Multi-gamma-source azimuth inversion method, device and equipment and storage medium

Abstract

The invention relates to a multi-gamma-source azimuth inversion method, device and equipment and a storage medium, and the method comprises the steps: constructing a data set through employing a Monte Carlo numerical simulation method; establishing a convolutional neural network for multi-gamma-source inversion; wherein a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network to dynamically adjust the training process of the convolutional neural network, and a center loss function is introduced to reconstruct a loss function of the convolutional neural network to adjust the classification performance and convergence of the training process of the convolutional neural network; training a convolutional neural network based on the data set to obtain a multi-gamma-source orientation inversion model; and collecting a to-be-inverted sample, inputting the to-be-inverted sample into the multi-gamma-source azimuth inversion model, and outputting the quantity and the positions of gamma sources in the to-be-inverted sample based on the multi-gamma-source azimuth inversion model. According to the scheme, dual targets of source number classification and coordinate positioning can be covered, and an innovative solution is provided for real-time dynamic reconstruction of a radiation field.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear radiation field inversion, and in particular to a multi-gamma source azimuth inversion method, device, equipment and storage medium. Background Art

[0002] Radiation field inversion and reconstruction technology plays a crucial role in fields such as nuclear safety and environmental monitoring. It is an essential key technology for ensuring personnel safety and optimizing nuclear facility operations. This technology measures radiation field data by deploying a limited number of detectors within a specific area, thereby visualizing the radiation field environment. This provides strong support for the development of scientific and effective protective measures, thereby reducing personnel exposure and safeguarding their health. Furthermore, radiation field inversion and reconstruction technology can optimize nuclear facility operational procedures, improving operational efficiency and reducing radiation risks by preventing personnel from entering high-radiation areas.

[0003] In the event of a nuclear accident or other emergency, radiation field inversion and reconstruction technology can quickly and accurately assess the radiation environment, rapidly define the scope of the accident, provide a basis for the rapid development of emergency plans, guide effective rescue and cleanup efforts, and mitigate the consequences of the accident. Specifically, radiation field inversion and reconstruction technology uses limited monitoring data, mathematical models, and computational methods to infer the spatial distribution of the radiation field. Through this technology, the number, location, and activity of radioactive sources can be reconstructed, and the spatial distribution of the radiation field can be determined.

[0004] In terms of nuclear facility safety, using a limited number of detectors to measure radiation field data can visualize the radiation field environment, allowing for the development of more scientific and effective protective measures, reducing personnel exposure and safeguarding their health. This data can also be used to optimize nuclear facility operating procedures, improving operational efficiency and reducing radiation risks by preventing personnel from entering high-radiation areas. In the event of a nuclear accident or other emergency, rapid and accurate assessment of the radiation field environment allows for the swift definition of the accident's impact, providing a basis for the rapid development of emergency plans, guiding effective rescue and cleanup efforts, and mitigating the consequences of the accident.

[0005] With the development of artificial intelligence and deep learning technologies, neural network-based methods have shown significant advantages in complex data modeling, pattern recognition, and nonlinear problem solving. Using data collected by a small number of dose detectors to train a neural network can accurately invert the radiation dose field in key areas, improving the efficiency of radiation field assessment. This technological advancement makes it possible to instantly invert and reconstruct the radiation field, providing new scientific tools and methodological support for research and application in related fields. With the continuous advancement of science and technology, radiation field inversion and reconstruction technology is also constantly developing and improving. In the future, this technology will be more intelligent and automated, capable of inverting and reconstructing radiation fields in real time and accurately, providing more efficient and reliable technical support for fields such as nuclear safety. At the same time, this technology will be combined with other technologies to form a more comprehensive safety monitoring and early warning system, providing stronger support for ensuring personnel safety, optimizing nuclear facility operations, and other aspects.

[0006] In summary, radiation field inversion and reconstruction technology has broad application prospects and important practical significance in fields such as nuclear safety. With continued technological development and improvement, it is believed that this technology will play an even more important role in the future, providing stronger support for ensuring personnel safety and optimizing nuclear facility operations. However, radiation field inversion and reconstruction technology is still in its infancy and cannot effectively handle the problem of unknown source numbers or radiation field inversion for radioactive sources with unknown locations. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a multi-gamma source azimuth inversion method, device, equipment and storage medium.

[0008] To achieve the above object, the present invention provides a multi-gamma source azimuth inversion method, comprising the following steps: S1. Monte Carlo numerical simulation method is used to construct the data set; S2. Establishing a convolutional neural network for multi-gamma source inversion; wherein, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network to dynamically adjust the training process of the convolutional neural network, and a center loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; S3. Training the convolutional neural network based on the data set to obtain a multi-γ source azimuth inversion model; S4. Collect samples to be inverted and input them into the multi-γ source azimuthal inversion model, and output the number and positions of γ sources in the samples to be inverted based on the multi-γ source azimuthal inversion model.

[0009] According to one aspect of the present invention, in step S1, the step of constructing a data set using a Monte Carlo numerical simulation method includes: S11. Establish a ground environment model of preset dimensions based on MCNP software; S12. Arrange a detector array in the ground environment model; S13. Randomly arrange 1 to 4 gamma radiation sources in the standard atmospheric medium of the ground environment model; wherein the activity range of the gamma radiation source is set to 1Ci to 10Ci; S14. Using a geometric element card to define the spatial structure of the ground environment model, and using a material element card to configure the properties of different media; S15. Based on the MCNP software, the motion trajectory in the ground environment model and the radiation field data of the photoelectric effect, Compton scattering, and electron pair effect during photon transport are simulated and recorded to generate a data set for training.

[0010] According to one aspect of the present invention, in step S11, in the step of establishing a ground environment model of a preset size based on MCNP software, the ground environment model adopts a 10m×10m cement ground environment model; In step S12, in the step of arranging a detector array in the ground environment model, the detector array is arranged in a uniform coverage manner.

[0011] According to one aspect of the present invention, in step S2, in the step of establishing a convolutional neural network for multi-gamma source inversion, the convolutional neural network includes: an input layer, a feature extraction module connected to the input layer, a classification decision module connected to the feature extraction module, and an output layer connected to the classification decision module; The feature extraction module includes: several cascaded convolutional layers and maximum pooling layers; Each of the convolutional layers is followed by a ReLU activation function for extracting the spatial features of the radiation field intensity distribution of the samples in the dataset layer by layer, and gradually compressing the spatial dimensions through the maximum pooling layer; The classification decision module includes: a Flatten layer and multiple fully connected layers; The Flatten layer is connected to the last maximum pooling layer of the feature extraction module and is used to expand the features output by the feature extraction module; The fully connected layer is provided with three layers, and the first fully connected layer is connected to the Flatten layer, and the third fully connected layer is connected to the output layer; The fully connected layer is configured with a tanh activation function to output the probability of the number of γ sources, and a Softmax function is configured to output the probability of the classification of the γ source coordinates.

[0012] According to one aspect of the present invention, in step S2, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network, and in the step of dynamically adjusting the training process of the convolutional neural network, the learning rate gradually decays from the initial value to a preset minimum value according to a cosine curve; wherein the learning rate is expressed as: ; in, represents the learning rate, represents the lower limit of the learning rate, represents the upper limit of the learning rate, Indicates the current training step number, Indicates the total number of steps in the training cycle.

[0013] According to one aspect of the present invention, in step S2, in the step of establishing a convolutional neural network for multi-gamma source inversion, the reconstructed loss function is a composite loss function that integrates a cross entropy loss function and a center loss function; The composite loss function is expressed as: ; ; ; in, represents the composite loss function, represents the cross entropy loss function, represents the center loss function, Represents the weight coefficient that balances the two loss functions, Represents the classification label of the coordinates, Indicates the number of classification categories, which includes 1 to 4 gamma radiation sources and gamma radiation source coordinates, Represents the probability values ​​corresponding to different classification labels predicted by the trained convolutional neural network, Represents the different classification labels predicted by the trained convolutional neural network, Represents the sample data in the input dataset, represents the center value of the label, Indicates the total number of labels for the classification category, subscript Indicates the order of label classification.

[0014] To achieve the above-mentioned object, the present invention provides a multi-gamma source azimuth inversion device, comprising: A data set generation module is used to construct a data set using the Monte Carlo numerical simulation method; A convolutional neural network building module is used to establish a convolutional neural network for multi-gamma source inversion; wherein a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network to dynamically adjust the training process of the convolutional neural network, and a center loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; A training module, which trains the convolutional neural network based on the data set to obtain a multi-gamma source azimuth inversion model; The identification module is used to load the obtained multi-γ source azimuthal inversion model to collect samples to be inverted and input the multi-γ source azimuthal inversion model, and output the number and position of γ sources in the samples to be inverted based on the multi-γ source azimuthal inversion model.

[0015] To achieve the above object of the invention, the present invention provides a device comprising at least one processor, at least one memory and a data bus; The processor and the memory communicate with each other via the data bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the aforementioned multi-γ source azimuth inversion method.

[0016] To achieve the above-mentioned object of the invention, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program implements the above-mentioned multi-γ source azimuth inversion method when executed by a processor.

[0017] According to one solution of the present invention, a systematic approach based on Monte Carlo simulation and convolutional neural networks is proposed to address the challenge of identifying the number and coordinates of sources in multi-source radiation field inversion. This approach successfully solves the challenge of simultaneously determining the number of unknown radioactive sources and their azimuthal coordinates in radiation field inversion. By optimizing the detector array layout, introducing a cosine annealing strategy, and employing a composite loss function, the robustness of the resulting multi-gamma source azimuthal inversion model is improved in complex scenarios.

[0018] According to a solution of the present invention, this solution can cover the dual goals of source quantity classification and coordinate positioning, and provides an innovative solution for real-time dynamic reconstruction of the radiation field.

[0019] According to a solution of the present invention, in the identification of the number of radioactive sources, the recognition accuracy of 2 to 4 gamma radiation sources reaches 89.25%; in coordinate inversion, the innovative strategy of joint calculation of composite loss function is adopted, which can make the single-source positioning accuracy exceed 98% (for example, the x-coordinate and y-coordinate positioning accuracy are 99.2% and 98.9% respectively), the dual-source positioning accuracy exceeds 81% (for example, the x-coordinate and y-coordinate positioning accuracy of the first source are 83.55% and 85.25% respectively, and the x-coordinate and y-coordinate positioning accuracy of the second source are 81.8% and 81.85% respectively), and the three-source positioning accuracy exceeds 50% (for example, the x-coordinate and y-coordinate positioning accuracy of the first source are 83.55% and 85.25% respectively, and the x-coordinate and y-coordinate positioning accuracy of the second source are 81.8% and 81.85% respectively). The accuracy rates are 76.13% and 75.97% respectively, the x-coordinate and y-coordinate positioning accuracy rates of the second source are 50.4% and 50.13% respectively, and the x-coordinate and y-coordinate positioning accuracy rates of the third source are 72.57% and 71.93% respectively). The four-source positioning accuracy exceeds 41% (the x-coordinate and y-coordinate positioning accuracy rates of the first source are 68.78% and 69.75% respectively, the x-coordinate and y-coordinate positioning accuracy rates of the second source are 42.2% and 41.98% respectively, the x-coordinate and y-coordinate positioning accuracy rates of the third source are 41.93% and 42.15% respectively, and the x-coordinate and y-coordinate positioning accuracy rates of the fourth source are 67.68% and 66.85% respectively).

[0020] According to one solution of the present invention, a cosine annealing scheduler is used to dynamically adjust the learning rate. This strategy leverages the cosine annealing scheduler's periodic restart mechanism, allowing the convolutional neural network in this solution to escape local minima when approaching the optimal solution, enhancing parameter fine-tuning capabilities. In the task of radiation source identification, the introduction of a cosine annealing scheduler (with a period of 100 epochs) significantly increased the rate of decrease in the loss value during convolutional neural network training, exhibiting fluctuation characteristics synchronized with the learning rate period. In particular, the introduction of the cosine annealing scheduler to adjust the learning rate effectively alleviates the local optimality problem caused by the coupling of multi-source features, significantly improving the performance of multi-source identification.

[0021] According to a solution of the present invention, this solution achieves high-precision inversion of the number and spatial position of unknown radioactive sources, provides a new development idea for radiation field inversion, and has important application value for improving nuclear emergency response capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A diagram showing the steps of a multi-gamma source azimuth inversion method according to an embodiment of the present invention; Figure 2 Schematic diagram of a gate cell card according to an embodiment of the present invention, wherein: Figure 2 (a) is a schematic diagram of the geometric grid card. Figure 2 (b) Schematic diagram of the material grid cell card; Figure 3 A structural diagram of a convolutional neural network according to an embodiment of the present invention; Figure 4 Schematic diagram of single-source radiation field measurement in an example of one embodiment of the present invention; Figure 5 This is a comparison diagram of the experimental measurement results of a single-source radiation field in an example of an embodiment of the present invention and this solution; Figure 6 Schematic diagram of multi-source radiation field measurement in an example of an embodiment of the present invention; Figure 7 This is a comparison diagram of the experimental measurement results of the multi-source radiation field in an example of an embodiment of the present invention and this solution. DETAILED DESCRIPTION

[0023] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments cannot be described one by one here, but the embodiments of the present invention are not limited to the following embodiments.

[0024] like Figure 1 As shown, according to one embodiment of the present invention, a multi-gamma source azimuth inversion method of the present invention includes the following steps: S1. Monte Carlo numerical simulation method is used to construct the data set; S2. Establishing a convolutional neural network for multi-gamma source inversion; wherein, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network for dynamically adjusting the training process of the convolutional neural network, and a center loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; S3. Train a convolutional neural network based on the dataset to obtain a multi-γ source azimuth inversion model; S4. Collect samples to be inverted and input them into a multi-γ source azimuthal inversion model, and output the number and positions of γ sources in the samples to be inverted based on the multi-γ source azimuthal inversion model.

[0025] like Figure 1 As shown, according to one embodiment of the present invention, in step S1, the step of constructing a data set using a Monte Carlo numerical simulation method includes: S11. Establish a ground environment model of a preset size using MCNP software. MCNP software is a universal particle transport simulation platform capable of handling neutron, photon, and electron transport problems within complex three-dimensional geometries. Therefore, constructing a dataset using MCNP software generates a large amount of high-precision simulation data, thereby addressing the high cost and time required for data acquisition. Furthermore, in the step of establishing a ground environment model of a preset size using MCNP software, a 10m x 10m concrete floor model is used.

[0026] S12. Arrange the detector array in the ground environment model; in this embodiment, the detector array is arranged in a uniform coverage manner in the ground environment model. For example, if 10,000 detectors are evenly placed in the ground environment model (XY plane), the detectors are arranged at intervals of 0.1m×0.1m. Of course, if the number of detectors is limited by cost and placement environment conditions, the number of detectors can be reduced to 2500 and the detectors are arranged at intervals of 0.2m×0.2m, 1600 and the detectors are arranged at intervals of 0.25m×0.25m, or 100 and the detectors are arranged at intervals of 1m×1m. In this way, different detection accuracies can be achieved through different detector arrays, and the data accuracy of the data set can be flexibly adjusted. In this embodiment, since the uniform distribution of the detectors makes the intervals between them fixed, the layout area can be divided into a matching grid based on the layout of the detectors, thereby matching the detection results of the detectors to the corresponding grid coordinates.

[0027] S13. Randomly arrange 1 to 4 gamma radiation sources in a standard atmospheric medium of a ground environment model; wherein the activity range of the gamma radiation sources is set to 1Ci to 10Ci; and the emission direction / position of the gamma radiation sources is isotropic.

[0028] S14. Use the geometry grid element card to define the spatial structure of the ground environment model, and use the material grid element card to configure the properties of different media; such as Figure 2 As shown in (a), the geometry grid card records the size and material allocation of the three-dimensional area division in the ground environment model. For example, the size is set to 10m×10m, the ground material is configured as cement, and the environment is configured as air. Figure 2 As shown in (b), the material grid card records the element composition, density and nuclide cross section library in the MCNP DATA data set using MCNP software.

[0029] S15. Using MCNP software, simulate and record the motion trajectory within the ground environment model and the radiation field data from the photoelectric effect, Compton scattering, and electron pair effect during photon transport to generate a training dataset. The resulting dataset also records the corresponding absorbed dose rate. In this embodiment, a dataset consisting of 210,000 sets of three-dimensional radiation field data can be constructed for subsequent convolutional neural network training. Furthermore, 200,000 sets of three-dimensional radiation field data are selected from the constructed dataset as training samples, and 10,000 sets of three-dimensional radiation field data are selected as test samples.

[0030] In this embodiment, the obtained three-dimensional radiation field data is structured to facilitate the training of the convolutional neural network. Specifically, the absorbed dose rate in the data set is normalized to construct a two-dimensional dose matrix, and the gamma radiation source parameters are encoded as a label vector [x1, y1, a1, ..., x n ,y n ,a n ]. Through the grid coordinate mapping strategy, the continuous coordinate regression problem is transformed into a discrete classification problem, realizing the unification of the source quantity prediction and coordinate positioning algorithms. Therefore, all data are serialized into a "dataset.pth" binary file for the input of training convolutional neural networks. Among them, the FloatTensor of the PyTorch framework is used to store the dose matrix and the IntTensor is used to store the label vector.

[0031] like Figure 3As shown, according to one embodiment of the present invention, in step S2, in the step of establishing a convolutional neural network for multi-gamma source inversion, the convolutional neural network includes: an input layer, a feature extraction module connected to the input layer, a classification decision module connected to the feature extraction module, and an output layer connected to the classification decision module. In this embodiment, the input layer is used to receive samples, and its size can be set to 1×10×10, indicating that the input is a single-channel two-dimensional image or matrix of size 10×10. Furthermore, the feature extraction module includes: a plurality of cascaded convolutional layers and maximum pooling layers. In this embodiment, there are two convolutional layers, and the maximum pooling layers are arranged in a one-to-one correspondence with the convolutional layers. Among them, the size of the features that can be output by the first convolution layer and the first maximum pooling layer becomes 32×11×11. Specifically, the input dimension is 1×10×10, a 2×2 convolution kernel is used, the step size is 1, and a layer of white edges is added, then the output dimension becomes 32×11×11; the size of the features that can be output by the second convolution layer and the first maximum pooling layer becomes 64×12×12. Specifically, the input dimension is 32×11×11, a 2×2 convolution kernel is used, the step size is 1, and a layer of white edges is added, then the output dimension becomes 64×12×12. In this embodiment, each convolution layer is followed by a ReLU activation function to extract the spatial features of the radiation field intensity distribution of the samples in the data set layer by layer, and gradually compress the spatial dimensions through the maximum pooling layer.

[0032] Furthermore, the classification decision module includes: a Flatten layer and multiple fully connected layers; wherein the Flatten layer is connected to the last maximum pooling layer of the feature extraction module, and is used to expand the features output by the feature extraction module; wherein the Flatten layer uses the feature expansion output by the view function; the fully connected layer is provided with three layers, and the first fully connected layer is connected to the Flatten layer, and the third fully connected layer is connected to the output layer; in this embodiment, the first fully connected layer is used to flatten the features output by the feature extraction module into a 1×256 vector, specifically, its input dimension is 1×9216, using a fully connected function, and using tanh The activation function and dropout regularization are packaged to achieve an output dimension of 1×256. The second fully connected layer is used to reduce the dimension of the output of the first fully connected layer to 1×64. Specifically, its input dimension is 1×256, and the fully connected function is used. The softplus activation function and dropout regularization are used to package it, that is, the output dimension is 1×64. The third fully connected layer is used to reduce the dimension of the output of the second fully connected layer to 1×4 or 1×10 (depending on different scenario settings). Specifically, its input dimension is 1×64, and the fully connected function is used, and the output dimension is 1×4 or 1×10, obtaining different results and corresponding probability values. In this embodiment, the fully connected layer is configured with a tanh activation function to output the probability of the number of gamma sources, and a softmax function is configured to output the probability of the gamma source coordinate classification.

[0033] According to one embodiment of the present invention, the learning rate is a core hyperparameter for training a convolutional neural network. If the learning rate is too small, the training time will be long and it will be easy to fall into the dilemma of a local optimal solution. If the learning rate is too large, the learning parameters will change dramatically, making it difficult to reach the optimal solution. To this end, this solution introduces a cosine annealing scheduler, which optimizes the training process by dynamically adjusting the strategy, and verifies the generalization ability of the convolutional neural network of this solution in combination with Monte Carlo simulation data. Therefore, in step S2, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network. In the step of dynamically adjusting the training process of the convolutional neural network, the learning rate gradually decays from the initial value to the preset minimum value according to the cosine curve; wherein the learning rate is expressed as: ; in, represents the learning rate, represents the lower limit of the learning rate, represents the upper limit of the learning rate, Indicates the current training step number, Indicates the total number of steps in the training cycle. In this embodiment, the total number of steps in the training cycle Can be set to 100 epochs. In addition, the learning rate lower limit Can be set to 0, upper limit of learning rate Can be set to 0.001.

[0034] Through the above settings, a cosine annealing scheduler is used to dynamically adjust the learning rate. This strategy fully utilizes the cosine annealing scheduler's periodic restart mechanism, allowing the convolutional neural network in this scheme to escape local minima when approaching the optimal solution, enhancing parameter fine-tuning capabilities. In the task of radiation source identification, the introduction of a cosine annealing scheduler (period of 100 epochs) significantly increases the rate of decrease in the loss value during convolutional neural network training, and exhibits fluctuation characteristics synchronized with the learning rate cycle. In particular, the introduction of the cosine annealing scheduler to adjust the learning rate effectively alleviates the local optimality problem caused by the coupling of multi-source features, significantly improving the performance of multi-source identification.

[0035] According to one embodiment of the present invention, in step S2, in the step of establishing a convolutional neural network for multi-gamma source inversion, the reconstructed loss function is a composite loss function that integrates the cross entropy loss function and the center loss function; in this embodiment, the composite loss function is expressed as: ; ; ; in, represents the composite loss function, represents the cross entropy loss function, represents the center loss function, Represents the weight coefficient for balancing the two loss functions, which can be set to 0.1. Indicates the classification label of the coordinate, with a value from 1 to 10. Indicates the number of classification categories, which includes 1 to 4 gamma radiation sources and gamma radiation source coordinates, Represents the probability values ​​corresponding to different classification labels predicted by the trained convolutional neural network, Represents the different classification labels predicted by the trained convolutional neural network, Represents the sample data in the input dataset, Indicates the center value of the label, with values ​​of 0.5, 1.5, 2.5...9.5. Indicates the total number of labels for the classification category, and it can be set to 10 categories, subscript Indicates the order of label classification, with a value of 1 to 10.

[0036] According to one embodiment of the present invention, the present invention provides a multi-gamma source azimuthal inversion device, comprising: a data set generation module, a convolutional neural network construction module, a training module and an identification module; wherein the data set generation module is used to construct a data set using a Monte Carlo numerical simulation method; the convolutional neural network construction module is used to establish a convolutional neural network for multi-gamma source inversion; wherein a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network, which is used to dynamically adjust the training process of the convolutional neural network, and a central loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; the training module trains the convolutional neural network based on the data set to obtain a multi-gamma source azimuthal inversion model; the identification module is used to load the obtained multi-gamma source azimuthal inversion model to collect samples to be inverted and input them into the multi-gamma source azimuthal inversion model, and output the number and position of gamma sources in the samples to be inverted based on the multi-gamma source azimuthal inversion model.

[0037] The specific limitations of the multi-gamma source azimuth inversion device can be found in the limitations of the vector data extraction method based on high-fidelity scenes above and will not be repeated here. Each module in the multi-gamma source azimuth inversion device described above can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the modules can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0038] According to one embodiment of the present invention, a device is provided, comprising at least one processor, at least one memory, and a data bus. In this embodiment, the processor and the memory communicate with each other via the data bus; the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute the aforementioned multi-gamma source azimuthal inversion method.

[0039] In this embodiment, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0040] In this embodiment, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0041] In this embodiment, the device may also be provided with a display screen and an input device, wherein the display screen may be a liquid crystal display screen or an electronic ink display screen, and the input device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the outer shell, or an external keyboard, touchpad or mouse, etc.

[0042] According to one embodiment of the present invention, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program implements the aforementioned multi-γ source azimuth inversion method when executed by a processor.

[0043] To further illustrate this solution, further examples are given to illustrate it.

[0044] Example In this embodiment, a radiation field detection experiment was conducted in the laboratory to verify that the radiation field data inverted using this solution and the error range in the real environment were within an acceptable range. The experimental steps involved measuring the dose rate under different scenarios in the laboratory, then using this solution's multi-gamma source azimuth inversion method to simulate the results, and drawing conclusions by comparing the radiation field dose rates.

[0045] a) Single source radiation field verification Laboratory experiments were conducted in a shielded laboratory (background dose rate 0.10 μSv / h) using a RadEye G-10 detector to measure the activity of a Cs-137 source. ) dose rate distribution. The Cs-137 radiation source and the detector are spaced 20 cm to 500 cm apart along a straight line (see Figure 4 ), record the average and peak dose rates (in μSv / h) at each distance point. In this embodiment, a RadEye G-10 detector is used to measure the Cs-137 radiation source (activity ) in the process of dose rate distribution, each measurement time is 5 minutes and each interval distance is 20 cm.

[0046] This scheme: MCNP software is used to construct a single source model, and the photon energy is set to 0.661MeV (corresponding to decay spectrum), The particle count is calculated and converted into single particle dose rate, and the statistical error is controlled within 1%.

[0047] like Figure 5 As shown, the comparison between the inversion results based on this scheme and the laboratory experimental data shows that: at medium and short distances (20-300 cm): the error is less than 10%, and the inversion results of this scheme are highly consistent with the laboratory experimental data; at long distances (450 cm): the dose rate is as low as At the same level, the maximum error caused by environmental background interference is 50%, which is optimized to within 20% after background noise filtering.

[0048] b) Multi-source radiation field verification Laboratory experiment: High / low activity Cs-137 sources were arranged in a 250 cm×250 cm area ( and ), see Figure 6 , using RadEye G-10 detector to measure Cs-137 radioactive source (activity ) dose rate distribution, each measurement lasted 3 minutes. Furthermore, Latin Hypercube Sampling (LHS) was used to address the sampling challenge of the six variables (source and detector coordinates). Each coordinate variable was divided into 10 layers (25 cm per layer). Ten sets of samples were drawn to evenly cover the parameter space. The sampling results are shown in Table 1. In Table 1, S1_x and S1_y represent the x and y coordinates of the first source, in cm; S2_x and S2_y represent the x and y coordinates of the second source, in cm; D_x and D_y represent the x and y coordinates of the detector, in cm. Stratification represents the number of strata in the aforementioned Latin Hypercube Sampling, ranging from 1 to 10, and is unitless.

[0049] Table 1 Latin hypercube sampling results

[0050] This scheme: MCNP software is used to construct a dual-source model, and the photon energy is set to 0.661MeV (corresponding to decay spectrum), maintain The particle count is calculated and converted into single particle dose rate, and the statistical error is controlled within 1%.

[0051] like Figure 7As shown in the figure, the inversion results based on this scheme and the 10 sets of experimental results of laboratory experimental data are summarized to show their consistency: the relative error of 90% of the data points is ≤10%; in high fluctuation scenarios (such as experiments 3 / 4 / 9): when the peak dose rate is 1.0-1.2 μSv / h, the error is still stable. ; Therefore, comparative experimental results show that: the error of this scheme is less than 10% in the range of 20-300cm, and is optimized to 20% at long distances through noise filtering; LHS sampling achieves full coverage of the parameter space with 10 groups of experiments, and 90% of the data errors are ≤10%, meeting the requirements of complex scenarios, proving that this scheme has engineering practicality in both single-source and multi-source scenarios.

[0052] The above contents are merely examples of specific solutions of the present invention. For devices and structures not described in detail, it should be understood that they can be implemented by adopting general devices and methods available in the art.

[0053] The above description is merely one embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A multi-gamma source azimuth inversion method, characterized in that: The following steps are involved: S1. Monte Carlo numerical simulation method is used to construct the data set; S2. Establishing a convolutional neural network for multi-gamma source inversion; wherein, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network to dynamically adjust the training process of the convolutional neural network, and a center loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; S3. Training the convolutional neural network based on the data set to obtain a multi-γ source azimuth inversion model; S4. Collect samples to be inverted and input them into the multi-γ source azimuthal inversion model, and output the number and position of γ sources in the samples to be inverted based on the multi-γ source azimuthal inversion model.

2. The multi-gamma source azimuth inversion method according to claim 1, characterized in that: In step S1, the step of constructing a data set using the Monte Carlo numerical simulation method includes: S11. Establish a ground environment model of preset dimensions based on MCNP software; S12. Arrange a detector array in the ground environment model; S13. Randomly arrange 1 to 4 gamma radiation sources in the standard atmospheric medium of the ground environment model; wherein the activity range of the gamma radiation source is set to 1Ci to 10Ci; S14. Using a geometric element card to define the spatial structure of the ground environment model, and using a material element card to configure the properties of different media; S15. Based on the MCNP software, the motion trajectory in the ground environment model and the radiation field data of the photoelectric effect, Compton scattering, and electron pair effect during photon transport are simulated and recorded to generate a data set for training.

3. The multi-gamma source azimuth inversion method according to claim 2, characterized in that: In step S11, in the step of establishing a ground environment model of a preset size based on MCNP software, the ground environment model adopts a 10m×10m cement ground environment model; In step S12, in the step of arranging a detector array in the ground environment model, the detector array is arranged in a uniform coverage manner.

4. The multi-gamma source azimuth inversion method according to claim 3, characterized in that: In step S2, in the step of establishing a convolutional neural network for multi-gamma source inversion, the convolutional neural network includes: an input layer, a feature extraction module connected to the input layer, a classification decision module connected to the feature extraction module, and an output layer connected to the classification decision module; The feature extraction module includes: several cascaded convolutional layers and maximum pooling layers; Each of the convolutional layers is followed by a ReLU activation function for extracting the spatial features of the radiation field intensity distribution of the samples in the dataset layer by layer, and gradually compressing the spatial dimensions through the maximum pooling layer; The classification decision module includes: a Flatten layer and multiple fully connected layers; The Flatten layer is connected to the last maximum pooling layer of the feature extraction module and is used to expand the features output by the feature extraction module; The fully connected layer is provided with three layers, and the first fully connected layer is connected to the Flatten layer, and the third fully connected layer is connected to the output layer; The fully connected layer is configured with a tanh activation function to output the probability of the number of γ sources, and a Softmax function is configured to output the probability of the classification of the γ source coordinates.

5. The multi-gamma source azimuth inversion method according to claim 4, characterized in that: In step S2, a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network. In the step of dynamically adjusting the training process of the convolutional neural network, the learning rate gradually decays from the initial value to a preset minimum value according to a cosine curve; wherein the learning rate is expressed as: ； in, represents the learning rate, represents the lower limit of the learning rate, represents the upper limit of the learning rate, Indicates the current training step number, Indicates the total number of steps in the training cycle.

6. The multi-gamma source azimuth inversion method according to claim 5, characterized in that: In step S2, in the step of establishing a convolutional neural network for multi-γ source inversion, the reconstructed loss function is a composite loss function that integrates the cross entropy loss function and the center loss function; The composite loss function is expressed as: ； ； ； in, represents the composite loss function, represents the cross entropy loss function, represents the center loss function, Represents the weight coefficient that balances the two loss functions, Represents the classification label of the coordinates, Indicates the number of classification categories, which includes 1 to 4 gamma radiation sources and gamma radiation source coordinates, Represents the probability values ​​corresponding to different classification labels predicted by the trained convolutional neural network, Represents the different classification labels predicted by the trained convolutional neural network, Represents the sample data in the input dataset, represents the center value of the label, Indicates the total number of labels for the classification category, subscript Indicates the order of label classification.

7. A multi-gamma source azimuth inversion device, characterized in that: include: A data set generation module is used to construct a data set using the Monte Carlo numerical simulation method; A convolutional neural network building module is used to establish a convolutional neural network for multi-gamma source inversion; wherein a cosine annealing scheduler is introduced to adjust the learning rate of the convolutional neural network to dynamically adjust the training process of the convolutional neural network, and a center loss function is introduced to reconstruct the loss function of the convolutional neural network to adjust the classification performance and convergence of the convolutional neural network training process; A training module, which trains the convolutional neural network based on the data set to obtain a multi-gamma source azimuth inversion model; The identification module is used to load the obtained multi-γ source azimuthal inversion model to collect samples to be inverted and input the multi-γ source azimuthal inversion model, and output the number and position of γ sources in the samples to be inverted based on the multi-γ source azimuthal inversion model.

8. A device, characterized in that comprising at least one processor, at least one memory and a data bus; The processor and the memory communicate with each other via the data bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the multi-gamma source azimuth inversion method according to any one of claims 1 to 6.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-gamma source azimuth inversion method according to any one of claims 1 to 6 is implemented.

Images