A multi-level visualization modeling method based on semantics for radiation field

Abstract

The application discloses a kind of multi-level visualization modeling methods of radiation field based on semantics, belong to nuclear radiation detection and visualization technical field, the steps of method are as follows: three-dimensional geometric network model is constructed to target environment, and sparse measurement point set is constructed;Three-dimensional geometric network model is carried out voxel discretization processing, and material attribute is set for each voxel respectively, and prior generation and posterior estimation are carried out, obtain the posterior estimation value of voxel level radiation intensity and the variation coefficient of the uncertainty measurement of each voxel;Three-dimensional geometric network model is carried out instance segmentation and semantic classification, generates environment structure semantic label, and generates physical semantic label and statistical semantic label according to radiation transmission physical mechanism;HSV color space uniform mapping mechanism is constructed, and three-dimensional radiation field model is rendered and drawn.The present application solves the problem that existing radiation field information expression is fragmented and uncertainty expression is difficult.

Description

Technical Field

[0001] This invention belongs to the field of nuclear radiation detection and visualization technology, and in particular relates to a semantic-based multi-level visualization modeling method for radiation fields. Background Technology

[0002] In nuclear facility operation, decommissioning, and emergency scenarios, accurately acquiring and understanding the spatial distribution of the radiation field is fundamental for operational protection and decision-making. Traditional radiation mapping primarily uses portable detectors for point-by-point measurements combined with two-dimensional interpolation methods to generate dose rate distribution maps, but this approach has significant drawbacks: firstly, the spatial coverage of discrete measurement points is low, making it difficult to reflect changes in the radiation field gradient near small-scale structures; secondly, two-dimensional representations cannot describe vertical distribution differences and the physical modulation effects of shielding structures, pipes, and other structures on radiation transport. With the development of mobile measurement platforms and multimodal sensors, existing technologies have attempted to construct three-dimensional radiation field visualization schemes. Typical technical paths include: acquiring a three-dimensional voxel mesh of the environment based on lidar or visual reconstruction, performing forward physical simulations using the radiation transport equations, and then fusing measured data through Bayesian inference or iterative inversion algorithms to obtain an estimate of the three-dimensional radiation field and its uncertainty measure. However, the aforementioned existing technologies still have the following shortcomings in terms of visualization and interpretability: (1) The visual expression of multi-source information fusion is fragmented: Existing solutions usually overlay the radiation estimation results on the surface of the three-dimensional geometric model as pseudo-color cloud maps. There is no semantic relationship between the geometric structure and the radiation physical quantity, making it difficult to intuitively judge the causal correspondence between dose anomalies and specific components, such as shielding walls and pipe penetrations. (2) The spatial solvability of uncertainty information is poor: The uncertainty measures such as posterior variance and confidence interval given by Bayesian inference are mostly presented in the form of numerical values ​​or independent statistical graphs, which are disconnected from the three-dimensional spatial location. This makes it difficult to quickly identify whether a high radiation area corresponds to a high estimation error, and it is also difficult to trace the source of uncertainty. (3) There are obstacles to the cognitive transformation of physical and statistical models: The scattering attenuation mechanism described by the radiation transport equation and the prior and error propagation process in Bayesian inference are highly professional, making it difficult to understand their spatial semantic connotation through traditional numerical charts. Summary of the Invention

[0003] To address the aforementioned shortcomings in existing technologies, this invention provides a semantic-based multi-level visualization modeling method for radiation fields, which solves the problems of fragmented information representation and difficulty in expressing uncertainty in existing radiation field methods.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: This invention provides a semantic-based multi-level visualization modeling method for radiation fields, comprising the following steps: S1. Construct a three-dimensional geometric network model of the target environment and build a sparse measurement point set based on several spatial radiation data; S2. The three-dimensional geometric network model is discretized into voxels, and material properties are set for each voxel. Prior generation and posterior estimation are performed based on the sparse measurement point set to obtain the posterior estimate of voxel-level radiation intensity and the coefficient of variation of uncertainty measure for each voxel. S3. Perform instance segmentation and semantic classification on the three-dimensional geometric network model, generate environmental structure semantic labels, and generate physical semantic labels and statistical semantic labels based on the radiative transfer physical mechanism and the coefficient of variation of the uncertainty measure of each voxel. S4. Based on the environmental structural semantic tags, physical semantic tags, and statistical semantic tags, construct a unified mapping mechanism for the HSV color space, and render a three-dimensional radiation field model based on the posterior estimate of voxel-level radiation intensity and the coefficient of variation of the uncertainty measure of each voxel.

[0005] Further, step S1 includes the following steps: S11. Using a mobile measurement platform equipped with lidar, a visual camera and a radiation detector, move within the target environment along a preset path and simultaneously collect radar point cloud data, visual image data and radiation measurement data at the corresponding spatial locations. S12. Based on the real-time positioning and mapping method, the radar point cloud data and visual image data of each spatial location are integrated to reconstruct a three-dimensional geometric network model of the target environment. S13. Based on the pose at each spatial location and the corresponding radiation measurement data, form spatial radiation data pairs and construct a sparse measurement point set.

[0006] Further, step S2 includes the following steps: S21. Discretize the three-dimensional geometric network model into several individual elements; S22. Set material properties for each voxel according to the environmental materials, and use the discrete ordinate method to solve the steady-state Boltzmann transport equation to obtain the radiation detector response matrix and the prior distribution of the radiation field. The calculation expression for the steady-state Boltzmann transport equation is as follows: , in, This represents the unit vector indicating the direction of particle motion. The three-dimensional spatial gradient representing the particle angular flux density. Represents the three-dimensional spatial position vector of a particle. Represents particle energy. Indicates the linear attenuation coefficient of the medium bus. Represents particle angular flux density. Represents the energy of the particle before scattering. This represents the unit vector indicating the direction of particle motion before scattering. Represents the differential scattering coefficient. Represents the energy of the particle before scattering. Indicates a radiation source; S23. Based on the sparse measurement point set, the radiation detector response matrix, and the prior distribution of the radiation field, let the radiation measurement data follow a Gaussian distribution and calculate the likelihood probability density. The expression for calculating the likelihood probability density is as follows: , in, Represents the likelihood probability density. Represents a vector of radiation measurement data. This represents the voxel-level radiation intensity vector. This indicates the total number of radiation measurement points. Represents the measurement noise covariance matrix The determinant, Indicated by It is an exponential function of the basis constants. Indicates transpose. Represents the measurement noise covariance matrix The inverse matrix, Represents the response matrix of the radiation detector; S24. The spatial smoothness of the prior distribution of the radiation field is constrained by Markov random fields, and the posterior distribution of the radiation field is calculated according to Bayes' theorem. The calculation expression for the posterior distribution of the radiation field is as follows: , in, This represents the posterior probability density of the radiation field intensity under given measurement data conditions. Indicates proportional to, This represents the prior probability density of the radiation intensity vector; S25. By maximizing the posterior probability of the radiation field or based on the variational Bayesian method, the posterior estimate of the voxel-level radiation intensity and the posterior covariance matrix are calculated. S26. Based on the posterior estimate of voxel-level radiation intensity and the posterior covariance matrix, extract the coefficient of variation of the uncertainty measure for each voxel. The expression for calculating the coefficient of variation of the uncertainty measure for each voxel is as follows: , , in, Indicates the first The coefficient of variation, a measure of uncertainty in individual elements. Indicates the first The standard deviation of the uncertainty measure of individual elements. Indicates the first Posterior estimate of the radiation intensity of an individual element. Represents the posterior covariance matrix The first on the main diagonal Each element.

[0007] Further, step S3 includes the following steps: S31. Use a deep learning model to perform instance segmentation and recognition on a 3D geometric network model to obtain several semantic objects, and label the corresponding semantic objects as environmental structure semantic tags. S32. Define the semantics of the physical mechanism of radiative transfer, and identify and mark the shadow region and the scattering-dominant region as physical semantic labels; S33. Mark regions where the coefficient of variation of the uncertainty measure of voxels is greater than a preset coefficient of variation threshold as high uncertainty regions, and use them as statistical semantic labels.

[0008] Furthermore, the semantics of the radiative transfer physical mechanism include attenuation semantics and scattering semantics; The attenuation semantics are as follows: if a region is identified where the angle between the direction of the radiation field intensity gradient and the normal vector of the occluder is greater than 90°, then the region is marked as a shadow region. The scattering semantics are as follows: if a region is identified where the scattering contribution ratio of the radiation detector response is greater than a preset scattering contribution ratio threshold, then that region is marked as a scattering-dominant region.

[0009] Further, step S4 includes the following steps: S41. Based on environmental structure semantic tags, physical semantic tags and statistical semantic tags, construct a unified mapping mechanism for the HSV color space. The unified mapping mechanism for HSV color includes hue mapping based on radiation intensity, saturation mapping based on uncertainty, and brightness mapping based on semantic fusion. S42. Based on the unified mapping mechanism of HSV color space, and using the posterior estimate of voxel-level radiation intensity and the coefficient of variation of uncertainty measure for each voxel, a three-dimensional radiation field model including color, transparency and illumination is drawn in real time within the three-dimensional geometric network model using a graphics rendering engine.

[0010] Furthermore, the calculation expression for the hue mapping in S41 is as follows: , in, Representing three-dimensional spatial position The hue value at that location, Indicates the lower limit of the hue value. Indicates the upper limit of the hue value. Represents the logarithmic function. Representing three-dimensional spatial position The posterior estimate of the radiation intensity of the corresponding voxel. This represents the minimum radiation intensity within the target environment. This indicates the maximum radiation intensity within the target environment.

[0011] Furthermore, the calculation expression for the saturation mapping in S41 is as follows: , in, Representing three-dimensional spatial position The saturation value at that point, This indicates the upper limit of the saturation value. This represents the sigmoid function. This represents the saturation adjustment coefficient. Representing three-dimensional spatial position The coefficient of variation, representing the uncertainty measure of the corresponding voxel. This indicates that the threshold for the coefficient of variation is set.

[0012] Furthermore, the calculation expression for the brightness mapping in S41 is as follows: , in, Representing three-dimensional spatial position The brightness value at that location, The weighting coefficients represent the semantic weighting of the environmental structure. This indicates a semantic indicator function for the environment structure. Represents the weighting coefficients for physical semantics. Represents a physical semantic indicator function. Represents the weighting coefficients for statistical semantics. This represents a statistical semantic indicator function.

[0013] The beneficial effects of this invention are as follows: This invention provides a semantic-based multi-level visualization modeling method for radiation fields. By mapping physical quantities, geometric structures, and statistical uncertainties to a unified HSV color space, it enables direct understanding of multi-dimensional information from color, avoiding the cognitive burden of comparing multiple independent charts in traditional methods, and the fragmented visualization of radiation field information, environmental geometry, and uncertainty measures. This invention directly encodes uncertainty measures such as posterior variance and confidence intervals into a three-dimensional visualization model, enabling immediate identification of whether high-radiation areas are accompanied by high estimation errors and intuitive tracing of uncertainty sources, avoiding the disconnect between uncertainty information and three-dimensional spatial location in existing technologies. This invention generates three layers of semantic labels—environmental structure semantics, physical semantics, and statistical semantics—and transforms them into intuitive visual cues, such as semi-transparent display of occluded areas, effectively helping non-professional users understand the scattering attenuation mechanism described by the radiation transport equation and the error propagation process in Bayesian inference, thus improving the efficiency of decision support. This invention, through the fusion of physical-driven and statistical inversion, significantly improves reconstruction accuracy and the reliability of uncertainty information compared to pure physical models or pure data-driven methods.

[0014] Other advantages of the present invention will be analyzed in more detail in the following embodiments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the steps of a semantic-based multi-level visualization modeling method for radiation fields in an embodiment of the present invention. Detailed Implementation

[0017] 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a semantic-based multi-level visualization modeling method for radiation fields, comprising the following steps: S1. Construct a three-dimensional geometric network model of the target environment and build a sparse measurement point set based on several spatial radiation data; S1 includes the following steps: S11. Using a mobile measurement platform equipped with lidar, a visual camera and a radiation detector, move within the target environment along a preset path and simultaneously collect radar point cloud data, visual image data and radiation measurement data at the corresponding spatial locations. In this scheme, the radiation detector is a gamma spectrometer or dose rate meter, and the mobile measurement platform is a drone or robot. In this embodiment, a rotary-wing UAV platform equipped with a Riegl VUX-1 lidar, a FLIR Blackfly S-type industrial camera, and a CeBr3 scintillator detector is used as the mobile measurement platform; the target environment is a hot chamber; before measurement, an "S"-shaped trajectory is planned based on a known rough geometric model of the hot chamber to balance the scanning coverage and sampling point density; during the measurement process, the mobile measurement platform moves at a constant speed of 1.5 m / s for a total of 1200 seconds, obtaining radar point cloud data, visual image data, and radiation measurement data at each spatial location; S12. Based on the real-time positioning and mapping method, the radar point cloud data and visual image data of each spatial location are integrated to reconstruct a three-dimensional geometric network model of the target environment. In this scheme, a tightly coupled laser-vision-inertial odometry framework is used to process radar point cloud data and visual image data at various spatial locations to generate a high-precision three-dimensional geometric network model of the target environment. S13. Based on the pose at each spatial location and the corresponding radiation measurement data, form spatial radiation data pairs and construct a sparse measurement point set.

[0019] In this scheme, the radiation detector synchronously collects dose rate data at a frequency of 2Hz as radiation measurement data, and combines the pose calculated by SLAM at each spatial location with the corresponding radiation measurement data to obtain 1200 pairs of spatial radiation data, corresponding to 1200 effective measurement points, forming a sparse measurement point set.

[0020] S2. The three-dimensional geometric network model is discretized into voxels, and material properties are set for each voxel. Prior generation and posterior estimation are performed based on the sparse measurement point set to obtain the posterior estimate of voxel-level radiation intensity and the coefficient of variation of uncertainty measure for each voxel. S2 includes the following steps: S21. Discretize the three-dimensional geometric network model into several individual elements; In this design, the side length of each voxel is 0.05m, and the total length is approximately [missing information]. One effective volume element; S22. Based on the material properties of each voxel according to the environmental material, and using the discrete ordinate method to solve the steady-state Boltzmann transport equation, the radiation detector response matrix and the prior distribution of the radiation field are obtained. In this scheme, environmental materials include concrete, iron, air, etc.; material properties include density, atomic number, and attenuation coefficient; the Monte Carlo method can also be used to solve the steady-state Boltzmann transport equation to obtain the radiation detector response matrix and the prior distribution of the radiation field. Based on the material segmentation results, a linear attenuation coefficient is assigned to each voxel: air is set as... Concrete is set as Stainless steel is set as The primary source terms are set as Co-60 point sources, located within a known source container geometry, with primary source term energies of 1.17 MeV and 1.33 MeV.

[0021] The calculation expression for the steady-state Boltzmann transport equation is as follows: , in, This represents the unit vector indicating the direction of particle motion. The three-dimensional spatial gradient representing the particle angular flux density. Represents the three-dimensional spatial position vector of a particle. Represents particle energy. Indicates the linear attenuation coefficient of the medium bus. Represents particle angular flux density. Represents the energy of the particle before scattering. This represents the unit vector indicating the direction of particle motion before scattering. Represents the differential scattering coefficient. Represents the energy of the particle before scattering. Indicates a radiation source; In this plan, For position ,energy ,direction Particle angular flux density at that location, For position ,energy The linear attenuation coefficient of the medium bus. To obtain energy ,direction Scattered energy ,direction The differential scattering coefficient, For position ,energy ,direction The radiation source at that location; In this embodiment, during the forward modeling stage, the deterministic discrete ordinate method is used to solve the Boltzmann transport equation, and the resulting detector response matrix quantitatively describes the contribution of each voxel unit source strength to the detector response at each measurement point. S23. Based on the sparse measurement point set, the radiation detector response matrix, and the prior distribution of the radiation field, let the radiation measurement data follow a Gaussian distribution and calculate the likelihood probability density. The expression for calculating the likelihood probability density is as follows: , in, Represents the likelihood probability density. Represents a vector of radiation measurement data. This represents the voxel-level radiation intensity vector. This indicates the total number of radiation measurement points. Represents the measurement noise covariance matrix The determinant, Indicated by It is an exponential function of the basis constants. Indicates transpose. Represents the measurement noise covariance matrix The inverse matrix, Represents the response matrix of the radiation detector; In this scheme, the likelihood probability density Given a voxel-level radiation intensity vector Under certain conditions, the probability density of the radiation measurement data vector acquired by the radiation detector can reflect the degree of agreement between the model's predicted response and the measured radiation data; voxel-level radiation intensity vector The radiation intensity values ​​corresponding to all voxels after the target environment is discretized are arranged in order to characterize the distribution of the radiation field in the entire three-dimensional space. This is the residual vector between the measured data and the theoretical response data of the radiation detector, representing the deviation between the measured value and the calculated value from the model. In this embodiment, the likelihood function is set to a Gaussian distribution in the statistical inversion stage, and the standard deviation of the measurement noise is 5% of the measured value. S24. The spatial smoothness of the prior distribution of the radiation field is constrained by Markov random fields, and the posterior distribution of the radiation field is calculated according to Bayes' theorem. The calculation expression for the posterior distribution of the radiation field is as follows: , in, This represents the posterior probability density of the radiation field intensity under given measurement data conditions. Indicates proportional to, This represents the prior probability density of the radiation intensity vector; In this plan, The statistical confidence level characterizing the distribution of the radiation field after fusing measured information; Characterize the confidence level of the initial distribution of the radiation field obtained based on the physical laws of radiative transfer; In this embodiment, the spatial prior adopts a Gaussian Markov random field model to encourage the continuity of radiation intensity of adjacent voxels. S25. By maximizing the posterior probability of the radiation field or based on the variational Bayesian method, the posterior estimate of the voxel-level radiation intensity and the posterior covariance matrix are calculated. The formula for calculating the posterior estimate of the voxel-level radiation intensity is as follows: , in, This represents the posterior estimate of voxel-level radiation intensity. This represents the posterior estimate of the radiation intensity of the first voxel. This represents the posterior estimate of the radiation intensity of the second voxel. Indicates the first Posterior estimate of the radiation intensity of an individual element; S26. Based on the posterior estimate of voxel-level radiation intensity and the posterior covariance matrix, extract the coefficient of variation of the uncertainty measure for each voxel. The expression for calculating the coefficient of variation of the uncertainty measure for each voxel is as follows: , , in, Indicates the first The coefficient of variation, a measure of uncertainty in individual elements. Indicates the first The standard deviation of the uncertainty measure of individual elements. Indicates the first Posterior estimate of the radiation intensity of an individual element. Represents the posterior covariance matrix The first on the main diagonal Each element. In this scheme, For the first Variance of the posterior estimate of individual element radiation intensity.

[0022] In this embodiment, prior generation and posterior estimation experiments based on sparse measurement point sets were conducted in the obstructed area behind a stainless steel pipe. The experiments show that the prior estimate relying solely on the physical model is 0.5 mSv / h, while the posterior estimate is corrected to 0.7 mSv / h after fusing sparse measurement data. At the same time, the posterior uncertainty of this voxel is significantly increased, that is, the coefficient of variation of the uncertainty measure of the voxel reaches 0.4.

[0023] S3. Perform instance segmentation and semantic classification on the three-dimensional geometric network model, generate environmental structure semantic labels, and generate physical semantic labels and statistical semantic labels based on the radiative transfer physical mechanism and the coefficient of variation of the uncertainty measure of each voxel. S3 includes the following steps: S31. Use a deep learning model to perform instance segmentation and recognition on a 3D geometric network model to obtain several semantic objects, and label the corresponding semantic objects as environmental structure semantic tags. In this embodiment, the deep learning model uses a pre-trained PointNet++ model; the identified semantic objects include walls, pipes, containers, etc., and the corresponding output is environmental structure semantic labels such as concrete wall, stainless steel pipe, source container, etc. S32. Define the semantics of the physical mechanism of radiative transfer, and identify and mark the shadow region and the scattering-dominant region as physical semantic labels; The semantics of the physical mechanism of radiative transfer include attenuation semantics and scattering semantics; The attenuation semantics are as follows: if a region is identified where the angle between the direction of the radiation field intensity gradient and the normal vector of the occluder is greater than 90°, then the region is marked as a shadow region. The scattering semantics are as follows: if a region is identified where the scattering contribution ratio of the radiation detector response is greater than a preset scattering contribution ratio threshold, then that region is marked as a scattering-dominant region.

[0024] In this scheme, the scattering contribution ratio is the proportion of the radiation detector response caused by scattered photons to the total radiation detector response at that location; the radiation detector response matrix is... Decomposed into direct response matrix and scattering response matrix And satisfy For three-dimensional spatial position The formula for calculating the scattering contribution ratio at the corresponding measurement point is as follows: , , in, Representing three-dimensional spatial position The scattering contribution ratio at the corresponding measurement point, Indicates the first The radiation emitted by an individual element undergoes at least one scattering before affecting the first element. The contribution coefficient of each measurement point to the response of the radiation detector. Indicates the first Posterior estimate of the radiation intensity of an individual element. Indicates the first The radiation emitted by the individual element reaches the first unscattered point. The contribution coefficient of each measurement point to the response of the radiation detector. Represents the elements of the overall detector response matrix; In this embodiment, the preset scattering contribution ratio threshold is preferably set to 0.5; if the calculated scattering contribution ratio If the ratio is greater than the preset scattering contribution ratio threshold, it is marked as a scattering-dominant region.

[0025] S33. Mark regions where the coefficient of variation of the uncertainty measure of voxels is greater than a preset coefficient of variation threshold as high uncertainty regions, and use them as statistical semantic labels.

[0026] S4. Based on the environmental structural semantic tags, physical semantic tags, and statistical semantic tags, construct a unified mapping mechanism for the HSV color space, and render a three-dimensional radiation field model based on the posterior estimate of voxel-level radiation intensity and the coefficient of variation of the uncertainty measure of each voxel.

[0027] S4 includes the following steps: S41. Based on environmental structure semantic tags, physical semantic tags and statistical semantic tags, construct a unified mapping mechanism for the HSV color space. The unified mapping mechanism for HSV color includes hue mapping based on radiation intensity, saturation mapping based on uncertainty, and brightness mapping based on semantic fusion. The calculation expression for the hue mapping in S41 is as follows: , in, Representing three-dimensional spatial position The hue value at that location, Indicates the lower limit of the hue value. Indicates the upper limit of the hue value. Represents the logarithmic function. Representing three-dimensional spatial position The posterior estimate of the radiation intensity of the corresponding voxel. This represents the minimum radiation intensity within the target environment. This indicates the maximum radiation intensity within the target environment.

[0028] In this embodiment, hue mapping refers to linearly mapping the estimated posterior dose rate values ​​within a preset posterior dose range to a hue range of 0° to 240° after logarithmic transformation, in order to clearly distinguish between high-dose and low-dose regions. The preset posterior dose range is... to 0° corresponds to red, and 240° corresponds to blue; The calculation expression for the saturation mapping in S41 is as follows: , in, Representing three-dimensional spatial position The saturation value at that point, This indicates the upper limit of the saturation value. This represents the sigmoid function. This represents the saturation adjustment coefficient. Representing three-dimensional spatial position The coefficient of variation, representing the uncertainty measure of the corresponding voxel. This indicates that the threshold for the coefficient of variation is set.

[0029] In this embodiment, saturation mapping refers to inputting the coefficient of variation of the uncertainty measure of voxels into the sigmoid function to adjust the saturation. The higher the coefficient of variation of the uncertainty measure of voxels, the lower the saturation and the more the color tends to be grayish-white, so as to intuitively convey the uncertainty of the result and indicate that the estimation result of this region is unreliable. The calculation expression for the brightness mapping in S41 is as follows: , in, Representing three-dimensional spatial position The brightness value at that location, The weighting coefficients represent the semantic weighting of the environmental structure. This indicates a semantic indicator function for the environment structure. Represents the weighting coefficients for physical semantics. Represents a physical semantic indicator function. Represents the weighting coefficients for statistical semantics. This represents a statistical semantic indicator function.

[0030] In this embodiment, for example, when displaying physical semantics, the transparency of voxels in the area marked as a shadow area is increased to 0.5, making them semi-transparent, which makes it easier to observe the structural relationships behind the occluded objects. S42. Based on the unified mapping mechanism of HSV color space, and using the posterior estimate of voxel-level radiation intensity and the coefficient of variation of uncertainty measure for each voxel, a three-dimensional radiation field model including color, transparency and illumination is drawn in real time within the three-dimensional geometric network model using a graphics rendering engine.

[0031] The three-dimensional radiation field model constructed by this scheme can achieve multi-level visualization. When compared with the traditional pseudo-color cloud map that only uses hue to represent dose, the visualization model of this scheme showed a significant advantage in expert interpretation tests. Operators can instantly identify areas with high uncertainty caused by data sparsity or physical occlusion based on the saturation and transparency of the colors, thereby avoiding over-reliance on the dose estimate of that area and providing direct visual basis for path planning and supplementary measurements.

[0032] In a practical example of the present invention, a controlled simulation experimental environment was constructed to quantitatively evaluate the reconstruction accuracy and uncertainty calibration of the present scheme under known truth conditions.

[0033] In this embodiment, a virtual scene with geometric dimensions of 10m × 10m × 5m is constructed using the Monte Carlo particle transport procedure. The scene contains the following elements: a 0.5m thick concrete wall, a 0.4m diameter stainless steel pipe, and an isotropic ¹³ 7 Cs gamma source (energy 0.662 MeV); through large-scale particle simulation, a high-fidelity radiation dose rate field at 0.1 m voxel resolution was generated for the entire space as the true value for evaluation; In this embodiment, to verify the advancement of the method of the present invention, the following two methods are selected for comparison: Comparison Method A is a pure physical model: only the forward transport calculation in step S2 is performed, and the prior distribution is used as the final visualization basis; Comparison Method B is Kriging interpolation: ignoring the geometric structure and physical model, ordinary Kriging interpolation is directly performed on the 1200 simulated measurement points, and the simulated measurement data is generated by superimposing 5% Gaussian noise at the corresponding position in the true field.

[0034] In this embodiment, normalized root mean square error (RMSE) and uncertainty calibration degree are used as indicators for quantitative analysis. RMSE is the normalized RMSE between the reconstructed field and the true field. Uncertainty calibration degree assesses the coverage of the posterior confidence interval to the true error. If the uncertainty calibration is accurate, the proportion of voxels whose true values ​​fall within the 95% confidence interval should be close to 95%. The indicator analysis results are shown in Table 1. Table 1 Experimental results show that the method of this invention significantly outperforms the two comparative methods in terms of normalized root mean square error (RMSE). Comparative method A, by ignoring actual measurement information, exhibits a systematic bias in source term intensity estimation. Comparative method B, by failing to consider the shielding and scattering effects of geometric structures on the radiation field, produces an erroneous estimate in the region behind the pipe due to excessive smoothing. This proposed solution, by fusing physical priors and observational data through a Bayesian framework, effectively suppresses measurement noise and corrects model bias. The proportion of true values ​​falling within the 95% confidence interval is 93.8%, very close to the theoretical value of 95%, demonstrating the reliability of the provided uncertainty information. In contrast, the other two methods fail to provide any information regarding the credibility of the estimation results.

[0035] Through real-world scenario verification and quantitative simulation experiments, this scheme demonstrates its ability to achieve high-precision radiation field reconstruction and provide accurate quantitative uncertainty measurement for the reconstruction results. Combined with a multi-level semantic HSV color space mapping mechanism, this scheme successfully transforms complex physical inversion results into an intuitive and interpretable visualization model, providing strong technical support for safety decision-making in nuclear radiation environments.

[0036] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A semantic-based multi-level visualization modeling method for radiation fields, characterized in that, Includes the following steps: S1. Construct a three-dimensional geometric network model of the target environment and build a sparse measurement point set based on several spatial radiation data; S2. The three-dimensional geometric network model is discretized into voxels, and material properties are set for each voxel. Prior generation and posterior estimation are performed based on the sparse measurement point set to obtain the posterior estimate of voxel-level radiation intensity and the coefficient of variation of uncertainty measure for each voxel. S3. Perform instance segmentation and semantic classification on the three-dimensional geometric network model, generate environmental structure semantic labels, and generate physical semantic labels and statistical semantic labels based on the radiative transfer physical mechanism and the coefficient of variation of the uncertainty measure of each voxel. S4. Based on the environmental structural semantic tags, physical semantic tags, and statistical semantic tags, construct a unified mapping mechanism for the HSV color space, and render a three-dimensional radiation field model based on the posterior estimate of voxel-level radiation intensity and the coefficient of variation of the uncertainty measure of each voxel.

2. The semantic-based multi-level visualization modeling method for radiation fields according to claim 1, characterized in that, S1 includes the following steps: S11. Using a mobile measurement platform equipped with lidar, a visual camera and a radiation detector, move within the target environment along a preset path and simultaneously collect radar point cloud data, visual image data and radiation measurement data at the corresponding spatial locations. S12. Based on the real-time positioning and mapping method, the radar point cloud data and visual image data of each spatial location are integrated to reconstruct a three-dimensional geometric network model of the target environment. S13. Based on the pose at each spatial location and the corresponding radiation measurement data, form spatial radiation data pairs and construct a sparse measurement point set.

3. The semantic-based multi-level visualization modeling method for radiation fields according to claim 2, characterized in that, S2 includes the following steps: S21. Discretize the three-dimensional geometric network model into several individual elements; S22. Set material properties for each voxel according to the environmental materials, and use the discrete ordinate method to solve the steady-state Boltzmann transport equation to obtain the radiation detector response matrix and the prior distribution of the radiation field. The calculation expression for the steady-state Boltzmann transport equation is as follows: ， in, The unit vector representing the direction of particle motion. The three-dimensional spatial gradient representing the particle angular flux density. Represents the three-dimensional spatial position vector of a particle. Represents particle energy. Indicates the linear attenuation coefficient of the medium bus. Represents particle angular flux density. Represents the energy of the particle before scattering. This represents the unit vector indicating the direction of particle motion before scattering. Represents the differential scattering coefficient. Represents the energy of the particle before scattering. Indicates a radiation source; S23. Based on the sparse measurement point set, the radiation detector response matrix, and the prior distribution of the radiation field, let the radiation measurement data follow a Gaussian distribution and calculate the likelihood probability density. The expression for calculating the likelihood probability density is as follows: ， in, Represents the likelihood probability density. Represents a vector of radiation measurement data. This represents the voxel-level radiation intensity vector. This indicates the total number of radiation measurement points. Represents the measurement noise covariance matrix The determinant, Indicates An exponential function of the basis constants. Indicates transpose. Represents the measurement noise covariance matrix The inverse matrix, Represents the response matrix of the radiation detector; S24. The spatial smoothness of the prior distribution of the radiation field is constrained by Markov random fields, and the posterior distribution of the radiation field is calculated according to Bayes' theorem. The calculation expression for the posterior distribution of the radiation field is as follows: ， in, This represents the posterior probability density of the radiation field intensity under given measurement data conditions. Indicates proportional to, This represents the prior probability density of the radiation intensity vector; S25. By maximizing the posterior probability of the radiation field or based on the variational Bayesian method, the posterior estimate of the voxel-level radiation intensity and the posterior covariance matrix are calculated. S26. Based on the posterior estimate of voxel-level radiation intensity and the posterior covariance matrix, extract the coefficient of variation of the uncertainty measure for each voxel. The expression for calculating the coefficient of variation of the uncertainty measure for each voxel is as follows: ， ， in, Indicates the first The coefficient of variation, a measure of uncertainty in individual elements. Indicates the first The standard deviation of the uncertainty measure of individual elements. Indicates the first Posterior estimate of the radiation intensity of an individual element. Represents the posterior covariance matrix The first on the main diagonal Each element.

4. The semantic-based multi-level visualization modeling method for radiation fields according to claim 3, characterized in that, S3 includes the following steps: S31. Use a deep learning model to perform instance segmentation and recognition on a 3D geometric network model to obtain several semantic objects, and label the corresponding semantic objects as environmental structure semantic tags. S32. Define the semantics of the physical mechanism of radiative transfer, and identify and mark the shadow region and the scattering-dominant region as physical semantic labels; S33. Mark regions where the coefficient of variation of the uncertainty measure of voxels is greater than a preset coefficient of variation threshold as high uncertainty regions, and use them as statistical semantic labels.

5. The semantic-based multi-level visualization modeling method for radiation fields according to claim 4, characterized in that, The semantics of the physical mechanism of radiative transfer include attenuation semantics and scattering semantics; The attenuation semantics are as follows: if a region is identified where the angle between the direction of the radiation field intensity gradient and the normal vector of the occluder is greater than 90°, then the region is marked as a shadow region. The scattering semantics are as follows: if a region is identified where the scattering contribution ratio of the radiation detector response is greater than a preset scattering contribution ratio threshold, then that region is marked as a scattering-dominant region.

6. The semantic-based multi-level visualization modeling method for radiation fields according to claim 4, characterized in that, S4 includes the following steps: S41. Based on environmental structure semantic tags, physical semantic tags and statistical semantic tags, construct a unified mapping mechanism for the HSV color space. The unified mapping mechanism for HSV color includes hue mapping based on radiation intensity, saturation mapping based on uncertainty, and brightness mapping based on semantic fusion. S42. Based on the unified mapping mechanism of HSV color space, and using the a posteriori estimate of voxel-level radiation intensity and the coefficient of variation of the uncertainty measure of each voxel, a three-dimensional radiation field model including color, transparency and illumination is drawn in real time within the three-dimensional geometric network model using a graphics rendering engine.

7. The semantic-based multi-level visualization modeling method for radiation fields according to claim 6, characterized in that, The calculation expression for the hue mapping in S41 is as follows: ， in, Representing three-dimensional spatial position The hue value at that location, Indicates the lower limit of the hue value. Indicates the upper limit of the hue value. Represents the logarithmic function. Representing three-dimensional spatial position The posterior estimate of the radiation intensity of the corresponding voxel. This represents the minimum radiation intensity within the target environment. This indicates the maximum radiation intensity within the target environment.

8. The semantic-based multi-level visualization modeling method for radiation fields according to claim 6, characterized in that, The calculation expression for the saturation mapping in S41 is as follows: ， in, Representing three-dimensional spatial position The saturation value at that point, This indicates the upper limit of the saturation value. This represents the sigmoid function. This represents the saturation adjustment coefficient. Representing three-dimensional spatial position The coefficient of variation, representing the uncertainty measure of the corresponding voxel. This indicates that the threshold for the coefficient of variation is set.

9. The semantic-based multi-level visualization modeling method for radiation fields according to claim 6, characterized in that, The calculation expression for the brightness mapping in S41 is as follows: ， in, Representing three-dimensional spatial position The brightness value at that location, The weighting coefficients represent the semantic weighting of the environmental structure. This indicates a semantic indicator function for the environment structure. Represents the weighting coefficients for physical semantics. Represents a physical semantic indicator function. Represents the weighting coefficients for statistical semantics. This represents a statistical semantic indicator function.

Images