Beta-gamma coincidence energy spectrum analysis method, system and equipment based on AI and medium

Abstract

The invention relates to the field of energy spectrum analysis, in particular to an AI-based beta-gamma coincidence energy spectrum analysis method, system and device and a medium, and the method comprises the steps: inputting a to-be-processed energy spectrum into an input convolution module for image feature extraction, and obtaining initial image features; performing multiple times of spatial dimension reduction on the initial image features through a down-sampling module, and extracting the image features after each time of spatial dimension reduction to obtain compression energy spectrum features of different sizes; performing spatial dimension improvement and fusion on the compressed energy spectrum features through an up-sampling module to obtain a tail end decoding feature map with the same size as the to-be-processed energy spectrum map; and through an output module, calculating the probability that each pixel in the tail end decoding feature map corresponds to each nuclide category, generating a prediction mask map according to the probability, and carrying out nuclide analysis on the to-be-processed energy spectrum map based on the prediction mask map. According to the method, automatic and high-precision nuclide region segmentation is realized in a complex scene, and compared with a traditional method, the division efficiency and the anti-interference capability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of energy spectrum analysis, and in particular to an AI-based β-γ coincidence energy spectrum analysis method, system, equipment and medium. Background Art

[0002] The β-γ coincidence spectrum is high-dimensional data acquired through a β-γ coincidence measurement system. It uses β-ray energy, γ-ray energy, and coincidence counts as three-dimensional coordinates to form an energy distribution matrix that contains the decay characteristics of nuclides. In nuclear monitoring (such as nuclear test ban verification), the β-γ coincidence spectrum requires precise delineation of the energy distribution regions of different nuclides (such as 214Pb and 133Xe) to quantitatively analyze nuclide activity and determine their origin.

[0003] However, due to the presence of multi-nuclide signal superposition, low-count noise interference and energy resolution limitations in the β-γ coincidence energy spectrum, traditional methods rely on manual experience to set energy thresholds or fit energy spectrum curves, resulting in low efficiency and high subjectivity in the nuclide region division, making it difficult to meet real-time monitoring needs. Summary of the Invention

[0004] To solve the above problems, the present invention provides an AI-based β-γ coincidence spectrum analysis method, system, device and medium.

[0005] The first aspect of the present invention discloses an AI-based β-γ coincidence spectrum analysis method, comprising:

[0006] Inputting the energy spectrum to be processed into the input convolution module of the mask prediction model for image feature extraction to obtain initial image features; wherein the energy spectrum to be processed refers to an analysis spectrum including energy characteristics of a specific nuclide category generated by simultaneously detecting coincident events of beta decay and accompanying gamma ray generation;

[0007] Performing multiple spatial dimensionality reductions on the initial image features through a downsampling module of the mask prediction model, and extracting image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes;

[0008] The compressed energy spectrum features are spatially enhanced and fused through the upsampling module of the mask prediction model to obtain a terminal decoding feature map with the same size as the energy spectrum map to be processed;

[0009] The output module of the mask prediction model calculates the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category, and generates a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed;

[0010] The nuclide analysis is performed on the energy spectrum to be processed based on the predicted mask image.

[0011] Furthermore, the input convolution module includes multiple cascaded input convolution networks; wherein, the input data of the first input convolution network is the energy spectrum to be processed, the input data of other input convolution networks are the output features of the previous input convolution network in the cascade, and the output features of the last input convolution network are the initial image features;

[0012] And, each input convolutional network processes its input data in the following way:

[0013] The image features of the input data are extracted through the convolution layer of the input convolutional network to obtain the initial convolution features;

[0014] Performing batch normalization on the initial convolutional features by inputting a normalization layer of a convolutional network to obtain an initial normalized feature map;

[0015] The initial normalized feature map is subjected to nonlinear activation processing by a nonlinear activation layer of the input convolutional network to obtain output features of the input convolutional network.

[0016] Furthermore, the downsampling module includes multiple cascaded downsampling networks, wherein the input data of the first downsampling network is the initial image feature, and the input data of other downsampling networks are the compressed energy spectrum features of corresponding sizes output by the previous downsampling network in the cascade;

[0017] And, each downsampling network processes its input data by:

[0018] Performing spatial dimensionality reduction on the input data through the maximum pooling layer of the downsampling network to obtain a first compressed energy spectrum;

[0019] Performing feature extraction on the first compressed energy spectrum through the convolutional layer of the downsampling network to obtain first compressed energy spectrum features;

[0020] performing batch normalization processing on the first compressed energy spectrum feature through the normalization layer of the downsampling network to obtain a normalized compressed energy spectrum feature;

[0021] The normalized compressed energy spectrum feature is nonlinearly activated through the nonlinear activation layer of the downsampling network to obtain the compressed energy spectrum feature of the corresponding size of the downsampling network.

[0022] Furthermore, the upsampling module includes multiple cascaded upsampling networks, wherein the output of each upsampling network is a spatial reconstruction fusion feature of a corresponding size; and the input data of the first upsampling network is a compressed energy spectrum feature of a corresponding size and a compressed energy spectrum feature of a minimum size, and the input of other downsampling networks is the spatial reconstruction fusion feature output by the previous upsampling network and the compressed energy spectrum feature of the corresponding size, and the spatial reconstruction fusion feature output by the last downsampling network is the terminal decoding feature map;

[0023] And, the upsampling processing steps of each upsampling network include:

[0024] Performing an upsampling operation on the feature map to be expanded through the upsampling layer of the upsampling network to obtain a spatial expansion map corresponding to its target size; wherein the upsampling operation refers to a bilinear interpolation operation or a transposed convolution operation;

[0025] The spatial expansion image and the compressed energy spectrum features having the same target size are spliced ​​together through the splicing layer of the upsampling network to obtain a fused energy spectrum feature map;

[0026] Performing feature extraction on the fused energy spectrum feature map through the convolution layer of the upsampling network to obtain a fused energy spectrum feature;

[0027] Performing batch normalization processing on the fused energy spectrum features through the normalization layer of the upsampling network to obtain normalized fused energy spectrum features;

[0028] The normalized fusion energy spectrum feature is nonlinearly activated through the nonlinear activation layer of the upsampling network to obtain a spatial reconstruction fusion feature.

[0029] Furthermore, the mask prediction model is trained by adjusting model parameters through a backpropagation algorithm and a gradient descent optimizer according to a predefined loss function; wherein the loss function is a weighted sum of a distribution focus loss function and a smoothing loss function; the distribution focus loss function represents the probability distribution difference loss between the predicted probability of the nuclide category of all pixels in the β-γ conformity energy spectrum sample during training and its true nuclide category; the smoothing loss function represents the similarity loss between the pixel area corresponding to each nuclide category in the predicted mask image and the nuclide area actually corresponding to the nuclide category.

[0030] Furthermore, the distribution focus loss function is:

[0031]

[0032] in, L df3represents the distribution focus loss function, b represents the β-γ energy spectrum sample number during training, B represents the total number of β-γ energy spectrum samples during training; i represents the pixel number in the β-γ energy spectrum sample, I represents the number of pixels in the β-γ energy spectrum sample; c represents the nuclide number, C represents the number of nuclides, y b,i,c represents the true probability that the pixel belongs to the cth nuclide category, p b,i,c represents the predicted probability that the pixel belongs to the cth nuclide category.

[0033] Furthermore, the smoothing loss function is:

[0034]

[0035] Among them, L Dice represents the smoothing loss function, b represents the β-γ spectrum sample number during training, B represents the total number of β-γ spectrum samples during training, c represents the nuclide number, C represents the number of nuclides, i represents the pixel number in the β-γ spectrum sample, I represents the number of pixels in the β-γ spectrum sample, y b,i,c It represents the true probability that the i-th pixel in the b-th β-γ spectrum sample is the c-th nuclide category, p b,i,c represents the predicted probability that the pixel belongs to the cth nuclide category.

[0036] The second aspect of the present invention discloses an AI-based β-γ coincidence spectrum analysis system, comprising:

[0037] an extraction module configured to input the energy spectrum to be processed into the input convolution module of the mask prediction model for image feature extraction to obtain initial image features; wherein the energy spectrum to be processed is an analysis spectrum including energy characteristics of a specific nuclide category generated by simultaneously detecting coincident events of beta decay and accompanying gamma ray generation;

[0038] A dimensionality reduction module, configured to perform multiple spatial dimensionality reductions on the initial image features through the downsampling module of the mask prediction model, and extract image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes;

[0039] A fusion module is used to perform spatial dimension enhancement and fusion on the compressed energy spectrum features through the upsampling module of the mask prediction model to obtain a terminal decoding feature map with the same size as the energy spectrum map to be processed;

[0040] A classification module, configured to calculate the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category through the output module of the mask prediction model, and generate a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed;

[0041] An analysis module is used to perform nuclide analysis on the energy spectrum to be processed based on the predicted mask image.

[0042] The third aspect of the present invention discloses an electronic device, which includes a memory, a processor, and a computer program stored in the memory and runable on the processor. The device is characterized in that when the processor executes the computer program, it implements the steps of any one of the AI-based β-γ coincidence energy spectrum analysis methods disclosed in the first aspect of the present invention.

[0043] The fourth aspect of the present invention discloses a storage medium storing a computer program, wherein the storage medium is characterized in that when the computer program is executed by a processor, the steps of any one of the AI-based β-γ coincidence energy spectrum analysis methods disclosed in the first aspect of the present invention are implemented.

[0044] The encoder-decoder architecture constructed in this paper gradually abstracts the high-level semantic features of the nuclide region through multiple downsampling operations in the encoder path. Simultaneously, the decoder path utilizes jump connections to fuse shallower detail features, achieving a precise mapping from global energy distribution to local pixel classification. This method not only avoids the subjective errors of manual rules but also exploits the nuclide energy correlation patterns implicit in the energy spectrum through end-to-end training, ultimately achieving automated, high-precision nuclide region segmentation in complex scenarios, significantly improving segmentation efficiency and anti-interference capabilities compared to traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 This is a flow chart of an AI-based β-γ coincidence spectrum analysis method disclosed in an embodiment of the present invention;

[0047] Figure 2 This is a schematic structural diagram of an AI-based β-γ coincidence spectrum analysis system disclosed in an embodiment of the present invention;

[0048] Figure 3 It is a schematic structural diagram of an electronic device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0050] The terms "first," "second," and so on, in the description and claims of the present invention and the accompanying drawings are used to distinguish between different objects, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, or product comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, apparatus, or product.

[0051] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0052] See also Figure 1 As shown, Figure 1 This is a flow chart of an AI-based β-γ coincidence spectrum analysis method disclosed in an embodiment of the present invention. Figure 1 As shown, the AI-based β-γ coincidence spectrum analysis method may include the following operations:

[0053] S101. Inputting the energy spectrum to be processed into the input convolution module of the mask prediction model to extract image features and obtain initial image features; wherein the energy spectrum to be processed refers to an analysis spectrum including energy characteristics of a specific nuclide category generated by simultaneously detecting coincident events of β decay and accompanying γ ray generation;

[0054] In an optional embodiment, the backbone model of the mask prediction model is a U-Net network.

[0055] In an optional embodiment, the input convolution module includes multiple cascaded input convolutional networks; wherein the input data of the first input convolutional network is the energy spectrum to be processed, the input data of the other input convolutional networks are the output features of the previous input convolutional network in the cascade, and the output features of the last input convolutional network are the initial image features;

[0056] And, each input convolutional network processes its input data in the following way:

[0057] The image features of the input data are extracted through the convolution layer of the input convolutional network to obtain the initial convolution features;

[0058] The initial convolution features are batch normalized by inputting the normalization layer of the convolutional network to obtain the initial normalized feature map;

[0059] The initial normalized feature map is processed by nonlinear activation layer of the input convolutional network to obtain the output features of the input convolutional network.

[0060] In this optional embodiment, the input convolution module includes two cascaded input convolutional networks. The first convolution extracts low-level features, and the second convolution extracts high-level features based on these low-level features, thereby obtaining more comprehensive and robust image features. Moreover, while the first convolution extracts features, it may also cause some loss of detailed information. The second convolution can recapture these details on the output of the first convolution, reducing information loss.

[0061] In this optional embodiment, batch normalization is used to normalize the intermediate features of the neural network. Its main purpose is to reduce internal covariate shift, accelerate network convergence, and improve the generalization ability of the network. Batch normalization normalizes the features of each batch of data to make the data distribution more consistent between different batches, thereby making the network learning process more stable. In this optional embodiment, batch normalization normalizes each feature dimension independently so that the mean of each dimension is 0 and the variance is 1.

[0062] In this optional embodiment, the nonlinear activation process is processed by the ReLU activation function. The nonlinear activation process enables the network to learn and represent nonlinear feature patterns by applying a nonlinear function to the output of the neuron.

[0063] It can be seen that this optional embodiment adopts double convolution operations and uses a cascaded input convolutional network to perform multi-level feature extraction on the energy spectrum to be processed. Combined with batch normalization processing and nonlinear activation processing, it can better extract the energy characteristics of the β-γ coincidence energy spectrum, reduce information loss, and improve the quality and robustness of feature representation, thereby providing a more accurate and reliable data basis for subsequent nuclide identification and analysis tasks, and further improving the efficiency and accuracy of β-γ coincidence energy spectrum analysis.

[0064] S102, performing multiple spatial dimensionality reductions on the initial image features through a downsampling module of a mask prediction model, and extracting image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes;

[0065] In another optional embodiment, the downsampling module includes multiple cascaded downsampling networks, wherein the input data of the first downsampling network is the initial image feature, and the input data of the other downsampling networks are the compressed energy spectrum features of corresponding sizes output by the previous downsampling network in the cascade;

[0066] And, each downsampling network processes its input data by:

[0067] The input data is spatially reduced through the maximum pooling layer of the downsampling network to obtain the first compressed energy spectrum;

[0068] Performing feature extraction on the first compressed energy spectrum through the convolutional layer of the downsampling network to obtain first compressed energy spectrum features;

[0069] Performing batch normalization processing on the first compressed energy spectrum feature through the normalization layer of the downsampling network to obtain a normalized compressed energy spectrum feature;

[0070] The normalized compressed energy spectrum features are nonlinearly activated through the nonlinear activation layer of the downsampling network to obtain the compressed energy spectrum features of the corresponding size of the downsampling network.

[0071] In this optional embodiment, the max pooling layer divides the input data into several non-overlapping regions and selects the largest eigenvalue within each region as the representative value of that region, thereby obtaining a reduced-size feature map. In the present invention, the max pooling layer is used to perform spatial dimensionality reduction on the initial image features or the compressed energy spectrum features output by the previous downsampling network. By gradually reducing the size of the feature map, it can effectively extract and compress key information in the β-γ coincidence energy spectrum, while reducing computational complexity and memory consumption, and improving the efficiency and generalization ability of the model.

[0072] In an optional embodiment, the downsampling module utilizes four cascaded downsampling networks, each with a convolutional layer that has input data of a different size. The input data for the convolutional layer of the first downsampling network is the initial image features, which are the same size as the energy spectrum to be processed. The input data for the convolutional layers of subsequent downsampling networks is the output features of the previous downsampling network, whose size has been reduced through convolution and pooling operations. This design allows for the gradual extraction and fusion of feature information from the energy spectrum at different scales. Shallower downsampling networks can capture more localized and detailed features, while deeper downsampling networks can achieve more global and abstract feature representations.

[0073] In an optional embodiment, a downsampling network may include two convolutional layers. The output features of the first convolutional layer are processed in sequence through a normalization layer and a nonlinear activation layer and then input into the second convolutional layer. The output features of the second convolutional layer are also processed in sequence through a normalization layer and a nonlinear activation layer and then input into the next downsampling network.

[0074] It can be seen that this optional embodiment achieves spatial dimensionality reduction through the maximum pooling layer to retain key features, the convolution layer extracts deep semantic information, the normalization layer improves training stability, and the nonlinear activation enhances feature expression capabilities, thereby efficiently capturing the nuclide distribution characteristics of complex β-γ energy spectrum data during the multi-scale compression process, significantly improving the model's ability to analyze energy spectrum structure and the accuracy and robustness of mask generation.

[0075] S103, performing spatial dimension enhancement and fusion on the compressed energy spectrum features through the upsampling module of the mask prediction model to obtain a terminal decoding feature map with the same size as the energy spectrum map to be processed;

[0076] In an optional embodiment, the upsampling module includes multiple cascaded upsampling networks, wherein the output of each upsampling network is a spatial reconstruction fusion feature of a corresponding size; and the input data of the first upsampling network is a compressed energy spectrum feature of a corresponding size and a compressed energy spectrum feature of a minimum size, and the input of other downsampling networks is the spatial reconstruction fusion feature output by the previous upsampling network and the compressed energy spectrum feature of the corresponding size, and the spatial reconstruction fusion feature output by the last downsampling network is a terminal decoding feature map;

[0077] And, the upsampling processing steps of each upsampling network include:

[0078] An upsampling operation is performed on the feature map to be expanded through the upsampling layer of the upsampling network to obtain a spatial expansion map corresponding to its target size; wherein the upsampling operation refers to a bilinear interpolation operation or a transposed convolution operation;

[0079] Through the splicing layer of the upsampling network, the spatial expansion image and the compressed energy spectrum features with the same target size are spliced ​​together to obtain a fused energy spectrum feature map;

[0080] Through the convolution layer of the upsampling network, the fusion energy spectrum feature map is extracted to obtain the fusion energy spectrum feature;

[0081] The fused energy spectrum features are batch normalized through the normalization layer of the upsampling network to obtain the normalized fused energy spectrum features;

[0082] The normalized fusion energy spectrum features are nonlinearly activated through the nonlinear activation layer of the upsampling network to obtain the spatial reconstruction fusion features.

[0083] In this optional embodiment, bilinear interpolation generates new pixel values ​​by linearly interpolating between pixels in the original image, thereby achieving image magnification and resolution improvement. Specifically, for a target pixel to be interpolated, bilinear interpolation first finds its four adjacent pixels in the original image. It then performs linear interpolation in both the horizontal and vertical directions, and finally combines the interpolation results in both directions to obtain the target pixel value. Bilinear interpolation effectively smooths images, reducing aliasing and distortion, while preserving the original image's primary structure and details.

[0084] In this optional embodiment, the input of the transposed convolution operation is a low-resolution feature map, and the output is a high-resolution feature map. By performing a convolution operation at each pixel position in the input feature map and superimposing the result at the corresponding position in the output feature map, the spatial expansion of the feature map is achieved. In this process, the convolution kernel parameters of the transposed convolution are learned, so the upsampling method can be adaptively adjusted to better restore and reconstruct the details of the original image.

[0085] It can be seen that this optional embodiment realizes the gradual recovery and improvement from low-resolution compressed energy spectrum features to high-resolution spatial reconstruction fusion features by using multiple cascaded upsampling networks. In this process, the upsampling network fully utilizes the compressed energy spectrum features at different scales, and effectively combines local details and global context information through cross-scale feature fusion and interaction, thereby obtaining a more accurate and complete energy spectrum reconstruction result. At the same time, through multiple upsampling and feature extraction, this embodiment can gradually refine and optimize the reconstructed energy spectrum, and obtain a high-quality decoding feature map with the same size as the original input at the end, which provides a good foundation for subsequent accurate mask map generation and nuclide category prediction, and significantly improves the performance and practical value of β-γ coincidence energy spectrum analysis.

[0086] S104, calculating the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category through the output module of the mask prediction model, and generating a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed;

[0087] S105 , performing nuclide analysis on the energy spectrum to be processed based on the predicted mask image.

[0088] In this optional embodiment, performing nuclide analysis on the energy spectrum to be processed refers to performing a quantitative or qualitative analysis of the spatial distribution, type and proportion of nuclides based on the nuclide category information marked on each pixel in the predicted mask image and the energy characteristics of the original energy spectrum.

[0089] In an optional embodiment, the mask prediction model is trained by adjusting the model parameters through a back-propagation algorithm and a gradient descent optimizer according to a predefined loss function; wherein the loss function is a weighted sum of a distribution focus loss function and a smoothing loss function; the distribution focus loss function represents the probability distribution difference loss between the predicted probability of the nuclide category of all pixels in the β-γ conforming energy spectrum sample during training and its true nuclide category; the smoothing loss function represents the similarity loss between the pixel area corresponding to each nuclide category in the predicted mask image and the nuclide area actually corresponding to the nuclide category.

[0090] It can be seen that this optional embodiment, by combining the backpropagation algorithm with the gradient descent optimizer and adopting the weighted sum of the distribution focus loss function and the smoothing loss function as the training objective, can simultaneously take into account the distribution matching accuracy of pixel-level nuclide category prediction and the structural consistency of the mask area when optimizing the model parameters, thereby improving the model's ability to analyze the mixed distribution of multiple nuclides in complex β-γ energy spectrum samples, while avoiding the problem of blurred mask boundaries or category misjudgment caused by overfitting of a single loss function.

[0091] In an optional embodiment, the distribution focus loss function is:

[0092]

[0093] in, L df3 represents the distribution focus loss function, n represents the number of β-γ energy spectrum samples during training, B represents the total number of β-γ energy spectrum samples during training; i represents the pixel number in the β-γ energy spectrum sample, I represents the number of pixels in the β-γ energy spectrum sample; c represents the nuclide number, C represents the number of nuclides, y b,i,c represents the true probability of the pixel being the cth nuclide category, p b,i,c Represents the predicted probability that the pixel belongs to the cth nuclide category.

[0094] It can be seen that in this optional embodiment, the distribution focus loss function calculates the cross entropy loss on a sample-by-sample and pixel-by-pixel basis, and constrains the model's prediction of the probability distribution of the nuclide category at each pixel point from a microscopic level to ensure a matching degree with the true distribution. This can significantly improve the model's recognition sensitivity to low-proportion nuclide categories and reduce missed detections and misjudgments caused by background noise or interference from the main nuclide.

[0095] In an optional embodiment, the smoothing loss function is:

[0096]

[0097] Among them, L Dicerepresents the smoothing loss function, b represents the number of β-γ energy spectrum samples during training, B represents the total number of β-γ energy spectrum samples during training, c represents the nuclide number, C represents the number of nuclides, i represents the pixel number in the β-γ energy spectrum sample, I represents the number of pixels in the β-γ energy spectrum sample, y b,i,c It represents the true probability that the i-th pixel in the b-th β-γ spectrum sample is the c-th nuclide category, p b,i,c Represents the predicted probability that the pixel belongs to the cth nuclide category.

[0098] It can be seen that in this optional embodiment, the smoothing loss function is designed based on the Dice coefficient. By measuring the overlap between the predicted mask and the real nuclide area, the continuity and spatial consistency of the mask are optimized from a macro level, and the appearance of isolated noise points or fragmented prediction areas is effectively suppressed. The nuclide distribution mask output by the model is more in line with the physical characteristics of the nuclide aggregation area in the real scene, thereby improving the interpretability and practical application value of the segmentation results.

[0099] See also Figure 2 As shown, Figure 2 1 is a schematic structural diagram of an AI-based β-γ coincidence spectrum analysis system disclosed in an embodiment of the present invention, including:

[0100] An extraction module 201 is configured to input the energy spectrum to be processed into the input convolution module of the mask prediction model for image feature extraction to obtain initial image features. The energy spectrum to be processed is an analysis spectrum including energy characteristics of a specific nuclide class generated by simultaneously detecting coincident events of β decay and accompanying γ ray generation.

[0101] A dimensionality reduction module 202 is configured to perform multiple spatial dimensionality reductions on the initial image features through a downsampling module of the mask prediction model, and extract image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes;

[0102] A fusion module 203 is configured to perform spatial dimension enhancement and fusion on the compressed energy spectrum features through an upsampling module of the mask prediction model to obtain a terminal decoding feature map of the same size as the energy spectrum map to be processed;

[0103] A classification module 204 is configured to calculate the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category through the output module of the mask prediction model, and generate a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed;

[0104] The analysis module 205 is configured to perform nuclide analysis on the energy spectrum to be processed based on the predicted mask image.

[0105] For the specific limitations of the AI-based β-γ coincidence spectrum analysis system, please refer to the limitations of the AI-based β-γ coincidence spectrum analysis method above, which will not be repeated here. The various modules in the above-mentioned AI-based β-γ coincidence spectrum analysis system can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the electronic device in hardware format, or can be stored in the memory of the electronic device in software format, so that the processor can call the corresponding operations of the above modules.

[0106] It should be noted that, in order to highlight the innovative part of the present invention, this embodiment does not introduce modules that are not closely related to solving the technical problem proposed by the present invention, but this does not mean that there are no other modules in this embodiment.

[0107] like Figure 3 As shown, the electronic device 1 provided by the present invention may include a memory 11, a processor 12 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 12, such as an AI-based β-γ coincidence spectrum analysis program.

[0108] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Furthermore, the memory 11 can also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed on the electronic device 1, such as the code for AI-based β-γ coincidence spectrum analysis, but can also be used to temporarily store data that has been output or is to be output.

[0109] In some embodiments, the processor 12 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 12 is the control core (Control Unit) of the electronic device 1, connecting the various components of the entire electronic device 1 using various interfaces and lines. It executes or runs programs or modules stored in the memory 11 (such as an AI-based β-γ coincidence spectrum analysis program) and calls data stored in the memory 11 to perform various functions of the electronic device 1 and process data.

[0110] The processor 12 executes the operating system and various installed applications of the electronic device 1. The processor 12 executes the applications to implement the steps in the above-mentioned AI-based β-γ coincidence spectrum analysis method.

[0111] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory 11 and executed by the processor 12 to complete the present application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program may be divided into an extraction module 201, a dimensionality reduction module 202, a fusion module 203, a classification module 204, and an analysis module 205.

[0112] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The above-mentioned software function module is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, computer device, or network device, etc.) or a processor to perform part of the functions of the AI-based β-γ coincidence spectrum analysis method described in various embodiments of the present application.

[0113] In summary, the AI-based β-γ coincidence spectrum analysis method, system, device, and medium disclosed in this invention not only avoids the subjective errors of manual rules but also, through end-to-end training, mines the nuclide energy correlation patterns implicit in the spectrum. Ultimately, it achieves automated, high-precision nuclide region segmentation in complex scenarios, significantly improving segmentation efficiency and anti-interference capabilities compared to traditional methods. Therefore, this invention effectively overcomes the various shortcomings of the existing technology and has high industrial application value.

[0114] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A β-γ coincidence spectrum analysis method based on AI, characterized in that: The method comprises: Inputting the energy spectrum to be processed into the input convolution module of the mask prediction model for image feature extraction to obtain initial image features; wherein the energy spectrum to be processed refers to an analysis spectrum including energy characteristics of a specific nuclide category generated by simultaneously detecting coincident events of beta decay and accompanying gamma ray generation; Performing multiple spatial dimensionality reductions on the initial image features through a downsampling module of the mask prediction model, and extracting image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes; The compressed energy spectrum features are spatially enhanced and fused through the upsampling module of the mask prediction model to obtain a terminal decoding feature map with the same size as the energy spectrum map to be processed; The output module of the mask prediction model calculates the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category, and generates a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed; The nuclide analysis is performed on the energy spectrum to be processed based on the predicted mask image.

2. The AI-based β-γ coincidence spectrum analysis method according to claim 1, characterized in that: The input convolution module includes multiple cascaded input convolution networks; wherein the input data of the first input convolution network is the energy spectrum to be processed, the input data of the other input convolution networks are the output features of the previous input convolution network in the cascade, and the output features of the last input convolution network are the initial image features; And, each input convolutional network processes its input data in the following way: The image features of the input data are extracted through the convolution layer of the input convolutional network to obtain the initial convolution features; Performing batch normalization on the initial convolutional features by inputting a normalization layer of a convolutional network to obtain an initial normalized feature map; The initial normalized feature map is subjected to nonlinear activation processing by a nonlinear activation layer of the input convolutional network to obtain output features of the input convolutional network.

3. The AI-based β-γ coincidence spectrum analysis method according to claim 1, characterized in that: The downsampling module includes multiple cascaded downsampling networks, wherein the input data of the first downsampling network is the initial image feature, and the input data of other downsampling networks are the compressed energy spectrum features of corresponding sizes output by the previous downsampling network in the cascade; And, each downsampling network processes its input data by: Performing spatial dimensionality reduction on the input data through the maximum pooling layer of the downsampling network to obtain a first compressed energy spectrum; Performing feature extraction on the first compressed energy spectrum through the convolutional layer of the downsampling network to obtain first compressed energy spectrum features; performing batch normalization processing on the first compressed energy spectrum feature through the normalization layer of the downsampling network to obtain a normalized compressed energy spectrum feature; The normalized compressed energy spectrum feature is nonlinearly activated through the nonlinear activation layer of the downsampling network to obtain the compressed energy spectrum feature of the corresponding size of the downsampling network.

4. The AI-based β-γ coincidence spectrum analysis method according to claim 1, characterized in that: The upsampling module includes multiple cascaded upsampling networks, wherein the output of each upsampling network is a spatial reconstruction fusion feature of a corresponding size; and the input data of the first upsampling network is a compressed energy spectrum feature of a corresponding size and a compressed energy spectrum feature of a minimum size, and the input of other downsampling networks is the spatial reconstruction fusion feature output by the previous upsampling network and the compressed energy spectrum feature of the corresponding size, and the spatial reconstruction fusion feature output by the last downsampling network is the terminal decoding feature map; And, the upsampling processing steps of each upsampling network include: Performing an upsampling operation on the feature map to be expanded through the upsampling layer of the upsampling network to obtain a spatial expansion map corresponding to its target size; wherein the upsampling operation refers to a bilinear interpolation operation or a transposed convolution operation; The spatial expansion image and the compressed energy spectrum features having the same target size are spliced ​​together through the splicing layer of the upsampling network to obtain a fused energy spectrum feature map; Performing feature extraction on the fused energy spectrum feature map through the convolution layer of the upsampling network to obtain a fused energy spectrum feature; Performing batch normalization processing on the fused energy spectrum features through the normalization layer of the upsampling network to obtain normalized fused energy spectrum features; The normalized fusion energy spectrum feature is nonlinearly activated through the nonlinear activation layer of the upsampling network to obtain a spatial reconstruction fusion feature.

5. The AI-based β-γ coincidence spectrum analysis method according to claim 1, characterized in that: The mask prediction model is trained by adjusting model parameters using a backpropagation algorithm and a gradient descent optimizer according to a predefined loss function; wherein the loss function is a weighted sum of a distribution focus loss function and a smoothing loss function; the distribution focus loss function represents the probability distribution difference loss between the predicted probability of the nuclide category of all pixels in the β-γ conformance spectrum sample during training and their true nuclide category; the smoothing loss function represents the similarity loss between the pixel area corresponding to each nuclide category in the predicted mask image and the nuclide area actually corresponding to the nuclide category.

6. The AI-based β-γ coincidence spectrum analysis method according to claim 5, characterized in that: The distribution focus loss function is: in, L df3 represents the distribution focus loss function, b represents the β-γ energy spectrum sample number during training, B represents the total number of β-γ energy spectrum samples during training; i represents the pixel number in the β-γ energy spectrum sample, I represents the number of pixels in the β-γ energy spectrum sample; c represents the nuclide number, C represents the number of nuclides, y b,i,c represents the true probability that the pixel belongs to the cth nuclide category, p b,i,c represents the predicted probability that the pixel belongs to the cth nuclide category.

7. The AI-based β-γ coincidence spectrum analysis method according to claim 5, characterized in that: The smoothing loss function is: Among them, L Dice represents the smoothing loss function, b represents the β-γ spectrum sample number during training, B represents the total number of β-γ spectrum samples during training, c represents the nuclide number, C represents the number of nuclides, i represents the pixel number in the β-γ spectrum sample, I represents the number of pixels in the β-γ spectrum sample, y b,i,c It represents the true probability that the i-th pixel in the b-th β-γ spectrum sample is the c-th nuclide category, p b,i,c represents the predicted probability that the pixel belongs to the cth nuclide category.

8. An AI-based β-γ coincidence spectrum analysis system, characterized in that: include: an extraction module configured to input the energy spectrum to be processed into the input convolution module of the mask prediction model for image feature extraction to obtain initial image features; wherein the energy spectrum to be processed is an analysis spectrum including energy characteristics of a specific nuclide category generated by simultaneously detecting coincident events of beta decay and accompanying gamma ray generation; A dimensionality reduction module, configured to perform multiple spatial dimensionality reductions on the initial image features through the downsampling module of the mask prediction model, and extract image features after each spatial dimensionality reduction to obtain compressed energy spectrum features of different sizes; A fusion module is used to perform spatial dimension enhancement and fusion on the compressed energy spectrum features through the upsampling module of the mask prediction model to obtain a terminal decoding feature map with the same size as the energy spectrum map to be processed; A classification module, configured to calculate the probability that each pixel in the terminal decoding feature map corresponds to each nuclide category through the output module of the mask prediction model, and generate a prediction mask map based on the probability; the prediction mask map includes the nuclide category corresponding to each pixel in the energy spectrum map to be processed; An analysis module is used to perform nuclide analysis on the energy spectrum to be processed based on the predicted mask image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the AI-based β-γ coincidence spectrum analysis method according to any one of claims 1 to 7 are implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the AI-based β-γ coincidence spectrum analysis method according to any one of claims 1 to 7 are implemented.

Images