Nucleus identification method, electronic device, and storage medium

Abstract

The application provides a nuclide identification method, an electronic device and a storage medium. The nuclide identification method comprises: obtaining a to-be-identified nuclide spectrum diagram; the to-be-identified nuclide spectrum diagram is a single nuclide spectrum diagram or a mixed nuclide spectrum diagram; performing nuclide identification on the to-be-identified nuclide spectrum diagram by using a target nuclide identification model to obtain a corresponding target identification result; the target nuclide identification model is obtained based on model optimization training on a to-be-trained nuclide identification model; the to-be-trained nuclide identification model is obtained by adding an attention mechanism module in a convolutional neural network model; the attention mechanism module is used for spatial and channel weighting processing on a feature map output by a convolution layer of the convolutional neural network model; and the target identification result is used for representing a nuclide corresponding to the to-be-identified nuclide spectrum diagram. The technical solution of the application can efficiently and accurately identify the nuclide of the to-be-identified nuclide spectrum diagram, thereby efficiently and accurately identifying the nuclide corresponding to the to-be-identified nuclide spectrum diagram.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to methods for identifying nuclides, electronic devices, and storage media. Background Technology

[0002] In fields such as nuclear safety, environmental monitoring, and medical diagnosis, radiation detection is often performed on the radioactive source first to obtain the energy spectrum of the nuclide to be identified. Then, the nuclide is identified by performing nuclide identification on the energy spectrum of the nuclide to be identified, thereby determining the nuclide corresponding to the energy spectrum of the nuclide to be identified, and thus determining the nuclide present in the radioactive source.

[0003] When identifying a nuclide from its energy spectrum, if the spectrum represents a single nuclide, the nuclide can usually be determined by analyzing its characteristic peaks. However, if the spectrum represents a mixture of nuclides, the peaks of multiple nuclides may overlap, causing the characteristic peaks of different nuclides to intertwine in the spectrum. In this case, it is often difficult to determine the nuclide from the spectrum by analyzing the characteristic peaks alone.

[0004] However, in practical applications, radioactive sources may contain a single nuclide or a mixture of nuclides. That is, in practical applications, the energy spectrum of the nuclide to be identified may be a single nuclide spectrum or a spectrum of the nuclide to be identified. Therefore, in order to perform efficient and accurate nuclide identification on the energy spectrum of the nuclide to be identified, regardless of whether it is a single nuclide spectrum or a spectrum of mixed nuclides, a nuclide identification method is needed that can perform efficient and accurate nuclide identification on both single and mixed nuclide spectra. Summary of the Invention

[0005] This application provides a nuclide identification method, an electronic device, and a storage medium to solve one or more of the above-mentioned technical problems.

[0006] In a first aspect, embodiments of this application provide a method for identifying nuclides, including:

[0007] Obtain the energy spectrum of the nuclide to be identified; the energy spectrum of the nuclide to be identified may be a single nuclide energy spectrum or a mixed nuclide energy spectrum.

[0008] The target nuclide identification model is used to identify the nuclide in the energy spectrum of the nuclide to be identified, and the corresponding target identification result is obtained. The target nuclide identification model is obtained by optimizing and training the nuclide identification model to be trained. The nuclide identification model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model. The attention mechanism module is used to perform spatial and channel weighting on the feature map output by the convolutional layer of the convolutional neural network model. The target identification result is used to represent the nuclide corresponding to the energy spectrum of the nuclide to be identified.

[0009] Optionally, the attention mechanism module can be added after the convolutional layer.

[0010] Optionally, the steps for optimizing the nuclide recognition model to obtain the target nuclide recognition model are as follows:

[0011] Obtain training and testing sets; the training and testing sets each include multiple sample nuclide energy spectra and the corresponding annotation results for each sample nuclide energy spectrum; the multiple sample nuclide energy spectra include single nuclide energy spectra and mixed nuclide energy spectra;

[0012] The training set is used to iterate the nuclide recognition model under training for multiple rounds until the set iteration stopping condition is met.

[0013] When the iteration stopping condition is met, the nuclide recognition model to be trained is tested using the test set, and if the test results meet the preset standards, the nuclide recognition model to be trained is used as the target nuclide recognition model.

[0014] Optionally, the steps for each round of iterative training are as follows:

[0015] Select the nuclide energy spectrum of the target sample in the training set;

[0016] The target sample nuclide energy spectrum is input into the nuclide recognition model to be trained to obtain the corresponding sample recognition result;

[0017] Calculate the target loss value between the sample recognition result and the nuclide annotation result corresponding to the target sample image;

[0018] The internal model parameters in the nuclide recognition model to be trained are updated using the target loss value.

[0019] Optionally, the test results include the accuracy of the nuclide recognition model to be trained.

[0020] Optionally, the energy spectra of sample nuclides that are mixed energy spectra can be obtained through the following steps:

[0021] Obtain the energy spectra of individual nuclides for various nuclides;

[0022] In the single-nuclide energy spectrum diagrams corresponding to multiple nuclides, at least two single-nuclide energy spectrum diagrams corresponding to different nuclides are randomly selected.

[0023] The energy spectra of individual nuclides corresponding to at least two nuclides are mixed to obtain mixed nuclide energy spectra, and the mixed nuclide energy spectra are used as sample nuclide energy spectra.

[0024] Optionally, the energy spectrum of a single nuclide can be a real energy spectrum of a nuclide obtained through radiation detection, or a simulated energy spectrum of a nuclide generated through computer simulation.

[0025] Optionally, the attention mechanism module is a convolutional block attention module.

[0026] Secondly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor implements the method provided in any embodiment of this application when executing the computer program.

[0027] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method provided in any embodiment of this application.

[0028] Compared with the prior art, this application has the following advantages:

[0029] The technical solution of this application is that the target nuclide recognition model is obtained by optimizing the nuclide recognition model to be trained. The nuclide recognition model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model. The attention mechanism module is used to perform spatial and channel weighted processing on the feature map output by the convolutional layer of the convolutional neural network model.

[0030] By applying spatial weighting to the feature map, the target nuclide identification model can highlight key regions in the energy spectrum of the target nuclide while suppressing irrelevant or redundant regions. This allows the target nuclide identification model to more accurately extract spatial features relevant to nuclide identification, enhances its ability to focus on regions crucial for nuclide identification, and reduces interference from background noise and irrelevant information, thereby enabling the model to better identify nuclide features.

[0031] Meanwhile, by applying channel weighting to the feature map, the target nuclide identification model can highlight the most contributing feature channels while suppressing irrelevant or redundant channels. This allows the target nuclide identification model to more accurately identify key channel features in the nuclide energy spectrum, thereby improving the effectiveness and accuracy of nuclide identification.

[0032] Therefore, by performing spatial weighting and channel weighting on the feature map, the target nuclide identification model can more accurately identify the nuclide corresponding to the energy spectrum of the nuclide to be identified, thereby more accurately identifying the nuclide corresponding to the energy spectrum of the nuclide to be identified.

[0033] Furthermore, the technical solution of this application can identify nuclides in both single-nuclide and mixed-nuclide energy spectrum patterns, thereby identifying the nuclide corresponding to the energy spectrum pattern.

[0034] Furthermore, since the target nuclide identification model is used to identify the nuclide energy spectrum to obtain the corresponding target identification result, thereby identifying the nuclide corresponding to the nuclide energy spectrum, the technical solution of this application can identify the nuclide energy spectrum in a relatively efficient manner, thereby identifying the nuclide corresponding to the nuclide energy spectrum in a relatively efficient manner.

[0035] Therefore, the technical solution of this application can efficiently and accurately identify nuclides in the energy spectrum of the nuclide to be identified, whether it is a single nuclide energy spectrum or a mixed nuclide energy spectrum, thereby efficiently and accurately identifying the nuclide corresponding to the energy spectrum of the nuclide to be identified.

[0036] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of this application will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0037] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0038] Figure 1 A flowchart of a nuclide identification method provided in an embodiment of this application is shown;

[0039] Figure 2 A schematic diagram of a nuclide identification device provided in an embodiment of this application is shown; and

[0040] Figure 3 A block diagram of an electronic device used to implement embodiments of this application is shown. Detailed Implementation

[0041] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other forms than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0042] Figure 1 A flowchart of a nuclide identification method 100 provided in an embodiment of this application is shown. The method may include steps S101-S102.

[0043] In step S101, the energy spectrum of the nuclide to be identified is obtained; the energy spectrum of the nuclide to be identified is either a single nuclide energy spectrum or a mixed nuclide energy spectrum.

[0044] In step S102, the target nuclide recognition model is used to identify the nuclide energy spectrum to obtain the corresponding target recognition result; the target nuclide recognition model is obtained by optimizing and training the nuclide recognition model to be trained; the nuclide recognition model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model; the attention mechanism module is used to perform spatial and channel weighting processing on the feature map output by the convolutional layer of the convolutional neural network model; the target recognition result is used to represent the nuclide corresponding to the nuclide energy spectrum to be identified.

[0045] In the embodiments of this application, the nuclide energy spectrum Figure 1 Generally, an energy spectrum is a graph showing the relationship between energy and count obtained by detecting radiation from a radioactive source using a detector. Alternatively, it can be a graph generated through computer simulation, such as an energy spectrum generated using the Monte Carlo simulation method. The nuclide energy spectrum reflects the radiation characteristics of the radioactive nuclides present in the radioactive source.

[0046] In this embodiment of the application, the nuclide energy spectrum obtained by radiation detection is called the real nuclide energy spectrum, and the nuclide energy spectrum generated by computer simulation is called the simulated nuclide energy spectrum.

[0047] In this application, a radioactive source generally refers to a substance capable of emitting gamma rays. Radioactive sources are typically composed of naturally occurring or artificially produced radioactive nuclides, which release specific types of radiation (such as gamma rays) during their decay process. Common radioactive sources include cobalt-60 (Co-60) and cesium-137 (Cs-137).

[0048] In this embodiment of the application, a single-nuctopic energy spectrum refers to an energy spectrum containing the radiation characteristics of only one nuclide. A mixed-nuctopic energy spectrum refers to an energy spectrum composed of the superimposed radiation characteristics of two or more different nuclides, or an energy spectrum composed of the superimposed energy spectra of two or more single-nuctopic nuclides.

[0049] Generally, a Convolutional Neural Network (CNN) typically consists of an input layer, convolutional layers, activation layers, pooling layers, fully connected layers, and an output layer. Since the attention mechanism is used to spatially and channel-weight the feature maps output from the convolutional layers of the CNN model, it is usually added after the convolutional layers.

[0050] The attention mechanism module in this application can be a CBAM (Convolutional Block Attention Module), a BAM (Bottleneck Attention Module), a Dual Attention Network (DAN), an Attention-based Feature Fusion (AFNet), or a Focal Attention Network (FAN), etc. However, due to the advantages of the CBAM module, such as high efficiency, lightweight design, ease of integration, and high flexibility, the attention mechanism module will be configured as a CBAM module in most cases.

[0051] When configuring the attention mechanism module as a CBAM module, the CBAM module is often added after the convolutional blocks of the convolutional neural network model. In one example, the convolutional neural network model contains four convolutional blocks, and a different number of CBAM modules are added after each convolutional block: 3 CBAM modules after the first convolutional block, 4 CBAM modules after the second convolutional block, 6 CBAM modules after the third convolutional block, and 3 CBAM modules after the fourth convolutional block. With this configuration, the CBAM module performs spatial and channel weighting on the output feature map of each convolutional block.

[0052] A convolutional block typically includes: convolutional layers, activation function layers, and pooling layers.

[0053] In this embodiment, the CBAM module includes a spatial attention submodule and a channel attention submodule, and the CBAM module is obtained by concatenating the spatial attention submodule and the channel attention submodule.

[0054] The spatial attention submodule is used to perform spatial weighting on the feature maps. In one example, the structure of the spatial attention submodule may include the following parts:

[0055] The global pooling layer performs global max pooling and global average pooling operations on the input feature map of size C×L, respectively, to obtain two feature maps of size 1×L. By extracting global information at each spatial location, the global pooling layer provides a basis for subsequent weighting, helping the network focus on global features at different locations.

[0056] The concatenation layer combines the results of global max pooling and global average pooling to generate a feature map of size 2×L. By combining the two pooling results, the concatenation layer enhances the model's ability to capture spatial features, enabling the model to utilize information from both max pooling and average pooling simultaneously.

[0057] The convolutional layer performs one-dimensional convolution on the concatenated feature maps to obtain a feature map of size 1×L. The convolution operation extracts spatial features through a one-dimensional convolution kernel, further adjusting and optimizing the feature representation. Specifically, the kernel size can be set to 7, the input channels can be set to 2 (corresponding to the concatenated feature maps), and the output channel can be 1.

[0058] The activation function layer uses the sigmoid activation function to activate the convolutional result, generating a spatial attention weight matrix. This matrix is ​​used to perform spatial weighting on the input feature map, thereby highlighting key regions and suppressing irrelevant or redundant regions, allowing the model to focus more on important information areas.

[0059] The channel attention submodule is used to perform channel-weighted processing on the feature map. In one example, the structure of the channel attention submodule may include the following parts:

[0060] The global pooling layer performs global average pooling and max pooling operations on each channel of a C×L feature map, resulting in two C×1 feature maps. By extracting global information for each channel, the global pooling layer helps the model capture global features at the channel level, providing a basis for subsequent weighting and strengthening the model's focus on important channels.

[0061] The convolutional layer takes the two C×1 feature maps and performs convolution operations on them separately, convolving them into feature maps of size (C / 4)×1. Then, a ReLU activation function is used to perform a non-linear transformation on the convolution result, resulting in an activated feature map. Next, a convolution operation is performed to restore the activated feature map to a C×1 feature map. It is important to note that when performing convolution operations on the two C×1 feature maps separately, it is crucial to ensure that both convolution operations use the same kernel parameters. This guarantees consistency in channel weighting, avoids differences in weighting methods between different channels, and thus maintains the effectiveness and stability of the weighting, ensuring consistency and optimization results in feature weighting.

[0062] The addition and activation function layer adds the two C×1 feature maps after convolution to obtain a single C×1 feature map. Then, the sigmoid activation function is used to activate this feature map, generating a channel attention weight matrix. This matrix represents the weight of each channel and is used to weight the input feature map along the channel dimension, highlighting key channels and suppressing irrelevant channels.

[0063] In this embodiment of the application, before optimizing the nuclide recognition model to be trained to obtain the target nuclide recognition model, it is often necessary to first construct the nuclide recognition model to be trained. Once the nuclide recognition model to be trained is constructed, the following steps can be followed to optimize and train it to obtain the target nuclide recognition model:

[0064] First, obtain training and testing sets. Then, use the training set to iterate the nuclide recognition model under training for multiple rounds until the set iteration stopping condition is met. After that, when the iteration stopping condition is met, use the testing set to test the nuclide recognition model under training. If the test results meet the preset standard, the nuclide recognition model under training is used as the target nuclide recognition model.

[0065] In this embodiment, the training set and the test set each include multiple sample nuclide energy spectra and the corresponding annotation results for each sample nuclide energy spectrum. The multiple sample nuclide energy spectra include single nuclide energy spectra and mixed nuclide energy spectra. Simultaneously configuring single nuclide energy spectra and mixed nuclide energy spectra in the multiple sample nuclide energy spectra helps the target nuclide identification model learn the characteristics of different types of nuclides, enabling the model to accurately identify nuclides regardless of whether the energy spectrum of the nuclide to be identified is a single nuclide or a mixed nuclide energy spectrum.

[0066] In the embodiments of this application, during the process of collecting sample nuclide energy spectra to construct training and testing sets, it is often necessary to obtain single nuclide energy spectra corresponding to multiple nuclides, and add some or all of the single nuclide energy spectra corresponding to multiple nuclides to the training and testing sets.

[0067] For sample nuclide energy spectra that are mixed nuclide energy spectra from multiple sample nuclide energy spectra, the following steps are taken: First, obtain the individual nuclide energy spectra corresponding to each of the multiple nuclides. Then, randomly select at least two individual nuclide energy spectra corresponding to each of the multiple nuclides. Next, mix the individual nuclide energy spectra corresponding to the at least two nuclides to obtain the mixed nuclide energy spectrum, and use the mixed nuclide energy spectrum as the sample nuclide energy spectrum.

[0068] Generally, to simplify the training process and improve data diversity, when randomly selecting single-nuctopic energy spectra of at least two nuclides, only the single-nuctopic energy spectra of the two nuclides are usually selected for mixing. That is, in multiple sample nuctopic energy spectra, the sample nuctopic energy spectrum that serves as the mixed nuctopic energy spectrum is usually formed by mixing the single-nuctopic energy spectra corresponding to the two nuclides respectively.

[0069] It should be noted that when the sample nuclide energy spectrum, which is used as a mixed nuclide energy spectrum, is composed of a mixture of the energy spectra of two individual nuclides, in order for the target nuclide identification model to still be able to perform nuclide identification efficiently and accurately when faced with mixed nuclide energy spectra of three or more nuclides, it is necessary to ensure that the trained target nuclide identification model has strong generalization ability.

[0070] Furthermore, to ensure the target nuclide identification model has strong generalization ability, the diversity of mixed nuclide energy spectra can be appropriately increased during model training. Therefore, when randomly selecting single nuclide energy spectra corresponding to at least two nuclides, it is not limited to randomly selecting only single nuclide energy spectra corresponding to two nuclides. In this way, the sample nuclide energy spectra used as mixed nuclide energy spectra can be composed of a mixture of single nuclide energy spectra of two nuclides, or a mixture of single nuclide energy spectra of three or more nuclides. This effectively increases the diversity of mixed nuclide energy spectra.

[0071] In this embodiment, "multiple nuclides" can refer to 48 nuclides. Among these nuclides, some nuclides can have their corresponding single-nuctopic energy spectra obtained through radiation detection. For these nuclides, their single-nuctopic energy spectra can be obtained through actual radiation detection. However, some nuclides are difficult to prepare or are inaccessible due to objective limitations, and therefore their corresponding single-nuctopic energy spectra cannot be obtained through radiation detection. In this case, it is necessary to simulate the single-nuctopic energy spectra of these nuclides through computer simulation, for example, by using the Monte Carlo simulation method to simulate the single-nuctopic energy spectra corresponding to these nuclides. That is to say, in this embodiment, the single-nuctopic energy spectrum can be either a real nuctopic energy spectrum obtained through radiation detection or a simulated nuctopic energy spectrum generated by computer simulation.

[0072] In this embodiment, to improve the robustness of the target nuclide identification model, image enhancement processing can be performed on the single-nuclide energy spectrum and / or mixed-nuclide energy spectrum before adding them to the training set. For example, noise can be added to the single-nuclide energy spectrum and / or mixed-nuclide energy spectrum. This enhancement processing enables the model to better adapt to noise interference, improving its robustness in practical applications. However, it should be noted that to avoid the adverse effects of noise on the energy spectrum data, when generating noise for each data point using a random function, the cumulative error of noise on the energy spectrum counting should be controlled to effectively reduce the same-direction error and ensure that the counting error of the entire energy spectrum eventually tends to zero. This processing method ensures that the added noise does not destroy the statistical properties of the energy spectrum, thereby improving the model's generalization ability and adaptability to noise.

[0073] In this embodiment, during the multi-round iterative training of the nuclide recognition model to be trained, the steps of each round of iterative training are as follows: First, select the target sample nuclide energy spectrum image from the training set. Then, input the target sample nuclide energy spectrum image into the nuclide recognition model to be trained to obtain the corresponding sample recognition result. Next, calculate the target loss value between the sample recognition result and the nuclide annotation result corresponding to the target sample image. Finally, update the internal model parameters in the nuclide recognition model to be trained using the target loss value.

[0074] In this embodiment of the application, the iteration stopping condition can be: the model target loss value converges, the number of iterations has reached the set maximum number of iterations, or other set conditions.

[0075] In this embodiment, the model's internal parameters include: convolutional layer weights, bias parameters, and fully connected layer weights.

[0076] In this embodiment of the application, the nuclide labeling result refers to the actual nuclide corresponding to the pre-labeled sample nuclide energy spectrum.

[0077] In this embodiment, a test set is used to test the nuclide recognition model to be trained, and the nuclide recognition model to be trained is used as the target nuclide recognition model only when the test results meet the preset standards, which can ensure that the recognition performance of the model meets the requirements.

[0078] In one example, the test results include the accuracy of the nuclide identification model to be trained.

[0079] In another example, the test results further include the recall and / or mean precision of the nuclide recognition model to be trained.

[0080] When the test results include precision, recall, and mean precision, the preset standards include precision thresholds, recall thresholds, and mean precision thresholds. In this case, the test results meeting the preset standards means that the precision of the nuclide recognition model to be trained reaches the precision threshold, the recall of the nuclide recognition model to be trained reaches the recall threshold, and the mean precision of the nuclide recognition model to be trained reaches the mean precision threshold.

[0081] The nuclide identification method provided in this application embodiment is obtained by optimizing and training the target nuclide identification model based on the nuclide identification model to be trained. The nuclide identification model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model. The attention mechanism module is used to perform spatial and channel weighted processing on the feature map output by the convolutional layer of the convolutional neural network model.

[0082] By applying spatial weighting to the feature map, the target nuclide identification model can highlight key regions in the energy spectrum of the target nuclide while suppressing irrelevant or redundant regions. This allows the target nuclide identification model to more accurately extract spatial features relevant to nuclide identification, enhances its ability to focus on regions crucial for nuclide identification, and reduces interference from background noise and irrelevant information, thereby enabling the model to better identify nuclide features.

[0083] Meanwhile, by applying channel weighting to the feature map, the target nuclide identification model can highlight the most contributing feature channels while suppressing irrelevant or redundant channels. This allows the target nuclide identification model to more accurately identify key channel features in the nuclide energy spectrum, thereby improving the effectiveness and accuracy of nuclide identification.

[0084] Therefore, by performing spatial weighting and channel weighting on the feature map, the target nuclide identification model can more accurately identify the nuclide corresponding to the energy spectrum of the nuclide to be identified, thereby more accurately identifying the nuclide corresponding to the energy spectrum of the nuclide to be identified.

[0085] Furthermore, the nuclide identification method provided in this application embodiment can identify nuclides in both single-nuclide and mixed-nuclide energy spectrum diagrams.

[0086] Furthermore, since the target nuclide identification model is used to identify the nuclide energy spectrum to obtain the corresponding target identification result, thereby identifying the nuclide corresponding to the nuclide energy spectrum, the nuclide identification method provided in this application embodiment can identify the nuclide energy spectrum in a relatively efficient manner, thereby identifying the nuclide corresponding to the nuclide energy spectrum in a relatively efficient manner.

[0087] Therefore, the nuclide identification method provided in this application embodiment can efficiently and accurately identify the nuclide energy spectrum, whether it is a single nuclide energy spectrum or a mixed nuclide energy spectrum, thereby efficiently and accurately identifying the nuclide corresponding to the nuclide energy spectrum.

[0088] Corresponding to the nuclide identification method provided in the embodiments of this application, the embodiments of this application also provide a nuclide identification device. For example... Figure 2 The diagram shown is a structural block diagram of a radionuclide identification device 200 according to an embodiment of this application. The device 200 may include:

[0089] The module 201 for acquiring the energy spectrum of the nuclide to be identified is used to acquire the energy spectrum of the nuclide to be identified; the energy spectrum of the nuclide to be identified is either a single nuclide energy spectrum or a mixed nuclide energy spectrum.

[0090] The nuclide identification module 202 is used to identify the nuclide in the energy spectrum of the nuclide to be identified using the target nuclide identification model, and obtain the corresponding target identification result. The target nuclide identification model is obtained by optimizing and training the nuclide identification model to be trained. The nuclide identification model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model. The attention mechanism module is used to perform spatial and channel weighting processing on the feature map output by the convolutional layer of the convolutional neural network model. The target identification result is used to represent the nuclide corresponding to the energy spectrum of the nuclide to be identified.

[0091] In one possible implementation, the attention mechanism module is added after the convolutional layer.

[0092] In one possible implementation, the steps for optimizing the nuclide recognition model to obtain the target nuclide recognition model are as follows:

[0093] Obtain training and testing sets; the training and testing sets each include multiple sample nuclide energy spectra and the corresponding annotation results for each sample nuclide energy spectrum; the multiple sample nuclide energy spectra include single nuclide energy spectra and mixed nuclide energy spectra;

[0094] The training set is used to iterate the nuclide recognition model under training for multiple rounds until the set iteration stopping condition is met.

[0095] When the iteration stopping condition is met, the nuclide recognition model to be trained is tested using the test set, and if the test results meet the preset standards, the nuclide recognition model to be trained is used as the target nuclide recognition model.

[0096] In one possible implementation, the steps for each round of iterative training are as follows:

[0097] Select the nuclide energy spectrum of the target sample in the training set;

[0098] The target sample nuclide energy spectrum is input into the nuclide recognition model to be trained to obtain the corresponding sample recognition result;

[0099] Calculate the target loss value between the sample recognition result and the nuclide annotation result corresponding to the target sample image;

[0100] The internal model parameters in the nuclide recognition model to be trained are updated using the target loss value.

[0101] In one possible implementation, the test results include the accuracy of the nuclide identification model to be trained.

[0102] In one possible implementation, the nuclide energy spectrum acquisition module 201 is specifically used to acquire the single nuclide energy spectrum corresponding to each of the multiple nuclides; randomly select at least two single nuclide energy spectra corresponding to each of the multiple nuclides; perform a mixing process on the single nuclide energy spectra corresponding to each of the at least two nuclides to obtain a mixed nuclide energy spectrum, and use the mixed nuclide energy spectrum as the sample nuclide energy spectrum.

[0103] In one possible implementation, the energy spectrum of a single nuclide is either a real energy spectrum obtained through radiation detection or a simulated energy spectrum generated through computer simulation.

[0104] In one possible implementation, the attention mechanism module is a convolutional block attention module.

[0105] The functions of each module in each device in the embodiments of this application can be found in the corresponding description in the above method, and they have corresponding beneficial effects, which will not be repeated here.

[0106] Figure 3This is a block diagram of an electronic device used to implement embodiments of this application. For example... Figure 3 As shown, the electronic device includes a memory 301 and a processor 302. The memory 301 stores a computer program that can run on the processor 302. When the processor 302 executes the computer program, it implements the method described in the above embodiments. The number of memories 301 and processors 302 can be one or more.

[0107] The electronic device also includes:

[0108] Communication interface 303 is used to communicate with external devices and perform data exchange and transmission.

[0109] If the memory 301, processor 302, and communication interface 303 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0110] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0111] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in this application.

[0112] This application also provides a chip, which includes a processor for calling and executing instructions stored in a memory, causing a communication device on which the chip is installed to perform the method provided in this application.

[0113] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0114] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting Advanced Reduced Instruction Set Machines (ARM) architecture.

[0115] Further, optionally, the aforementioned memory may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory may be volatile or non-volatile, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0116] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0117] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0118] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0119] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0120] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0121] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0122] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0123] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying nuclides, comprising: Obtain the energy spectrum of the nuclide to be identified; The energy spectrum of the nuclide to be identified is either a single nuclide energy spectrum or a mixed nuclide energy spectrum; The target nuclide is identified by using a target nuclide identification model to identify the energy spectrum of the nuclide to be identified, and the corresponding target identification result is obtained. The target nuclide identification model is obtained by optimizing and training the nuclide identification model to be trained. The nuclide recognition model to be trained is obtained by adding an attention mechanism module to the convolutional neural network model; the attention mechanism module is used to perform spatial and channel weighting processing on the feature map output by the convolutional layer of the convolutional neural network model; the target recognition result is used to represent the nuclide corresponding to the energy spectrum map of the nuclide to be identified; The steps for optimizing and training the nuclide recognition model to obtain the target nuclide recognition model are as follows: Obtain a training set and a test set; the training set and the test set each include multiple sample nuclide energy spectra and the corresponding annotation results for each sample nuclide energy spectrum; the multiple sample nuclide energy spectra include the energy spectrum of the single nuclide and the energy spectrum of the mixed nuclide; The nuclide identification model to be trained is subjected to multiple rounds of iterative training using the training set until the set iteration stopping condition is met; When the iteration stopping condition is met, the nuclide recognition model to be trained is tested using the test set, and if the test result meets the preset standard, the nuclide recognition model to be trained is used as the target nuclide recognition model.

2. The method according to claim 1, wherein, The attention mechanism module is added after the convolutional layer.

3. The method according to claim 1, wherein, The steps for each round of iterative training are as follows: Select the target sample nuclide energy spectrum from the training set; The target sample nuclide energy spectrum is input into the nuclide recognition model to be trained to obtain the corresponding sample recognition result; Calculate the target loss value between the sample recognition result and the nuclide annotation result corresponding to the target sample image; The internal model parameters in the nuclide recognition model to be trained are updated using the target loss value.

4. The method according to claim 1, wherein, The test results include the accuracy of the nuclide recognition model to be trained.

5. The method according to claim 1, wherein the energy spectrum of the sample nuclide in the plurality of sample nuclide energy spectra is the energy spectrum of the mixed nuclide, is obtained by the following steps: Obtain the energy spectra of individual nuclides for various nuclides; Among the single-nuclide energy spectrum diagrams corresponding to the various nuclides, at least two single-nuclide energy spectrum diagrams corresponding to the various nuclides are randomly selected. The energy spectra of the individual nuclides corresponding to the at least two nuclides are mixed to obtain the mixed nuclide energy spectrum, and the mixed nuclide energy spectrum is used as the sample nuclide energy spectrum.

6. The method according to claim 1, wherein, The single nuclide energy spectrum is either a real nuclide energy spectrum obtained through radiation detection, or a simulated nuclide energy spectrum generated through computer simulation.

7. The method according to claim 1, wherein, The attention mechanism module is a convolutional block attention module.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein the processor, when executing the computer program, implements the method of any one of claims 1-7.

9. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images