology is difficult to effectively distinguish gamma-ray pulse signals from background noise, and lacks a multi-channel signal fusion mechanism and adaptive noise suppression capability. This makes the existing system perform poorly in low dose rate environments, strong noise interference and long-term continuous monitoring conditions, especially when facing complex factors such as ambient background radiation, electronic device noise and temperature drift, making it difficult to achieve high-sensitivity and low-cost accurate gamma dose rate monitoring.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for monitoring gamma dose rate based on 8-channel SiPIN, comprising the following steps: The pulse signals output from 8 SiPIN detectors are acquired, and the pulse signals are input into an adaptive time-domain convolutional neural network embedded with the exponential decay response function of the SiPIN detector to obtain a stacked pulse feature vector. The stacked pulse feature vector is processed using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. Based on the pulse arrival time, the pulse amplitude, and the confidence parameter, a graph neural network is constructed to generate node features. The graph neural network is then used to perform fusion calculations on the node features to obtain the γ dose rate measurement value. The measured γ dose rate is compared with a preset γ dose rate threshold, and a corresponding monitoring status identifier is generated based on the comparison result to obtain the γ dose rate monitoring result.

[0006] Based on the above technical solutions, preferably, the multi-channel synchronization constraint loss function of the hierarchical variational autoencoder with geometric constraints is used to process the stacked pulse feature vector to obtain pulse arrival time, pulse amplitude, and confidence parameters, including: The geometrically constrained hierarchical variational autoencoder adopts a cross-channel multi-scale latent variable mapping method to construct an independent variational latent space for the feature vectors of each stacked pulse, and introduces a cross-channel set of physical constraint parameters for similar pulses to collaboratively model the correlation characteristics between multiple spatial channels, thereby obtaining pulse arrival time, pulse amplitude, and confidence parameters.

[0007] Based on the above technical solutions, preferably, the multipath physical constraint parameter set of the variational latent space includes: The prior distribution of latent variables is dynamically corrected based on the actual crosstalk matrix of the detector, the noise covariance of each amplifier, and the time drift trend. During the training process, the cosine distance, variance difference, and pulse continuity target between the prior distribution of latent variables and the output distribution are jointly optimized by the multi-channel synchronization constraint loss function to obtain the pulse arrival time, pulse amplitude, and confidence parameters.

[0008] Based on the above technical solution, preferably, the step of acquiring the pulse signals output by the 8-channel SiPIN detector and inputting the pulse signals into an adaptive time-domain convolutional neural network embedded in the exponential decay response function of the SiPIN detector to obtain a stacked pulse feature vector includes: Each output pulse signal of the eight SiPIN detectors is associated with the physical spatial arrangement parameters. An array temporal coupling characteristic model is established based on the spatial arrangement relationship and response function of each SiPIN detector. The eight pulse signals are then jointly modeled based on the array temporal coupling characteristic model to obtain the stacked pulse feature vector embedded with the array spatial structure parameters.

[0009] Based on the above technical solutions, preferably, in the array time-domain coupling characteristic model, the coupling correction factor between each SiPIN detector and its corresponding adjacent detector is integrated, and the pulse signal of each detector is dynamically compensated according to the real-time crosstalk correction factor of each detector to adjust and output the stacked pulse feature vector, which includes spatial position, response characteristics and coupling correction.

[0010] Based on the above technical solutions, preferably, the step of constructing node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and then performing fusion calculation processing on the node features through the graph neural network to obtain the γ dose rate measurement value includes: Node feature encoding is performed on the pulse arrival time, the pulse amplitude, and the confidence parameter to construct the corresponding node feature vector; The node feature vectors are input into a graph neural network, and the γ dose rate measurement value is obtained through information transmission and aggregation calculation between nodes.

[0011] Based on the above technical solution, preferably, the step of comparing the measured γ dose rate value with a preset γ dose rate threshold, generating a corresponding monitoring status identifier based on the comparison result, and obtaining the γ dose rate monitoring result includes: The measured γ dose rate value is compared with a multi-level preset γ dose rate threshold to obtain the dose rate level judgment result. Based on the dose rate level judgment result and the working status parameters of the 8-channel SiPIN detector, a monitoring status identifier is generated, which includes the dose rate level and detector status information.

[0012] Secondly, the present invention also provides a γ dose rate monitoring device based on an 8-channel SiPIN, the device comprising: The feature extraction module is used to acquire the pulse signals output by 8 SiPIN detectors, and input the pulse signals into an adaptive time-domain convolutional neural network that embeds the exponential decay response function of the SiPIN detector to obtain the stacked pulse feature vector. The feature processing module is used to process the stacked pulse feature vector using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. The dose measurement module is used to construct node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and to perform fusion calculation processing on the node features through the graph neural network to obtain the γ dose rate measurement value. The dose monitoring module is used to compare the measured γ dose rate with a preset γ dose rate threshold, generate a corresponding monitoring status identifier based on the comparison result, and obtain the γ dose rate monitoring result.

[0013] Thirdly, the present invention also provides an electronic device, comprising: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to implement the steps of a gamma dose rate monitoring method based on 8-channel SiPIN.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing computer instructions that enable a computer to perform steps such as those of an 8-channel SiPIN-based gamma dose rate monitoring method.

[0015] The γ dose rate monitoring method and device based on 8-channel SiPIN of the present invention have the following advantages over the prior art: (1) By integrating the multi-channel parallel acquisition of 8 SiPIN detectors, the adaptive temporal convolution processing with embedded physical response functions, the decoupling of geometrically constrained hierarchical variational autoencoders and graph neural network fusion calculation, the convolution kernel parameters are dynamically optimized based on the exponential decay response function of the SiPIN detector. The hierarchical variational autoencoder architecture with geometric constraints is adopted, and the multi-channel synchronous constraint loss function and graph neural network node feature fusion mechanism are introduced to accurately extract features and decouple multi-dimensional parameters of the stacked pulse signal. This ensures high-precision identification of pulse arrival time, amplitude and confidence parameters and effective fusion of multi-channel signals, realizing high-sensitivity real-time monitoring of γ dose rate, dynamic pulse stacking processing and hierarchical state early warning. (2) By adopting a geometrically constrained hierarchical variational autoencoder for cross-channel multi-scale latent variable mapping, combined with physical prior factors such as the actual crosstalk matrix of the detector, the noise covariance of the amplifier, and the time drift trend, the prior distribution of latent variables is dynamically corrected and joint optimization is performed using a multi-channel synchronous constraint loss function. This achieves high-precision decoupling and parameter extraction of 8-channel SiPIN detector signals, effectively eliminating crosstalk interference between channels, compensating for the effects of amplifier noise and time drift, and improving the accuracy of pulse arrival time, amplitude, and confidence parameters identification under the stacked pulse environment. (3) By associating each output pulse signal of the 8-channel SiPIN detector with the physical spatial arrangement parameters, an array temporal coupling characteristic model is established and the coupling correction factor between adjacent detectors is fused. Based on the real-time crosstalk correction factor, dynamic compensation is performed on each pulse signal, which realizes the effective combination of the spatial structure information and temporal response characteristics of the multi-channel detector, improves the spatial positioning accuracy and temporal response accuracy of the stacked pulse feature vector, and effectively eliminates crosstalk interference between adjacent channels in the detector array. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0017] Figure 1 This is a flowchart of a γ dose rate monitoring method based on 8-channel SiPIN according to the present invention. Detailed Implementation

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

[0019] Please see Figure 1 This invention provides a method for monitoring gamma dose rate based on 8-channel SiPIN, comprising the following steps: The pulse signals output from 8 SiPIN detectors are acquired, and the pulse signals are input into an adaptive time-domain convolutional neural network embedded with the exponential decay response function of the SiPIN detector to obtain a stacked pulse feature vector. The stacked pulse feature vector is processed using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. Based on the pulse arrival time, the pulse amplitude, and the confidence parameter, a graph neural network is constructed to generate node features. The graph neural network is then used to perform fusion calculations on the node features to obtain the γ dose rate measurement value. The measured γ dose rate is compared with a preset γ dose rate threshold, and a corresponding monitoring status identifier is generated based on the comparison result to obtain the γ dose rate monitoring result.

[0020] Specifically, this embodiment integrates multi-channel parallel acquisition from eight SiPIN detectors, adaptive temporal convolution processing with embedded physical response functions, decoupling of geometrically constrained hierarchical variational autoencoders, and graph neural network fusion calculations. Based on the exponential decay response function of the SiPIN detector, the convolution kernel parameters are dynamically optimized. A geometrically constrained hierarchical variational autoencoder architecture is adopted, and a multi-channel synchronous constraint loss function and graph neural network node feature fusion mechanism are introduced to accurately extract features and decouple multi-dimensional parameters of the stacked pulse signal. This ensures high-precision identification of pulse arrival time, amplitude, and confidence parameters and effective fusion of multiple signals, achieving high-sensitivity real-time monitoring of γ dose rate, dynamic pulse stacking processing, and graded state early warning.

[0021] The process involves acquiring pulse signals output from eight SiPIN detectors, inputting these pulse signals into an adaptive time-domain convolutional neural network embedded with the exponential decay response function of the SiPIN detector, to obtain a stacked pulse feature vector, including: Each output pulse signal of the eight SiPIN detectors is associated with the physical spatial arrangement parameters. An array temporal coupling characteristic model is established based on the spatial arrangement relationship and response function of each SiPIN detector. The eight pulse signals are then jointly modeled based on the array temporal coupling characteristic model to obtain the stacked pulse feature vector embedded with the array spatial structure parameters.

[0022] In one specific embodiment, the output calculation formula of the array temporal coupling characteristic model is: ; in, The time variable (seconds) represents the same time coordinate in all formulas of this invention; and For detector channel index (1≤ , ≤8), indicating the detector number; For the first The output signal of the circuit detector after coupling correction (in volts). For the first The raw pulse signal of the circuit detector (in volts). For the first Road detector The pulse signal at a given moment; For the first The attenuation constant (1 / second) of the path detector indicates that each detector has unique attenuation characteristics; For the first The attenuation constant (1 / second) of the road detector, and They are the same type of parameter but have different values; For the first The set of adjacent detectors for the 8-channel detector is determined based on the physical spatial arrangement of the 8-channel detector; For the first Road and the First The time-varying coupling correction factor (dimensionless) between paths is dynamically adjusted over time. For detector and The signal transmission delay (nanoseconds) between them.

[0023] In the array time-domain coupling characteristic model, the coupling correction factors between each SiPIN detector and its corresponding adjacent detector are integrated, and the pulse signal of each detector is dynamically compensated according to the real-time crosstalk correction factor of each detector to adjust and output the stacked pulse feature vector data. The stacked pulse feature vector data includes spatial position, response characteristics and coupling correction.

[0024] In one specific embodiment, the calculation formula for dynamically compensating each pulse signal based on the real-time crosstalk correction factor of each detector is as follows: ; in, For the first crosstalk compensation Road signal (unit: volts); For the first The output signal of the circuit detector after coupling correction (in volts). For frequency variables (Hertz), used for frequency domain analysis; For the first Road and the Frequency domain crosstalk coefficient between paths (dimensionless). For Fourier transform operators, This is the inverse Fourier transform operator; For detector and The frequency domain transfer function (dimensionless complex number) between them. For the first The output signal after coupling correction of the road detector.

[0025] In one specific embodiment, the formula for calculating the stacked pulse feature vector is: ; in, For the first The stacked pulse feature vector (column vector) of the path; For the first crosstalk compensation Road signal (unit: volts); For the first The spatial position parameter vector of the road detector contains x ,y , z Coordinate information; For the first The response characteristic parameter vector of the road detector, including parameters such as gain and bandwidth; This is the coupling correction parameter matrix, which contains the coupling coefficients between all detectors; This is the spatial weight matrix, based on the weight allocation of the physical arrangement of the detectors; This is the Kronecker product (tensor product) operator; This is the transpose symbol.

[0026] Specifically, this embodiment establishes an array temporal coupling characteristic model by associating each output pulse signal of the 8-channel SiPIN detector with the physical spatial arrangement parameters and fusing coupling correction factors between adjacent detectors. Based on the real-time crosstalk correction factor, dynamic compensation is performed on each pulse signal, which realizes the effective combination of spatial structure information and temporal response characteristics of multiple detectors, improves the spatial positioning accuracy and temporal response accuracy of the stacked pulse feature vector, and effectively eliminates crosstalk interference between adjacent channels in the detector array.

[0027] The multi-channel synchronization constraint loss function of the hierarchical variational autoencoder with geometric constraints processes the stacked pulse feature vector to obtain pulse arrival time, pulse amplitude, and confidence parameters, including: The geometrically constrained hierarchical variational autoencoder adopts a cross-channel multi-scale latent variable mapping method to construct an independent variational latent space for the feature vectors of each stacked pulse, and introduces a cross-channel set of physical constraint parameters for similar pulses to collaboratively model the correlation characteristics between multiple spatial channels, thereby obtaining a set of parameters that include and distinguish the pulse arrival time, pulse amplitude, and confidence level of 8 signal sources.

[0028] In one specific embodiment, the calculation formula for the cross-channel multi-scale latent variable mapping is: ; in, For the first Road detector The vector of latent variables in scale; For scale index (1≤ ≤ K ), representing different feature extraction scales; For the first Lu Di Variational mean parameter of the scale; For the first Lu Di The variational standard deviation parameter of the scale; A random sampling vector from a standard normal distribution ; This is the element-wise product (Hadamard product) operator; It is an exponential function; Index of scaling transformation function (1≤ ≤ M ); This represents the total number of scaling transformation functions; For the first A scale transformation function is used for multi-scale feature extraction; For the first Weighting coefficients of each scaling function; For the first The stacked pulse eigenvectors (column vectors) of the path.

[0029] The variational latent space includes a set of multiple physical constraint parameters, which are dynamically modified based on physical prior factors such as the actual crosstalk matrix of the detector, the noise covariance of each amplifier, and the time drift trend. During training, the cosine distance, variance difference, and pulse coherence target between the latent variable prior distribution and the output distribution are jointly optimized by the multiple synchronization constraint loss function to obtain pulse arrival time, pulse amplitude, and confidence parameters with physical constraint consistency.

[0030] In one specific embodiment, the formula for calculating the prior distribution of the latent variable is: ; in, Latent variables In parameters The probability density function under the given conditions; For a latent variable vector, and Related but indicates an overall latent variable; This is the set of model parameters, containing all learnable parameters; This is the notation for a multivariate normal distribution. The prior mean vector; It is an 8×8 inter-detector crosstalk matrix; This is a crosstalk bias vector to compensate for systematic bias. Here is the amplifier noise covariance matrix; This is a time drift trend matrix.

[0031] In one specific embodiment, the multi-path synchronization constraint loss function is calculated as follows: ; in, The value of the multi-path synchronization constraint loss function; For the first The hidden variables of the path, and Related but indicating the comprehensive latent variables of this path, For the first Hidden variables of the path; The prior latent variable vector serves as a reference benchmark; It is a cosine distance metric function; This is the variance calculation function; These are the weighting coefficients of the loss function (dimensionless). For pulse type index; For the first The coherence regularization term for impulse-like events.

[0032] In one specific embodiment, the formulas for calculating the pulse arrival time, pulse amplitude, and confidence parameters are as follows: ; in, The pulse arrival time is in nanoseconds. The pulse amplitude (volts); For confidence parameters (0≤ ≤1); This is a decoder function that maps latent variables to physical parameters; The total number of scales; For the first Scale-based fusion weighting coefficients; For the first Road detector The vector of latent variables in scale; This is a vector of physical constraint bias terms to ensure that the output conforms to physical laws.

[0033] Specifically, this embodiment employs a geometrically constrained hierarchical variational autoencoder for cross-channel multi-scale latent variable mapping. By combining physical prior factors such as the actual crosstalk matrix of the detector, the amplifier noise covariance, and the time drift trend, it dynamically corrects the prior distribution of latent variables and uses a multi-channel synchronous constraint loss function for joint optimization. This achieves high-precision decoupling and parameter extraction of 8-channel SiPIN detector signals, effectively eliminating crosstalk interference between channels, compensating for amplifier noise and time drift effects, and improving the accuracy of pulse arrival time, amplitude, and confidence parameters identification in the stacked pulse environment.

[0034] The process of constructing node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and then fusing and calculating these node features through the graph neural network to obtain the γ dose rate measurement includes: The pulse arrival time, pulse amplitude, and confidence parameter are encoded using node features to construct corresponding node feature vectors.

[0035] In one specific embodiment, node feature encoding of the pulse arrival time, the pulse amplitude, and the confidence parameter includes: The arrival time, pulse amplitude, and confidence parameters of pulses with the same timing sequence are combined into a node feature, and arranged according to the spatial arrangement order of the 8 SiPIN detectors to obtain a node feature vector group containing spatial arrangement information.

[0036] The node feature vectors are input into a graph neural network, and the γ dose rate measurement value is obtained through information transmission and aggregation calculation between nodes.

[0037] In one specific embodiment, the step of information transfer and aggregation calculation between nodes includes: A graph structure is constructed based on the actual physical adjacency relationship of the 8-channel SiPIN detector. In each layer of the graph neural network, the feature vectors of all nodes are iteratively fused through the information weighting update between adjacent nodes and the global feature aggregation mechanism. Finally, the features of all nodes are aggregated in the output layer to obtain the γ dose rate measurement value.

[0038] Specifically, this embodiment encodes the pulse arrival time, pulse amplitude, and confidence parameters into node features, constructs node feature vectors containing spatial information according to the spatial arrangement order of the 8 SiPIN detectors, establishes a graph structure based on the actual physical adjacency relationship of the detectors, and adopts a weighted update mechanism for adjacent nodes and a global feature aggregation mechanism. Through iterative fusion calculation of multi-layer graph neural networks, it realizes spatial correlation modeling and collaborative information processing of multi-detector signals, effectively utilizes the spatial topology relationship and adjacency correlation of the detector array, and improves the accuracy of γ dose rate measurement.

[0039] The step of comparing the measured γ dose rate value with a preset γ dose rate threshold, generating a corresponding monitoring status identifier based on the comparison result, and obtaining the γ dose rate monitoring result includes: The measured γ dose rate value is compared with a multi-level preset γ dose rate threshold to obtain the dose rate level judgment result.

[0040] In one specific embodiment, the multi-level preset γ dose rate threshold includes a normal monitoring threshold, an early warning monitoring threshold, and a danger monitoring threshold. The γ dose rate measurement value is compared sequentially with the normal monitoring threshold, the early warning monitoring threshold, and the danger monitoring threshold, respectively. Based on the comparison results, the dose rate level is divided into one of the normal level, the early warning level, and the danger level to obtain the dose rate level judgment result.

[0041] Based on the dose rate level judgment result and the working status parameters of the 8-channel SiPIN detector, a monitoring status identifier is generated, which includes the dose rate level and detector status information.

[0042] In one specific embodiment, the operating status parameters of the 8-channel SiPIN detector include the signal response intensity and noise level of each detector. The dose rate level judgment result is combined with the signal response intensity and the noise level for comprehensive encoding processing to generate a composite monitoring status identifier that includes a dose rate level identifier, a response intensity identifier, and a noise level identifier.

[0043] Specifically, this embodiment establishes a multi-level gamma dose rate threshold hierarchical monitoring system, sequentially compares and analyzes the normal monitoring threshold, early warning monitoring threshold, and dangerous monitoring threshold, and combines the signal response intensity and noise level of the 8-channel SiPIN detectors with the comprehensive coding processing mechanism to generate a composite monitoring status identifier that includes dose rate level identifier, response intensity identifier, and noise level identifier. This enables hierarchical risk assessment of gamma dose rate and real-time monitoring of equipment status, improving the early warning capability and fault diagnosis capability of the radiation monitoring system.

[0044] The present invention also provides a γ dose rate monitoring device based on an 8-channel SiPIN, the device comprising: The feature extraction module is used to acquire the pulse signals output by 8 SiPIN detectors, and input the pulse signals into an adaptive time-domain convolutional neural network that embeds the exponential decay response function of the SiPIN detector to obtain the stacked pulse feature vector. The feature processing module is used to process the stacked pulse feature vector using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. The dose measurement module is used to construct node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and to perform fusion calculation processing on the node features through the graph neural network to obtain the γ dose rate measurement value. The dose monitoring module is used to compare the measured γ dose rate with a preset γ dose rate threshold, generate a corresponding monitoring status identifier based on the comparison result, and obtain the γ dose rate monitoring result.

[0045] Specifically, this embodiment of a γ dose rate monitoring device based on an 8-channel SiPIN detector integrates technologies such as multi-channel parallel acquisition of 8 SiPIN detectors, adaptive temporal convolution processing with embedded physical response functions, decoupling of geometrically constrained hierarchical variational autoencoders, and graph neural network fusion calculation by constructing a feature extraction module, a feature processing module, a dose measurement module, and a dose monitoring module. Based on modular design, it realizes progressively refined processing of pulse signals and collaborative optimization calculation of multi-dimensional parameters. It adopts a distributed processing architecture and a real-time data stream transmission mechanism to ensure efficient data interaction between modules and reasonable allocation of computational load, thus achieving high integration and high real-time performance of the γ dose rate monitoring device.

[0046] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a gamma dose rate monitoring method based on 8-channel SiPIN.

[0047] This invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the γ dose rate monitoring method based on 8-channel SiPIN as described in the embodiments of this invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0048] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring gamma dose rate based on 8-channel SiPIN, characterized in that, Includes the following steps: The pulse signals output from 8 SiPIN detectors are acquired, and the pulse signals are input into an adaptive time-domain convolutional neural network embedded with the exponential decay response function of the SiPIN detector to obtain a stacked pulse feature vector. The stacked pulse feature vector is processed using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. Based on the pulse arrival time, the pulse amplitude, and the confidence parameter, a graph neural network is constructed to generate node features. The graph neural network is then used to perform fusion calculations on the node features to obtain the γ dose rate measurement value. The measured γ dose rate is compared with a preset γ dose rate threshold, and a corresponding monitoring status identifier is generated based on the comparison result to obtain the γ dose rate monitoring result.

2. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 1, characterized in that, The multi-channel synchronization constraint loss function of the hierarchical variational autoencoder with geometric constraints processes the stacked pulse feature vector to obtain pulse arrival time, pulse amplitude, and confidence parameters, including: The geometrically constrained hierarchical variational autoencoder adopts a cross-channel multi-scale latent variable mapping method to construct an independent variational latent space for the feature vectors of each stacked pulse, and introduces a cross-channel set of physical constraint parameters for similar pulses to collaboratively model the correlation characteristics between multiple spatial channels, thereby obtaining pulse arrival time, pulse amplitude, and confidence parameters.

3. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 2, characterized in that, The multipath physical constraint parameter set of the variational potential space includes: The prior distribution of latent variables is dynamically corrected based on the actual crosstalk matrix of the detector, the noise covariance of each amplifier, and the time drift trend. During the training process, the cosine distance, variance difference, and pulse continuity target between the prior distribution of latent variables and the output distribution are jointly optimized by the multi-channel synchronization constraint loss function to obtain the pulse arrival time, pulse amplitude, and confidence parameters.

4. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 1, characterized in that, The process involves acquiring pulse signals output from eight SiPIN detectors, inputting these pulse signals into an adaptive time-domain convolutional neural network embedded with the exponential decay response function of the SiPIN detector, to obtain a stacked pulse feature vector, including: Each output pulse signal of the eight SiPIN detectors is associated with the physical spatial arrangement parameters. An array temporal coupling characteristic model is established based on the spatial arrangement relationship and response function of each SiPIN detector. The eight pulse signals are then jointly modeled based on the array temporal coupling characteristic model to obtain the stacked pulse feature vector embedded with the array spatial structure parameters.

5. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 4, characterized in that, In the array time-domain coupling characteristic model, the coupling correction factors between each SiPIN detector and its corresponding adjacent detector are integrated, and the pulse signal of each detector is dynamically compensated according to the real-time crosstalk correction factor of each detector to adjust and output the stacked pulse feature vector. The stacked pulse feature vector includes spatial position, response characteristics and coupling correction.

6. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 1, characterized in that, The process of constructing node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and then fusing and calculating these node features through the graph neural network to obtain the γ dose rate measurement includes: Node feature encoding is performed on the pulse arrival time, the pulse amplitude, and the confidence parameter to construct the corresponding node feature vector; The node feature vectors are input into a graph neural network, and the γ dose rate measurement value is obtained through information transmission and aggregation calculation between nodes.

7. The γ dose rate monitoring method based on 8-channel SiPIN as described in claim 1, characterized in that, The step of comparing the measured γ dose rate value with a preset γ dose rate threshold, generating a corresponding monitoring status identifier based on the comparison result, and obtaining the γ dose rate monitoring result includes: The measured γ dose rate value is compared with a multi-level preset γ dose rate threshold to obtain the dose rate level judgment result. Based on the dose rate level judgment result and the working status parameters of the 8-channel SiPIN detector, a monitoring status identifier is generated, which includes the dose rate level and detector status information.

8. A γ dose rate monitoring device based on an 8-channel SiPIN, used to perform a γ dose rate monitoring method based on an 8-channel SiPIN as described in any one of claims 1-7, characterized in that, The device includes: The feature extraction module is used to acquire the pulse signals output by 8 SiPIN detectors, and input the pulse signals into an adaptive time-domain convolutional neural network that embeds the exponential decay response function of the SiPIN detector to obtain the stacked pulse feature vector. The feature processing module is used to process the stacked pulse feature vector using a multi-channel synchronous constraint loss function of a geometrically constrained hierarchical variational autoencoder to obtain pulse arrival time, pulse amplitude, and confidence parameters. The dose measurement module is used to construct node features of a graph neural network based on the pulse arrival time, the pulse amplitude, and the confidence parameter, and to perform fusion calculation processing on the node features through the graph neural network to obtain the γ dose rate measurement value. The dose monitoring module is used to compare the measured γ dose rate with a preset γ dose rate threshold, generate a corresponding monitoring status identifier based on the comparison result, and obtain the γ dose rate monitoring result.

9. An electronic device, characterized in that, include: At least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to perform the method as described in any one of claims 1 to 7.

Images