Integrated system based on radioactive substance detection and X-ray radiation imaging

Abstract

The invention relates to the technical field of security check, and discloses an integrated system based on radioactive substance detection and X-ray radiation imaging. The system comprises an initial modeling module, a detection planning module, a signal acquisition module, a data fusion module, a collaborative analysis module and a comprehensive judgment module. The initial modeling module establishes a synchronous initial space model, the detection planning module generates a key detection subspace according to historical attenuation probability distribution of radioactive substances, the signal acquisition module captures radioactive signals and extracts an X-ray density image, and the data fusion module aligns two types of data time sequences and peels off independent signal streams. The collaborative analysis module identifies suspected core points and calculates X-ray absorption coefficient abnormal fluctuation, and the comprehensive judgment module obtains a judgment result. The system can improve the radioactive signal capturing efficiency, avoid signal interference and enhance the accuracy of radioactive substance judgment, and is suitable for the field of radioactive substance detection.

Description

Technical Field

[0001] This invention relates to the field of security inspection technology, specifically to an integrated system based on radioactive material detection and X-ray radiation imaging. Background Technology

[0002] Current methods for detecting radioactive materials and X-ray radiation imaging are mostly carried out independently or in parallel. The detection process often covers the entire area of ​​the scanned object without taking into account the attenuation probability characteristics of radioactive materials in historical monitoring for targeted deployment. This results in dispersed detection resources, insufficient focus on high-risk areas, and difficulty in quickly capturing weak or concealed radioactive signals in complex scenarios. Although X-ray imaging can present structural information, when conducted simultaneously with radioactive detection, the two types of data are often directly correlated or simply superimposed, lacking strict matching of the time dimension and decoupling of signal levels. This causes radioactive signals and structural signals to become mixed, limiting the independent and accurate analysis of radioactive enrichment locations and corresponding structural features.

[0003] In scenarios requiring simultaneous acquisition of radioactive and structural information, conventional methods, lacking the utilization of historical attenuation probability distributions to delineate key detection subspaces, are prone to overlooking probabilistic high-value regions, increasing the possibility of missed detections. Furthermore, when fusing detection signals and X-ray images, the absence of temporal alignment and multi-level information separation prevents the separation of independent radioactive signal streams from X-ray structural signal streams. This makes it difficult to directly conduct targeted local absorption coefficient fluctuation analysis within the structural signal after locating suspected radioactive core points, resulting in a lack of clear data foundation for the linked determination of the presence of radioactive material and structural anomalies. Summary of the Invention

[0004] The purpose of this invention is to provide an integrated system based on radioactive material detection and X-ray radiation imaging to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an integrated system based on radioactive material detection and X-ray radiation imaging, the system comprising:

[0006] The initial modeling module establishes an initial spatial model that synchronizes radioactive detection and X-ray imaging based on the initial outline of the scanned object.

[0007] The detection planning module generates multiple probabilistic key detection subspaces in the initial spatial model based on the historical decay probability distribution of radioactive materials.

[0008] The signal acquisition module drives the detection device to perform radioactive signal capture on the multiple probabilistic key detection subspaces, forming an original detection signal set, and synchronously extracting multiple frames of X-ray density images of the scanned object during the capture period from the X-ray imaging device;

[0009] The data fusion module performs temporal alignment between the original detection signal set and the multi-frame X-ray density images to generate a spatiotemporally aligned fused data frame, and performs multi-level information stripping on it to separate the independently layered radioactive signal stream and X-ray structural signal stream.

[0010] The collaborative analysis module identifies potential core points of radioactive material enrichment in the radioactive signal stream based on the spatial gradient changes in signal intensity, delineates the corresponding material structure analysis range around the potential core points in the X-ray structural signal stream, and calculates the local abnormal fluctuation value of the X-ray absorption coefficient within the material structure analysis range.

[0011] The comprehensive judgment module combines the signal strength of the suspected core point with the local abnormal fluctuation value to calculate the comprehensive judgment result of the presence of radioactive material.

[0012] Preferably, the step of establishing an initial spatial model for simultaneous radioactive detection and X-ray imaging based on the initial outline of the scanned object includes:

[0013] Receive external shape data of the scanned object from a 3D contour scanning device or a preset 3D model database;

[0014] The external shape data of the scanned object is subjected to surface meshing to generate uniformly distributed surface mesh nodes;

[0015] Based on the surface mesh nodes, the normal direction is extended into the interior of the scanned object to generate a multi-layered nested internal voxel space;

[0016] Each voxel in the internal voxel space is assigned a unique spatiotemporal code, which includes a spatial location index and a timestamp index.

[0017] Establish a mapping relationship between the spatiotemporal coding and the radioactivity detection channel, and simultaneously establish a mapping relationship between the spatiotemporal coding and X-ray imaging pixels;

[0018] Based on the mapping relationship, an initial spatial model with a unified index is constructed, which synchronously points to the storage addresses of the radioactivity detection data stream and the X-ray imaging data stream.

[0019] Preferably, the step of generating multiple probabilistic key detection subspaces in the initial spatial model based on the historical decay probability distribution of radioactive materials includes:

[0020] Retrieve the spatial distribution statistical characteristics of various radionuclides in similar material objects from the historical detection database;

[0021] Extract the prior probability density function of radioactive material concentration as a function of spatial location from the statistical features;

[0022] The prior probability density function is superimposed onto each voxel of the initial spatial model to obtain the initial radioactivity probability value of each voxel;

[0023] Based on a preset probability threshold, all voxels whose initial radioactivity probability value exceeds the probability threshold are selected as a high-probability voxel set.

[0024] Perform spatial clustering analysis on the set of high-probability voxels to merge multiple high-probability voxels that are spatially adjacent into a single spatially connected component;

[0025] Calculate the geometric center of each spatially connected domain, and delineate a spherical subspace in the initial spatial model with the geometric center as the center of the sphere and a preset distance as the radius;

[0026] All defined spherical subspaces are treated as multiple probabilistic key detection subspaces.

[0027] Preferably, the step of temporally aligning the original detection signal set with the multi-frame X-ray density images to generate a spatiotemporally aligned fused data frame includes:

[0028] For each signal record in the original set of detection signals, mark the precise acquisition time of the clock built into the detection device;

[0029] Mark the precise exposure time of the image generating device for each frame of the multi-frame X-ray density images;

[0030] Establish a time offset correction relationship between the built-in clock of the detection device and the system clock of the image generation equipment;

[0031] Based on the time offset correction relationship, the acquisition time of the original detection signal set is uniformly converted to the system time reference of the image generation device;

[0032] Under a unified system time reference, find the X-ray density image frame corresponding to the exposure time with the smallest time difference for each acquisition time.

[0033] Signals with a time difference less than a preset synchronization threshold are bound to their corresponding X-ray density image frames to form signal-image pairs.

[0034] All signal image pairs belonging to the same probe subspace and within a continuous time window are packaged in chronological order to generate a spatiotemporally aligned fused data frame.

[0035] Preferably, the step of performing multi-level information stripping to separate the independently layered radioactive signal stream and X-ray structural signal stream includes:

[0036] Read the radioactive pulse sequence and X-ray pixel matrix contained in the spatiotemporally aligned fused data frame;

[0037] The radioactive pulse sequence was subjected to multi-scale decomposition based on wavelet transform to separate multiple sub-band signals representing different energy components;

[0038] Independent component analysis was applied to the X-ray pixel matrix to separate multiple independent components characterizing different material components;

[0039] From the multiple sub-band signals, sub-bands with energy ranges corresponding to the characteristic energy spectrum of the target radionuclide are selected and recombined into the target radioactive signal layer;

[0040] From the multiple independent components, the component that matches the absorption characteristics of the matrix material to which the target radioactive material may be attached is selected and recombined into the target matrix structure layer.

[0041] The target radioactive signal layer is merged with other sub-band signals besides the target radioactive signal layer to form a pure radioactive signal stream;

[0042] The target matrix structure layer is combined with other independent components to form a pure X-ray structure signal stream.

[0043] Preferably, the step of identifying potential core points of radioactive material enrichment in the radioactive signal stream based on spatial gradient changes in signal intensity includes:

[0044] A three-dimensional regular grid is established within the spatial range covered by the radioactive signal stream;

[0045] Calculate the radioactive signal intensity value at each grid point in the three-dimensional regular grid to generate a three-dimensional signal intensity field;

[0046] Calculate the gradient components of the three-dimensional signal intensity field in three orthogonal directions in space;

[0047] Calculate the gradient magnitude at each grid point based on the gradient components to generate a three-dimensional gradient magnitude field.

[0048] In the three-dimensional gradient magnitude field, locate all local gradient magnitude maxima points;

[0049] Calculate the integral value of the signal strength in the neighborhood surrounding each local gradient magnitude maxima.

[0050] The local gradient magnitude maxima where the signal strength integral value exceeds a preset strength threshold are marked as candidate core points;

[0051] Non-maximum suppression is performed on the candidate core points to eliminate redundant points that are too close in space, thus obtaining the final selected suspected core points.

[0052] Preferably, the step of delineating the corresponding material structure analysis range in the X-ray structural signal stream around the suspected core point includes:

[0053] Obtain the three-dimensional coordinates of the suspected core point in the initial spatial model;

[0054] A spherical analysis space is constructed with the three-dimensional coordinates as the center and a preset initial radius;

[0055] In the X-ray structural signal stream, the structural feature values ​​of all voxels located within the spherical analysis space are extracted;

[0056] Calculate the spatial distribution variance of the structural eigenvalues;

[0057] If the spatial distribution variance is greater than a preset heterogeneity threshold, the radius of the spherical analysis space is successively expanded.

[0058] After each expansion of the radius, the difference between the mean of the structural eigenvalues ​​of the newly included voxels and the mean of the structural eigenvalues ​​of the original voxel set is recalculated.

[0059] When the difference exceeds the preset difference tolerance, the radius expansion stops, and the spherical space range at this moment is finally determined as the range of material structure analysis.

[0060] Preferably, the step of calculating the local abnormal fluctuation value of the X-ray absorption coefficient within the range of the material structure analysis includes:

[0061] Within the spherical space corresponding to the range of material structure analysis, multiple spatial sampling directions are randomly selected;

[0062] Along each spatial sampling direction, from the boundary of the spherical space to the center of the suspected core point, the X-ray absorption coefficients of a series of sampling points are collected with a fixed step size;

[0063] For each spatial sampling direction, the X-ray absorption coefficient sequence is collected, and the difference in absorption coefficient between adjacent sampling points is calculated.

[0064] The absolute values ​​of the differences in all absorption coefficients across all spatial sampling directions are statistically analyzed to form a set of differences.

[0065] Calculate the standard deviation of the set of differences, and use the standard deviation as a local abnormal fluctuation value characterizing the degree of drastic spatial variation of the X-ray absorption coefficient.

[0066] Preferably, the step of calculating the comprehensive determination result of the presence of radioactive material by combining the signal intensity of the suspected core point with the local abnormal fluctuation value includes:

[0067] Obtain the normalized radioactive signal intensity value of the suspected core point;

[0068] Obtain the normalized fluctuation amplitude value of the local abnormal fluctuation value;

[0069] Based on a pre-trained neural network model, the normalized radioactive signal intensity value and the normalized fluctuation amplitude value are used as parallel input feature vectors.

[0070] The neural network model performs nonlinear transformation and feature fusion on the input feature vector through its hidden layers;

[0071] In the output layer of the neural network model, a continuous value between zero and one is generated as the overall confidence score for the presence of radioactive material.

[0072] The overall confidence score is compared with a pre-set judgment threshold;

[0073] When the overall confidence score is greater than or equal to the judgment threshold, a positive overall judgment result on the presence of radioactive material is generated.

[0074] When the overall confidence score is less than the judgment threshold, a negative overall judgment result indicating that the radioactive material does not exist is generated.

[0075] Preferably, the system further includes:

[0076] After generating the comprehensive determination result of the presence of the radioactive material, the coordinates of the suspected core point, the range of material structure analysis, and the corresponding fusion data frame identifier associated with the comprehensive determination result are stored together.

[0077] Based on the results of multiple historically accumulated associated storages, a probability update map of the distribution of radioactive materials is constructed in the initial spatial model;

[0078] Based on the probability update map, the generation strategy of the probabilistic key detection subspace in subsequent scans is dynamically adjusted to improve detection efficiency.

[0079] Compared with the prior art, the beneficial effects of the present invention are:

[0080] Based on the historical decay probability distribution of radioactive materials, multiple probabilistic key detection subspaces are generated in the initial spatial model. This allows the detection device to prioritize signal capture in areas where radioactive materials or their concentrations are historically more likely to be present. This concentrates detection activities on potentially high-risk locations, reduces invalid sampling in low-probability areas, and makes it easier to retain effective information from real radioactive enrichment areas in the original detection signal set. In the synchronously acquired X-ray density images, the images of the corresponding time period are also more likely to cover structural states strongly correlated with radioactive signals, providing a corresponding spatial premise for subsequently locking the analysis range in structural signals.

[0081] After temporally aligning the original detection signal set with multiple frames of X-ray density images, multi-level information stripping is performed to obtain independent, layered radioactive signal streams and X-ray structural signal streams. This process preserves the intensity characteristics of the radioactive signal stream, unaffected by structural grayscale interference, allowing for the identification of suspected core points based on spatial gradient changes. Simultaneously, the X-ray structural signal stream retains pure density information, enabling the direct calculation of local anomaly fluctuations in the absorption coefficient within the structural analysis range corresponding to suspected core points. The separation of these two signal streams allows for parallel processing of radioactivity intensity discrimination and structural anomaly analysis without contamination, making the combined determination of signal intensity and structural fluctuation values ​​a more direct reflection of the likelihood of radioactive material presence. Attached Figure Description

[0082] Figure 1 This is a timing diagram of the integrated system based on radioactive material detection and X-ray radiation imaging described in this invention;

[0083] Figure 2 A flowchart for establishing the initial spatial model;

[0084] Figure 3 A flowchart for generating spatiotemporally aligned fused data frames;

[0085] Figure 4 Map showing local abnormal fluctuations in X-ray absorption coefficients at different suspected core points;

[0086] Figure 5 This is a spatial thermogram representing the probability of the presence of radioactive materials. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] Please see Figure 1This invention provides an integrated system based on radioactive material detection and X-ray radiation imaging. The system includes: an initial modeling module that establishes an initial spatial model for simultaneous radioactive detection and X-ray imaging based on the initial outline of the scanned object. This model provides a unified spatiotemporal reference system for all subsequent data processing. A detection planning module operates within this initial spatial model, generating multiple probabilistic key detection subspaces based on the historical decay probability distribution of radioactive materials. These subspaces indicate spatial regions where radioactive materials are more likely to be present. A signal acquisition module is then driven to control the detection device to capture radioactive signals within the multiple probabilistic key detection subspaces, forming a raw detection signal set. Simultaneously, multiple frames of X-ray density images of the scanned object during signal capture are extracted from the X-ray imaging device. A data fusion module temporally aligns the raw detection signal set with the multiple frames of X-ray density images, generating a spatiotemporally aligned fused data frame. This frame undergoes multi-level information stripping to separate the independently layered radioactive signal stream and X-ray structural signal stream. The collaborative analysis module identifies potential core points of radioactive material enrichment in the radioactive signal stream based on the spatial gradient changes in signal intensity. Then, it delineates a corresponding material structure analysis range around each potential core point within the X-ray structural signal stream, and calculates the local abnormal fluctuation value of the X-ray absorption coefficient within this range. The comprehensive judgment module combines the signal intensity of the potential core points with the calculated local abnormal fluctuation value, and uses a preset algorithm to calculate a comprehensive judgment result indicating the presence of radioactive material, thus completing one detection process.

[0089] In one embodiment of the present invention, see [reference] Figure 2 The initial modeling module establishes an initial spatial model for simultaneous radioactivity detection and X-ray imaging based on the initial outline of the scanned object. This step is achieved as follows: The initial modeling module receives the external shape data of the scanned object from a 3D contour scanning device or a preset 3D model database. It performs surface meshing on the external shape data of the scanned object, generating uniformly distributed surface mesh nodes. Using the surface mesh nodes as a reference, it extends the normal direction into the scanned object, generating a multi-layered nested internal voxel space. A unique spatiotemporal code is assigned to each voxel in the internal voxel space. The spatiotemporal code includes a spatial location index and a timestamp index. A mapping relationship is established between the spatiotemporal code and the radioactivity detection channel, and simultaneously, a mapping relationship is established between the spatiotemporal code and the X-ray imaging pixels. Based on the mapping relationship, a unified indexed initial spatial model is constructed. This initial spatial model synchronously points to the storage addresses of the radioactivity detection data stream and the X-ray imaging data stream.

[0090] In the initial spatial model, the detection planning module generates multiple probabilistic key detection subspaces based on the historical decay probability distribution of radioactive materials. This step is achieved as follows: The detection planning module retrieves the spatial distribution statistical characteristics of various radionuclides in similar material objects from the historical detection database. It extracts the prior probability density function of radioactive material concentration varying with spatial location from the statistical characteristics and superimposes this prior probability density function onto each voxel in the initial spatial model to obtain the initial radioactivity probability value for each voxel. Based on a preset probability threshold, all voxels with initial radioactivity probability values ​​exceeding the threshold are selected as a high-probability voxel set. Spatial clustering analysis is performed on this high-probability voxel set, merging multiple spatially adjacent high-probability voxels into a single spatially connected domain. The geometric center of each spatially connected domain is calculated, and a spherical subspace is delineated in the initial spatial model with the geometric center as the sphere center and a preset distance as the radius. All delineated spherical subspaces are used as multiple probabilistic key detection subspaces.

[0091] In practice, the initial modeling module establishes an initial spatial model for simultaneous radioactive detection and X-ray imaging based on the initial outline of the scanned object. The module receives external shape data of the scanned object from a 3D contour scanning device or a pre-set 3D model database; this data consists of point clouds or triangular meshes describing the object's outer surface. The initial modeling module performs surface meshing on the external shape data of the scanned object, using the Delaunay triangulation algorithm to generate uniformly distributed surface mesh nodes, each containing 3D coordinate information. Using the surface mesh nodes as a reference, the model extends along their normal direction into the scanned object at equal or varying intervals, generating a multi-layered nested internal voxel space. Each voxel layer maintains a topological correspondence with the surface mesh. A unique spatiotemporal code is assigned to each voxel in the internal voxel space, containing a spatial location index for localization and a timestamp index for marking the data acquisition sequence. A mapping relationship is established between spatiotemporal coding and radioactivity detection channels. The reading of each radioactivity detection channel is associated with a voxel of a specific spatiotemporal coding. Simultaneously, a mapping relationship is established between spatiotemporal coding and X-ray imaging pixels, with the grayscale value of each X-ray imaging pixel also mapped to its corresponding voxel. Based on this mapping relationship, a unified indexed initial spatial model is constructed. This initial spatial model is a database or data structure containing all voxels, their spatiotemporal codes, and mapping relationships. It synchronously points to the physical or logical addresses of the radioactivity detection data stream and the X-ray imaging data stream in the storage system.

[0092] In some embodiments, the detection planning module operates within an initial spatial model. Based on the historical decay probability distribution of radioactive materials, it generates multiple probabilistic key detection subspaces. The module retrieves the spatial distribution statistical characteristics of various radionuclides in similar material objects from a historical detection database via a data interface. From these statistical characteristics, a priori probability density function is extracted to represent the variation of radioactive material concentration with spatial location. This function describes the probability of radioactive material appearing at different internal locations given the object type and structure. The priori probability density function is superimposed onto each voxel in the initial spatial model. Specifically, the initial radioactivity probability value of each voxel is obtained by calculating the function value of the center coordinates of each voxel in the priori probability density function. Based on a preset probability threshold, all voxels with initial radioactivity probability values ​​exceeding the threshold are selected as a high-probability voxel set. Spatial clustering analysis based on three-dimensional connected component analysis is performed on this high-probability voxel set, merging multiple spatially adjacent high-probability voxels into a single spatial connected component. For each spatially connected region, the arithmetic mean of the coordinates of all constituent voxels is calculated as the geometric center. Using this geometric center as the sphere's center and a distance preset according to the object's size ratio as the radius, a spherical subspace is defined within the initial spatial model. All defined spherical subspaces are output as multiple probabilistic key detection subspaces to guide the detection path planning of the signal acquisition module.

[0093] In one embodiment of the present invention, see [reference] Figure 3 The data fusion module aligns the original detection signal set with multiple frames of X-ray density images in time sequence to generate a spatiotemporally aligned fused data frame. This step is achieved as follows: The data fusion module marks the precise acquisition time of the detector's built-in clock for each signal record in the original detection signal set, and marks the precise exposure time of the image generation device for each image frame in the multiple frames of X-ray density images, establishing a time offset correction relationship between the detector's built-in clock and the image generation device's system clock. Based on the time offset correction relationship, the acquisition times of the original detection signal set are uniformly converted to the system time reference of the image generation device. Under the unified system time reference, the X-ray density image frame corresponding to the exposure time with the smallest time difference for each acquisition time is found. Signal records with a time difference less than a preset synchronization threshold are bound to their corresponding X-ray density image frames to form signal-image pairs. All signal-image pairs belonging to the same detection subspace and within a continuous time window are packaged in chronological order to generate a spatiotemporally aligned fused data frame.

[0094] In practical implementation, the data fusion module aligns the original set of detection signals with multiple frames of X-ray density images in time, generating spatiotemporally aligned fused data frames. The data fusion module records and marks the precise acquisition time of the detector's built-in clock for each signal in the original set of detection signals. This marking process occurs in the signal processing circuit of the radioactivity detector; for each captured radioactive pulse event, the count value of the high-precision crystal oscillator clock inside the detector is recorded as the acquisition time. The data fusion module also marks the precise exposure time of the image generation device for each frame of the multiple frames of X-ray density images. This marking process occurs in the control system of the X-ray imaging device; at each completion of X-ray tube exposure and image data readout, the moment of the image generation device's master control clock is recorded as the exposure time. The data fusion module establishes a time offset correction relationship between the detector's built-in clock and the image generation device's system clock. This process involves pre-sending a network time protocol signal or hardware trigger signal to both the detector and the image generation device as a time reference point, and recording the local clock readings of both devices at this reference point, thereby calculating the fixed offset and possible small drift coefficients between the two clocks.

[0095] In some embodiments, based on the time offset correction relationship, the data fusion module uniformly converts the acquisition times of the original detection signal set to the system time reference of the image generation device. The conversion process applies a linear correction function to all acquisition times. Under the unified system time reference, the data fusion module searches for the X-ray density image frame corresponding to the exposure time with the smallest time difference for each acquisition time. This search process involves traversing the exposure time sequence of multiple X-ray density images, calculating the absolute time difference between the target acquisition time and each exposure time, and selecting the image frame corresponding to the smallest absolute time difference. The data fusion module binds signal records with time differences less than a preset synchronization threshold with their corresponding X-ray density image frames to form signal-image pairs. The preset synchronization threshold is determined comprehensively based on the signal sampling period of the detection device and the frame rate of the X-ray imaging device. The data fusion module packages all signal-image pairs belonging to the same detection subspace and within a continuous time window in chronological order to generate a spatiotemporally aligned fused data frame. The data structure of the fused data frame includes a timestamp, a spatial sub-region identifier, radioactive pulse sequence data, and the corresponding X-ray density image data block.

[0096] Optionally, the time offset correction relationship can be modeled as a linear function, where the corrected acquisition time... Calculated using the following formula:

[0097]

[0098] in: This indicates the corrected time based on the system time of the image generating device. This indicates the original acquisition time recorded by the built-in clock of the detection device. The frequency scaling factor between the two clocks is used to correct for small drifts. This represents a fixed initial offset between two clocks. This represents the additional drift compensation value calculated based on the clock drift model. This formula is used to unify all acquisition times to the same time base.

[0099] In one embodiment of the present invention, the data fusion module performs multi-level information stripping on the spatiotemporally aligned fused data frame to separate independently layered radioactive signal streams and X-ray structural signal streams. This step is achieved as follows: The data fusion module reads the radioactive pulse sequence and X-ray pixel matrix contained in the spatiotemporally aligned fused data frame. It applies multi-scale decomposition based on wavelet transform to the radioactive pulse sequence to separate multiple sub-band signals representing different energy components. It applies independent component analysis to the X-ray pixel matrix to separate multiple independent components representing different material components. From the multiple sub-band signals, sub-bands with energy ranges corresponding to the characteristic energy spectrum of the target radionuclide are selected and recombined into a target radioactive signal layer. From the multiple independent components, components matching the absorption characteristics of the matrix material to which the target radioactive material may be attached are selected and recombined into a target matrix structural layer. The target radioactive signal layer is merged with other sub-band signals to form a pure radioactive signal stream. The target matrix structural layer is merged with other independent components to form a pure X-ray structural signal stream.

[0100] The collaborative analysis module identifies potential core points of radioactive material enrichment in a radioactive signal stream based on the spatial gradient changes in signal intensity. This step is achieved as follows: Within the spatial range covered by the radioactive signal stream, the collaborative analysis module establishes a three-dimensional regular grid, calculates the radioactive signal intensity value at each grid point, generates a three-dimensional signal intensity field, and calculates the gradient components of the three-dimensional signal intensity field in three orthogonal directions. Based on the gradient components, the gradient magnitude at each grid point is calculated, generating a three-dimensional gradient magnitude field. In this field, all local gradient magnitude maxima are located, and the signal intensity integral value in the neighborhood of each local gradient magnitude maxima is calculated. Local gradient magnitude maxima whose signal intensity integral value exceeds a preset intensity threshold are marked as candidate core points. Non-maximum suppression is applied to these candidate core points to eliminate redundant points that are too close in space, resulting in the final selected potential core points.

[0101] In practical implementation, the data fusion module performs multi-level information stripping on the spatiotemporally aligned fused data frames, separating independent layered radioactive signal streams and X-ray structural signal streams. The module reads the radioactive pulse sequence and X-ray pixel matrix contained within the spatiotemporally aligned fused data frames. The data fusion module applies multi-scale decomposition based on wavelet transform to the radioactive pulse sequence, using wavelet basis functions to decompose the pulse sequence in the time-frequency domain, separating multiple sub-band signals representing different energy components. Each sub-band signal corresponds to a specific range of energy spectrum components. The data fusion module applies independent component analysis to the X-ray pixel matrix, treating the pixel matrix as an observation signal composed of a linear mixture of multiple independent source signals. Through a demixing algorithm, statistically independent source signals are estimated, separating multiple independent components representing different material components. From the multiple sub-band signals, sub-bands with energy ranges corresponding to the characteristic energy spectrum of the target radionuclide are selected. This selection process is completed by calculating the matching degree between the energy distribution of each sub-band and a pre-stored target nuclide energy spectrum library, recombining them into a target radioactive signal layer. From multiple independent components, components that match the absorption characteristics of the matrix material to which the target radioactive material may be attached are selected. The matching process is achieved by comparing the absorption curves of the independent components with the characteristic absorption curves of known matrix materials, and then recombining them into a target matrix structure layer. The target radioactive signal layer is then merged with other sub-band signals to form a pure radioactive signal stream, and the target matrix structure layer is merged with other independent components to form a pure X-ray structure signal stream.

[0102] In some embodiments, the collaborative analysis module identifies potential core points of radioactive material enrichment in the radioactive signal stream based on the spatial gradient changes in signal intensity. Within the spatial range covered by the radioactive signal stream, the collaborative analysis module establishes a three-dimensional regular grid, the resolution of which is correlated with the voxel size in the initial spatial model. The radioactive signal intensity value at each grid point in the three-dimensional regular grid is calculated to generate a three-dimensional signal intensity field. The signal intensity value at each grid point is obtained by trilinear interpolation of its neighboring radioactive signal stream data. The gradient components of the three-dimensional signal intensity field in three orthogonal directions in space are calculated using the central difference method. The gradient magnitude at each grid point is calculated based on the gradient components, generating a three-dimensional gradient magnitude field. All local gradient magnitude maxima are located in the three-dimensional gradient magnitude field by comparing the gradient magnitude values ​​of each grid point with all grid points in its immediate neighborhood. The signal intensity integral value in the neighborhood surrounding each local gradient magnitude maxima is calculated. The neighborhood is defined as a cubic region centered at the point with sides of several grid units; the integral value is the sum of the signal intensity values ​​of all grid points within that region. Local gradient magnitude maxima where the signal strength integral value exceeds a preset strength threshold are marked as candidate core points. Non-maximum suppression is performed on the candidate core points to eliminate redundant points that are too close in space, thus obtaining the final selected suspected core points.

[0103] Optional, gradient magnitude The calculation can be performed using the following formula:

[0104]

[0105] in: Indicates the location in the 3D grid index The gradient magnitude at the grid point, Represents a three-dimensional signal intensity field. , , These represent the signal intensity field at points [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... along , , The gradient component in the direction. This formula quantifies the degree of drastic change in signal strength at a point in space.

[0106] In one embodiment of the present invention, the collaborative analysis module delineates the corresponding material structure analysis range in the X-ray structural signal stream around the suspected core point. This step is achieved as follows: The collaborative analysis module obtains the three-dimensional coordinates of the suspected core point in the initial spatial model. A spherical analysis space is constructed with the three-dimensional coordinates as the center and a preset initial radius. The structural feature values ​​of all voxels located within the spherical analysis space are extracted from the X-ray structural signal stream. The spatial distribution variance of the structural feature values ​​is calculated. If the spatial distribution variance is greater than a preset heterogeneity threshold, the radius of the spherical analysis space is successively expanded. After each expansion, the difference between the mean of the structural feature values ​​of the newly included voxels and the mean of the structural feature values ​​of the original voxel set is recalculated. When the difference exceeds a preset difference tolerance, the expansion of the radius is stopped, and the spherical space range at this moment is ultimately determined as the material structure analysis range.

[0107] Within the scope of the material structure analysis, the collaborative analysis module calculates the local anomaly fluctuation values ​​of the X-ray absorption coefficient. This step is achieved as follows: Within the spherical space corresponding to the material structure analysis scope, the collaborative analysis module randomly selects multiple spatial sampling directions. Along each spatial sampling direction, from the boundary of the spherical space towards the center of the suspected core point, it collects the X-ray absorption coefficients of a series of sampling points with a fixed step size. For the X-ray absorption coefficient sequence collected in each spatial sampling direction, the absorption coefficient difference between adjacent sampling points is calculated. The absolute values ​​of all absorption coefficient differences across all spatial sampling directions are statistically analyzed to form a set of differences. The standard deviation of the set of differences is calculated and used as the local anomaly fluctuation value characterizing the severity of spatial changes in the X-ray absorption coefficient.

[0108] In practice, the collaborative analysis module delineates the corresponding material structure analysis range within the X-ray structural signal stream around the suspected core point. The module acquires the three-dimensional coordinates of the suspected core point in the initial spatial model, derived from its index position in a three-dimensional regular grid through spatial transformation. The collaborative analysis module constructs a spherical analysis space centered on these three-dimensional coordinates with a preset initial radius, which is pre-set based on the overall size of the scanned object and the required detection accuracy. Within the X-ray structural signal stream, the module extracts the structural feature values ​​of all voxels located within the spherical analysis space. These feature values ​​can be equivalent to X-ray absorption coefficients, densities, or atomic numbers. The spatial distribution variance of the structural feature values ​​is calculated using the average of the sum of squared deviations of all voxel structural feature values ​​from their mean. If the spatial distribution variance exceeds a preset heterogeneity threshold, the radius of the spherical analysis space is progressively expanded. Each expansion step can be a fixed value or a fixed proportion of the current radius. After each radius expansion, the difference between the mean of the structural eigenvalues ​​of the newly included voxels and the mean of the structural eigenvalues ​​of the original voxel set is recalculated. This difference can be quantified by calculating the absolute or relative difference between the two means. When the difference exceeds a preset tolerance, the radius expansion stops, and the spherical spatial range at this point is ultimately determined as the material structure analysis range. The material structure analysis range is a spherical region centered on the suspected core point and measured by the final radius. See Table 1.

[0109] Table 1: Example Data Table of the Process for Defining the Scope of Material Structure Analysis

[0110]

[0111] Within the scope of the material structure analysis, the collaborative analysis module calculates the local anomaly fluctuation values ​​of the X-ray absorption coefficient. Within the spherical space corresponding to the material structure analysis scope, the module randomly selects multiple spatial sampling directions, the number of which is determined based on the required statistical significance. Along each spatial sampling direction, from the boundary of the spherical space towards the center of the suspected core point, a series of X-ray absorption coefficients are collected at a fixed step size, less than or equal to the spatial resolution of the X-ray imaging system. For the X-ray absorption coefficient sequence collected in each spatial sampling direction, the absorption coefficient difference between adjacent sampling points is calculated, i.e., the absorption coefficient of the later sampling point minus the absorption coefficient of the earlier sampling point. The absolute values ​​of all absorption coefficient differences across all spatial sampling directions are statistically analyzed to form a set of differences. The standard deviation of this set of differences is calculated and used as the local anomaly fluctuation value characterizing the severity of spatial variations in the X-ray absorption coefficient.

[0112] In some embodiments, when calculating the spatial distribution variance of structural eigenvalues, the variance value The calculation formula is:

[0113]

[0114] in: The variance representing the spatial distribution of structural eigenvalues. This represents the total number of voxels within the spherical analysis space. Indicates the first Structural characteristic values ​​of individual elements Indicates all The arithmetic mean of the individual elemental structural eigenvalues. This variance is used to quantify the homogeneity of the material structure within a spherical region.

[0115] See Figure 4 This is a map showing the local anomaly fluctuations in X-ray absorption coefficients at different suspected core points. Core point D has the highest fluctuation value, while core point A has the lowest. The overall distribution of fluctuation values ​​shows significant differences, reflecting the varying degrees of spatial variation in X-ray absorption coefficients at different core points. This type of chart is commonly used in scenarios such as radioactive material detection and industrial imaging analysis. It uses local fluctuation values ​​to determine the heterogeneity of the material structure in suspected areas, aiding in the identification of potential radioactive material enrichment regions. The differences in fluctuation values ​​at different core points reflect the homogeneity of the material structure in the corresponding areas, providing data for adjusting the subsequent "range of material structure analysis." Transforming the abstract concept of "the degree of variation in X-ray absorption coefficients" into a quantifiable and visual indicator facilitates comparative analysis across scenarios and samples.

[0116] In one embodiment of the present invention, the comprehensive judgment module combines the signal intensity of the suspected core point with the local abnormal fluctuation value to calculate a comprehensive judgment result of the presence of radioactive material. This step is implemented in the following way: The comprehensive judgment module obtains the normalized radioactive signal intensity value of the suspected core point and the normalized fluctuation amplitude value of the local abnormal fluctuation value. Based on a pre-trained neural network model, the normalized radioactive signal intensity value and the normalized fluctuation amplitude value are used as parallel input feature vectors. The neural network model performs nonlinear transformation and feature fusion on the input feature vectors through its hidden layer. In the output layer of the neural network model, a continuous value between zero and one is generated as a comprehensive confidence score of the presence of radioactive material. The comprehensive confidence score is compared with a pre-set judgment threshold. When the comprehensive confidence score is greater than or equal to the judgment threshold, a positive comprehensive judgment result of the presence of radioactive material is generated. When the comprehensive confidence score is less than the judgment threshold, a negative comprehensive judgment result of the absence of radioactive material is generated.

[0117] After generating a comprehensive determination result of the presence of radioactive materials, the coordinates of suspected core points, the scope of material structure analysis, and the corresponding fusion data frame identifier associated with the comprehensive determination result are stored together. Based on the results of multiple associated storages accumulated in history, a probability update map of the distribution of radioactive materials is constructed in the initial spatial model. Based on the probability update map, the generation strategy of the probabilistic key detection subspace in subsequent scans is dynamically adjusted.

[0118] In practical implementation, the comprehensive judgment module combines the signal intensity of suspected core points with local abnormal fluctuation values ​​to calculate a comprehensive judgment result on the presence of radioactive materials. The module obtains the normalized radioactive signal intensity value of the suspected core point, which is obtained by subtracting the mean of the system's background noise from the original signal intensity at the suspected core point and then dividing by the standard deviation of the background noise. The module also obtains the normalized fluctuation amplitude value of the local abnormal fluctuation values. The normalization process involves dividing the original local abnormal fluctuation value by the baseline fluctuation value statistically obtained from a pure matrix material sample without radioactive materials. Based on a pre-trained neural network model, the normalized radioactive signal intensity value and the normalized fluctuation amplitude value are used as parallel input feature vectors into the input layer of the neural network model. The neural network model performs nonlinear transformation and feature fusion on the input feature vectors through its hidden layers, which contain multiple fully connected layers and activation functions. In the output layer of the neural network model, a continuous value between zero and one is generated by a Sigmoid activation function as the overall confidence score for the presence of radioactive material. The overall confidence score is compared with a pre-set judgment threshold. When the overall confidence score is greater than or equal to the judgment threshold, a positive overall judgment result for the presence of radioactive material is generated. When the overall confidence score is less than the judgment threshold, a negative overall judgment result for the absence of radioactive material is generated.

[0119] Optionally, a neuron in the hidden layer of the neural network model Output It can be represented as:

[0120]

[0121] in: Indicates the first The output value of each neuron This represents the activation function (such as ReLU). Indicates the connection to the input layer. The node is connected to the first hidden layer. The weight of each node, This indicates the connection to the first hidden layer. The node is connected to the next hidden layer. The weight of each node, Represents the first feature vector of the input feature vector. Each component (i.e., the normalized radioactive signal intensity value or the normalized fluctuation amplitude value). and These represent the bias terms for the corresponding neurons. This formula illustrates the transfer and nonlinear transformation process of features in a neural network.

[0122] In some embodiments, after generating a comprehensive determination result for the presence of radioactive material, the coordinates of suspected core points, the scope of material structure analysis, and the corresponding fusion data frame identifier associated with the comprehensive determination result are linked and stored in a structured detection result database. Based on the results of multiple historically accumulated linked storage, a probability update map of radioactive material distribution is constructed in the initial spatial model. The construction method involves counting the number of times each voxel in the initial spatial model is included in the scope of material structure analysis in all historical determination results, and calculating its proportion of the total number of scans as the updated probability value. Based on the probability update map, the generation strategy of probabilistic key detection subspaces in subsequent scans is dynamically adjusted to improve detection efficiency. The adjustment strategy includes resetting the probability threshold based on the updated probability value or modifying the distance parameter used to merge voxels in spatial clustering analysis.

[0123] It is understandable that the pre-training process of the neural network model uses labeled historical data, with the labels representing the known presence or absence of radioactive materials. The threshold setting needs to strike a balance between false alarm and false negative rates. The mean and standard deviation of the background noise used for normalization need to be remeasured during each system calibration cycle, and the baseline values ​​for fluctuations in pure matrix material samples need to be experimentally calibrated before the system is put into use. It is also understandable that the associated database entries include metadata such as timestamps, operator identifiers, and scanned object identifiers. The construction of the probability update map is a continuous learning process; as detection data accumulates, the map will increasingly accurately reflect the potential distribution patterns of radioactive materials in specific types of scanned objects. The dynamic adjustment strategy can be fully automatic or semi-automatic, requiring operator review and confirmation before execution.

[0124] See Figure 5 This is a spatial heatmap showing the probability of radioactive material presence. It quickly identifies high-probability areas, providing precise spatial guidance for radioactive detection and reducing ineffective detection areas. Color gradients differentiate the risk levels of different areas, supporting the differentiated allocation of subsequent detection resources. It provides a visual basis for generating "probabilistic key detection subspaces," aiding in verifying the rationality of the detection plan. This type of chart is commonly used in fields such as radioactive material detection and security imaging, visually presenting the spatial heterogeneity of material distribution and improving the efficiency and accuracy of the detection process. The high-probability core areas directly identified through color gradients avoid the ambiguity of spatial location in pure numerical analysis, allowing for more focused detection targets.

[0125] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0126] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An integrated system based on radioactive material detection and X-ray radiation imaging, characterized in that, include: The initial modeling module establishes an initial spatial model that synchronizes radioactive detection and X-ray imaging based on the initial outline of the scanned object. The detection planning module generates multiple probabilistic key detection subspaces in the initial spatial model based on the historical decay probability distribution of radioactive materials. The signal acquisition module drives the detection device to perform radioactive signal capture on the multiple probabilistic key detection subspaces, forming an original detection signal set, and synchronously extracting multiple frames of X-ray density images of the scanned object during the capture period from the X-ray imaging device; The data fusion module performs temporal alignment between the original detection signal set and the multi-frame X-ray density images to generate a spatiotemporally aligned fused data frame, and performs multi-level information stripping on it to separate the independently layered radioactive signal stream and X-ray structural signal stream. The collaborative analysis module identifies potential core points of radioactive material enrichment in the radioactive signal stream based on the spatial gradient changes in signal intensity, delineates the corresponding material structure analysis range around the potential core points in the X-ray structural signal stream, and calculates the local abnormal fluctuation value of the X-ray absorption coefficient within the material structure analysis range. The comprehensive judgment module combines the signal strength of the suspected core point with the local abnormal fluctuation value to calculate the comprehensive judgment result of the presence of radioactive material.

2. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of establishing an initial spatial model for simultaneous radioactive detection and X-ray imaging based on the initial outline of the scanned object includes: Receive external shape data of the scanned object from a 3D contour scanning device or a preset 3D model database; The external shape data of the scanned object is subjected to surface meshing to generate uniformly distributed surface mesh nodes; Based on the surface mesh nodes, the normal direction is extended into the interior of the scanned object to generate a multi-layered nested internal voxel space; Each voxel in the internal voxel space is assigned a unique spatiotemporal code, which includes a spatial location index and a timestamp index. Establish a mapping relationship between the spatiotemporal coding and the radioactivity detection channel, and simultaneously establish a mapping relationship between the spatiotemporal coding and X-ray imaging pixels; Based on the mapping relationship, an initial spatial model with a unified index is constructed, which synchronously points to the storage addresses of the radioactivity detection data stream and the X-ray imaging data stream.

3. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 2, characterized in that, The step of generating multiple probabilistic key detection subspaces in the initial spatial model based on the historical decay probability distribution of radioactive materials includes: Retrieve the spatial distribution statistical characteristics of various radionuclides in similar material objects from the historical detection database; Extract the prior probability density function of radioactive material concentration as a function of spatial location from the statistical features; The prior probability density function is superimposed onto each voxel of the initial spatial model to obtain the initial radioactivity probability value of each voxel; Based on a preset probability threshold, all voxels whose initial radioactivity probability value exceeds the probability threshold are selected as a high-probability voxel set. Perform spatial clustering analysis on the set of high-probability voxels to merge multiple high-probability voxels that are spatially adjacent into a single spatially connected component; Calculate the geometric center of each spatially connected domain, and delineate a spherical subspace in the initial spatial model with the geometric center as the center of the sphere and a preset distance as the radius; All defined spherical subspaces are treated as multiple probabilistic key detection subspaces.

4. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of temporally aligning the original detection signal set with the multi-frame X-ray density images to generate a spatiotemporally aligned fused data frame includes: For each signal record in the original set of detection signals, mark the precise acquisition time of the clock built into the detection device; Mark the precise exposure time of the image generating device for each frame of the multi-frame X-ray density images; Establish a time offset correction relationship between the built-in clock of the detection device and the system clock of the image generation equipment; Based on the time offset correction relationship, the acquisition time of the original detection signal set is uniformly converted to the system time reference of the image generation device; Under a unified system time reference, find the X-ray density image frame corresponding to the exposure time with the smallest time difference for each acquisition time. Signals with a time difference less than a preset synchronization threshold are bound to their corresponding X-ray density image frames to form signal-image pairs. All signal image pairs belonging to the same probe subspace and within a continuous time window are packaged in chronological order to generate a spatiotemporally aligned fused data frame.

5. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of performing multi-level information stripping to separate the independently layered radioactive signal stream and X-ray structural signal stream includes: Read the radioactive pulse sequence and X-ray pixel matrix contained in the spatiotemporally aligned fused data frame; The radioactive pulse sequence was subjected to multi-scale decomposition based on wavelet transform to separate multiple sub-band signals representing different energy components; Independent component analysis was applied to the X-ray pixel matrix to separate multiple independent components characterizing different material components; From the multiple subband signals, subbands with energy ranges corresponding to the characteristic energy spectrum of the target radionuclide are selected and recombined into the target radioactive signal layer; From the multiple independent components, the component that matches the absorption characteristics of the matrix material to which the target radioactive material may be attached is selected and recombined into the target matrix structure layer. The target radioactive signal layer is merged with other sub-band signals besides the target radioactive signal layer to form a pure radioactive signal stream; The target matrix structure layer is combined with other independent components to form a pure X-ray structure signal stream.

6. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of identifying potential core points of radioactive material enrichment in the radioactive signal stream based on spatial gradient changes in signal intensity includes: A three-dimensional regular grid is established within the spatial range covered by the radioactive signal stream; Calculate the radioactive signal intensity value at each grid point in the three-dimensional regular grid to generate a three-dimensional signal intensity field; Calculate the gradient components of the three-dimensional signal intensity field in three orthogonal directions in space; Calculate the gradient magnitude at each grid point based on the gradient components to generate a three-dimensional gradient magnitude field. In the three-dimensional gradient magnitude field, locate all local gradient magnitude maxima points; Calculate the integral value of the signal strength in the neighborhood surrounding each local gradient magnitude maxima. Local gradient magnitude maxima where the signal strength integral value exceeds a preset strength threshold are marked as candidate core points; Non-maximum suppression is performed on the candidate core points to eliminate redundant points that are too close in space, thus obtaining the final selected suspected core points.

7. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of delineating the corresponding material structure analysis range in the X-ray structural signal stream around the suspected core point includes: Obtain the three-dimensional coordinates of the suspected core point in the initial spatial model; A spherical analysis space is constructed with the three-dimensional coordinates as the center and a preset initial radius; In the X-ray structural signal stream, the structural feature values ​​of all voxels located within the spherical analysis space are extracted; Calculate the spatial distribution variance of the structural eigenvalues; If the spatial distribution variance is greater than a preset heterogeneity threshold, the radius of the spherical analysis space is successively expanded. After each expansion of the radius, the difference between the mean of the structural eigenvalues ​​of the newly included voxels and the mean of the structural eigenvalues ​​of the original voxel set is recalculated. When the difference exceeds the preset difference tolerance, the radius expansion stops, and the spherical space range at this moment is finally determined as the range of material structure analysis.

8. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 7, characterized in that, The step of calculating the local abnormal fluctuation value of the X-ray absorption coefficient within the scope of the material structure analysis includes: Within the spherical space corresponding to the range of material structure analysis, multiple spatial sampling directions are randomly selected; Along each spatial sampling direction, from the boundary of the spherical space to the center of the suspected core point, the X-ray absorption coefficients of a series of sampling points are collected with a fixed step size; For each spatial sampling direction, the X-ray absorption coefficient sequence is collected, and the difference in absorption coefficient between adjacent sampling points is calculated. The absolute values ​​of the differences in all absorption coefficients across all spatial sampling directions are statistically analyzed to form a set of differences. Calculate the standard deviation of the set of differences, and use the standard deviation as a local abnormal fluctuation value characterizing the degree of drastic spatial variation of the X-ray absorption coefficient.

9. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 1, characterized in that, The step of calculating a comprehensive determination result of the presence of radioactive material by combining the signal intensity of the suspected core point with the local abnormal fluctuation value includes: Obtain the normalized radioactive signal intensity value of the suspected core point; Obtain the normalized fluctuation amplitude value of the local abnormal fluctuation value; Based on a pre-trained neural network model, the normalized radioactive signal intensity value and the normalized fluctuation amplitude value are used as parallel input feature vectors. The neural network model performs nonlinear transformation and feature fusion on the input feature vector through its hidden layers; In the output layer of the neural network model, a continuous value between zero and one is generated as the overall confidence score for the presence of radioactive material. The overall confidence score is compared with a pre-set judgment threshold; When the overall confidence score is greater than or equal to the judgment threshold, a positive overall judgment result on the presence of radioactive material is generated. When the overall confidence score is less than the judgment threshold, a negative overall judgment result indicating that the radioactive material does not exist is generated.

10. The integrated system based on radioactive material detection and X-ray radiation imaging according to claim 9, characterized in that, Also includes: After generating the comprehensive determination result of the presence of the radioactive material, the coordinates of the suspected core point, the range of material structure analysis, and the corresponding fusion data frame identifier associated with the comprehensive determination result are stored together. Based on the results of multiple historically accumulated associated storages, a probability update map of the distribution of radioactive materials is constructed in the initial spatial model; Based on the probability update map, the generation strategy of the probabilistic key detection subspace in subsequent scans is dynamically adjusted to improve detection efficiency.

Images