WebGL-based Radar Volume Rendering Performance Optimization System

Abstract

The present invention relates to the technical field of WebGL radar volume rendering, and discloses a WebGL-based radar volume rendering performance optimization system. The method includes a texture acquisition and reconstruction module, a dynamic decomposition and clustering module, a metric calculation module, a strategy determination module, and a compression and rendering module. After structuring and reconstructing the original radar volume data, the method uses energy diffusion simulation to achieve dynamic hierarchical decomposition, and then performs adaptive clustering according to the texture feature correlation. By calculating the texture change intensity and spatial consistency metrics of each rendering subset, the global rendering load is evaluated, and then an adaptive texture compression strategy is determined. Finally, the structured texture field is lossily compressed and sent to the WebGL pipeline for rendering. This method enables the hierarchical structure to automatically conform to the data features and organizes the data according to the feature consistency, realizing the adaptive allocation and efficient compression of rendering resources, and improving the rendering efficiency and drawing quality in the Web environment.

Description

Technical Field

[0001] This invention relates to the field of WebGL radar volume rendering technology, specifically a WebGL-based radar volume rendering performance optimization system. Background Technology

[0002] Real-time rendering of large-scale radar volume data in a WebGL environment faces performance bottlenecks. Existing technologies generally employ multi-level detail models based on fixed spatial partitioning and data chunking strategies based on spatial proximity. This approach uniformly divides the 3D texture field geometrically and simply organizes adjacent volume data into rendering units to manage data loading and rendering.

[0003] This conventional approach has its flaws. Its hierarchical structure fails to correlate with the non-uniform distribution of physical features within radar data, leading to inappropriate allocation of computational resources. Regions with significant feature variations may lose detail due to coarse representation, while uniform regions waste computational resources due to excessive subdivision. Furthermore, simple spatial clustering results in excessively large differences in texture features within individual rendering units, failing to provide a consistent data foundation for subsequent processing and limiting targeted compression and rendering optimization.

[0004] A rendering preprocessing method is needed, whose hierarchical structure can dynamically adapt to the inherent characteristics of the data, and whose data organization is guided by feature consistency. This invention requires a technical solution that can automatically generate non-uniform detail hierarchy partitions based on the topological and texture characteristics of volumetric data, and aggregate data with similar features to support differentiated and efficient compression and rendering. Summary of the Invention

[0005] The purpose of this invention is to provide a WebGL-based radar volume rendering performance optimization system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a WebGL-based radar volume rendering performance optimization system, the method comprising:

[0007] The texture acquisition and reconstruction module is used to acquire the original three-dimensional texture field of radar volume data and perform topological reconstruction based on voxel structure features on the original three-dimensional texture field to generate a structured texture field.

[0008] The dynamic decomposition and clustering module is used to perform dynamic hierarchical decomposition based on energy diffusion simulation on the structured texture field to obtain multiple detail level units, and to perform adaptive clustering based on texture feature correlation on the multiple detail level units to form several rendering subsets.

[0009] The metrics calculation module is used to calculate the texture change intensity and spatial consistency metrics for each rendering subset, and to calculate the global rendering load factor based on the texture change intensity and spatial consistency metrics of all rendering subsets.

[0010] The strategy determination module is used to determine a texture compression strategy based on the global rendering load factor and a preset rendering parameter threshold.

[0011] The compression rendering module is used to perform lossy compression encoding on the structured texture field according to the texture compression strategy, and input the compressed texture data into the WebGL rendering pipeline for volume rendering.

[0012] Preferably, the method for acquiring the original three-dimensional texture field of radar volume data and performing topological reconstruction based on voxel structure features on the original three-dimensional texture field to generate a structured texture field includes:

[0013] The spatial gradient vectors of each voxel point within the original 3D texture field are calculated to form a gradient field. Discrete divergence calculation is performed on the gradient field to obtain the structure tensor of each voxel point. Local anisotropy descriptors for voxel points are constructed using the eigenvalues ​​of the structure tensor. Region growing is performed on the voxel points based on the local anisotropy descriptors, merging voxel points with continuous characteristics into homogeneous regions. The boundary curvature and the uniformity of the internal voxel value distribution of each homogeneous region are calculated, and the boundary curvature and the uniformity are weighted and fused to generate the topological intensity value of the homogeneous region. The voxel points within the region are interpolated and resampled using the topological intensity value of each homogeneous region as a weight to generate a structured texture field.

[0014] Preferably, performing dynamic hierarchical decomposition of the structured texture field based on energy diffusion simulation includes:

[0015] A virtual energy field is constructed for the structured texture field, with each voxel point considered as an energy node, and the voxel value mapped to the initial energy of the node. A rule for energy transfer between adjacent nodes is defined, which is inversely proportional to the difference in voxel values ​​between nodes and directly proportional to the topological intensity value of the node. The energy transfer process is iteratively executed until the energy distribution reaches a stable state. The nodes are sorted according to their energy values ​​after stabilization, and abrupt changes in energy values ​​are selected as the hierarchical segmentation threshold. Based on the hierarchical segmentation threshold, the structured texture field is divided into multiple energy levels, each energy level corresponding to a detail level unit, where units in higher energy levels contain voxels with higher topological intensity values ​​and voxel values.

[0016] Preferably, performing adaptive clustering based on texture feature correlation on the plurality of detail level units includes:

[0017] For each level-of-detail (LHD) unit, its texture spectral features are extracted. These features are obtained by performing a Fast Fourier Transform on the voxel value sequence within the unit and taking the amplitude of the low-frequency components. The cosine similarity of the texture spectral features between any two LHD units is calculated. A dynamic similarity threshold is set, which is negatively correlated with the expected value of the global rendering load factor. Connection edges are established with LHD units as vertices, and the cosine similarity between units is greater than the dynamic similarity threshold, forming an undirected graph. The largest complete subgraph is found in the undirected graph, and the set of LHD units contained in each largest complete subgraph constitutes a rendering subset.

[0018] Preferably, the method for calculating the intensity of texture changes in a rendered subset includes:

[0019] For the rendered subset, obtain all its level-of-detail units; calculate the standard deviation of voxel values ​​within each level-of-detail unit; take a weighted average of the standard deviations of all level-of-detail units, with the weight being the proportion of voxel points contained in the level-of-detail unit to the total number of voxel points in the rendered subset, to obtain the internal variation intensity of the rendered subset; calculate the variance of the difference in the mean voxel values ​​among the level-of-detail units within the rendered subset, as the inter-unit variation intensity of the rendered subset; add the internal variation intensity to the inter-unit variation intensity to obtain the texture variation intensity of the rendered subset.

[0020] Preferably, the method for calculating the spatial consistency index of the rendered subset includes:

[0021] Calculate the spatial center coordinates of all detail level units in the rendered subset; calculate the covariance matrix of these spatial center coordinates; perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​in three principal directions; use the ratio of the minimum eigenvalue to the maximum eigenvalue as the spatial distribution dispersion of the rendered subset; calculate the Euclidean distance between each detail level unit and its spatial nearest neighbor unit on the texture spectral features, and take the average of all such distances as a measure of local feature continuity of the rendered subset; multiply the spatial distribution dispersion by the local feature continuity measure to obtain the spatial consistency index of the rendered subset, where a lower spatial consistency index indicates better spatial consistency.

[0022] Preferably, the method for calculating the global rendering load factor includes:

[0023] Traverse all rendering subsets and obtain the texture variation intensity and spatial consistency index for each subset. For each subset, calculate the product of its texture variation intensity and spatial consistency index as the subset rendering complexity. Multiply the subset rendering complexity by the total number of voxels contained in the subset to obtain the estimated load of the subset. Sum the estimated loads of all rendering subsets to obtain the total estimated load. Calculate the mean of the spatial consistency index of all rendering subsets as the average spatial consistency. Divide the total estimated load by the product of the average spatial consistency and a preset baseline factor to obtain the global rendering load factor.

[0024] Preferably, determining the texture compression strategy based on the global rendering load factor and a preset rendering parameter threshold includes:

[0025] The global rendering load factor is compared with preset high load thresholds and medium load thresholds. If the global rendering load factor is greater than the high load threshold, a multi-level compression strategy based on wavelet transform is adopted, and the compression level of the multi-level compression strategy is dynamically determined by the proportion by which the global rendering load factor exceeds the high load threshold. If the global rendering load factor is between the medium load threshold and the high load threshold, a hybrid compression strategy based on block truncation coding is adopted, and the block size and quantization bits during compression are obtained by linear interpolation of the global rendering load factor. If the global rendering load factor is less than the medium load threshold, a lossless compression strategy based on predictive coding is adopted.

[0026] Preferably, performing lossy compression encoding on the structured texture field according to the texture compression strategy includes:

[0027] When a multi-level compression strategy based on wavelet transform is adopted, a three-dimensional discrete wavelet transform is performed on the structured texture field to obtain multi-level sub-band coefficients. According to the compression level, a corresponding number of low-frequency sub-band coefficients are retained, high-frequency sub-band coefficients are threshold quantized, high-frequency coefficients below the threshold are discarded, and entropy encoding is performed on the quantized coefficients. When a hybrid compression strategy based on block truncation coding is adopted, the structured texture field is divided into multiple texture blocks of a size determined by the global rendering load factor. The mean and range of voxel values ​​are calculated for each texture block, and the voxel values ​​within the block are recalibrated using the quantization bit depth. When a lossless compression strategy based on predictive coding is adopted, a three-dimensional predictive filter is used to process the structured texture field, and arithmetic coding is performed on the prediction residuals.

[0028] Preferably, when employing a multi-level compression strategy based on wavelet transform, the step of performing a three-dimensional discrete wavelet transform on the structured texture field to obtain multi-level sub-band coefficients includes:

[0029] Select a preset wavelet basis function, which has compact support and regularity;

[0030] The structured texture field is subjected to one-dimensional wavelet transform in three spatial dimensions. The transform in each dimension includes low-pass filtering and high-pass filtering of the voxel data sequence, and downsampling of the filtering results.

[0031] The transformation results in the three dimensions are combined by tensor product to generate a three-dimensional wavelet coefficient matrix.

[0032] The repeated transformation process decomposes the low-frequency subband into multiple levels until the preset decomposition level is reached, resulting in a three-dimensional subband coefficient set containing multiple resolution levels.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] A dynamic hierarchical decomposition technique based on energy diffusion simulation is employed, treating the 3D texture field as an initial energy field. The generation of hierarchical structures is driven by simulating the transfer and dissipation of energy within this field. Energy naturally tends to diffuse rapidly and reach equilibrium in regions with gentle feature changes, while it is hindered and forms high energy gradients in areas with rich feature boundaries or details. This physical process simulation allows the boundaries of the final generated detailed hierarchical units to automatically conform to the inherent physical structure features of the data, achieving non-uniform, data-driven adaptive subdivision. Rendering resources can be allocated more precisely, retaining more detail in visually sensitive high-frequency feature regions and rationally merging in uniform regions, thereby reducing the number of voxels requiring real-time processing while maintaining visual fidelity.

[0035] An adaptive clustering technique based on texture feature correlation is employed. After hierarchical decomposition, units are reorganized according to their similarity in texture attributes such as color statistics and gradient distribution, rather than adhering to their spatial location. Units with highly correlated features are aggregated into the same rendering subset, making the texture features within each subset more consistent. This data reorganization guided by feature consistency creates favorable conditions for downstream processing. The rendering pipeline can select the lossy compression algorithm and parameters with the highest matching degree for each subset with consistent features. This improves the overall compression efficiency and quality, and because the data characteristics are consistent within the same subset, the GPU's texture sampling and shading calculations during rendering are also more efficient, reducing state switching and computational divergences. Attached Figure Description

[0036] Figure 1 This is a timing diagram of the WebGL-based radar volume rendering performance optimization system described in this invention.

[0037] Figure 2 A flowchart of a method for generating structured texture fields;

[0038] Figure 3 This is a flowchart of dynamic hierarchical decomposition.

[0039] Figure 4 Thermodynamic analysis diagram showing the sensitivity of the global load factor to parameter adjustments;

[0040] Figure 5 A heatmap showing the impact of texture block size and quantization bit depth on rendering efficiency in radar volume rendering. Detailed Implementation

[0041] 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.

[0042] Please see Figure 1 This invention provides a WebGL-based radar volume rendering performance optimization system. The method includes: a texture acquisition and reconstruction module acquiring the original 3D texture field of radar volume data and performing topological reconstruction based on voxel structure features to generate a structured texture field; a dynamic decomposition and clustering module performing dynamic hierarchical decomposition based on energy diffusion simulation on the structured texture field to obtain multiple detail level units, and performing adaptive clustering based on texture feature correlation on these units to form several rendering subsets; an index calculation module calculating the texture change intensity and spatial consistency index for each rendering subset, and calculating the global rendering load factor based on these indices for all rendering subsets; a strategy determination module determining a texture compression strategy based on the global rendering load factor and preset rendering parameter thresholds; and a compressed rendering module performing lossy compression encoding on the structured texture field according to the texture compression strategy, and inputting the compressed texture data into the WebGL rendering pipeline for volume rendering.

[0043] Example 1: See Figure 2The texture acquisition and reconstruction module acquires the original 3D texture field of the radar volume data and performs topological reconstruction based on voxel structural features to generate a structured texture field. It calculates the spatial gradient vector of each voxel point within the original 3D texture field to form a gradient field; performs discrete divergence calculation on the gradient field to obtain the structure tensor of each voxel point; constructs local anisotropy descriptors for voxel points using the eigenvalues ​​of the structure tensor; performs region growing on voxel points based on the local anisotropy descriptors, merging voxel points with continuous features into homogeneous regions; calculates the boundary curvature and uniformity of the internal voxel value distribution of each homogeneous region, and weights and fuses the boundary curvature and uniformity to generate the topological intensity value of the homogeneous region; and uses the topological intensity value of each homogeneous region as a weight to interpolate and resample the voxel points within the region to generate the structured texture field.

[0044] In specific implementation, the texture acquisition and reconstruction module acquires the original 3D texture field of the radar volume data and performs topological reconstruction based on voxel structural features to generate a structured texture field. It calculates the spatial gradient vector of each voxel point in the original 3D texture field to form a gradient field. It performs discrete divergence calculation on the gradient field to obtain the structure tensor of each voxel point. It constructs a local anisotropy descriptor for the voxel points using the eigenvalues ​​of the structure tensor. It performs region growing on the voxel points according to the local anisotropy descriptor to merge voxel points with continuous features into homogeneous regions. It calculates the boundary curvature and the uniformity of the internal voxel value distribution of each homogeneous region, and weights and fuses the boundary curvature and uniformity to generate the topological intensity value of the homogeneous region. It uses the topological intensity value of each homogeneous region as a weight to interpolate and resample the voxel points in the region to generate a structured texture field. In some embodiments, the spatial gradient vector is calculated using the central difference method. Each vector component in the gradient field represents the rate of change of the original three-dimensional texture field in the corresponding spatial direction. The discrete divergence is calculated based on the vector field of the gradient field. The structure tensor is obtained by taking the cross product of the gradient vectors and calculating the average value in the local neighborhood. The structure tensor is a symmetric matrix whose eigenvalues ​​reflect the degree of anisotropy of the local structure.

[0045] It can be understood that local anisotropy descriptors are constructed from the eigenvalues ​​of the structure tensor. The descriptor values ​​indicate the shape characteristics of the region where the voxel points are located, such as linear, planar, or spherical features. The region growth process iteratively merges adjacent voxel points from the seed point using the local anisotropy descriptors as a similarity criterion until the descriptor difference exceeds a preset threshold, thus forming a homogeneous region. In specific implementations, the boundary curvature is obtained by calculating the rate of change of the normal vector of the voxel points on the surface of the homogeneous region, and the uniformity of the internal voxel value distribution is obtained by calculating the variance of the voxel values ​​within the region. To enable linear weighted fusion of boundary curvature and uniformity, both need to be normalized. For the boundary curvature of the homogeneous region... Normalization is performed on the set of curvature values ​​across all homogeneous regions to obtain the normalized curvature. The uniformity of voxel value distribution within a homogeneous region is defined as the reciprocal of its variance. Then, normalize the uniformity values ​​in the set of all homogeneous regions to obtain the normalized uniformity. The formula for calculating the topological strength value is:

[0046]

[0047] in: Indicates homogeneous regions The topological strength value, Indicates homogeneous regions The normalized boundary curvature, Indicates homogeneous regions Normalized uniformity, and These are preset dimensionless weighting coefficients.

[0048] Optionally, the interpolation resampling process adjusts the sampling density based on the topological intensity value, using denser sampling points in regions with high topological intensity values ​​and sparser sampling points in regions with low topological intensity values, thereby generating a structured texture field. The structured texture field retains the key structural information of the original 3D texture field while reducing data redundancy. In some embodiments, eight-neighborhood or twenty-six-neighborhood connectivity is used during region growing to ensure continuity in 3D space. The eigenvalues ​​of the structure tensor are calculated using the Jacobi iteration method, and the magnitude relationship of the eigenvalues ​​is used to construct a local anisotropy descriptor, for example, quantifying anisotropy through the ratio of eigenvalues.

[0049] Example 2: See Figure 3The dynamic decomposition and clustering module performs dynamic hierarchical decomposition of the structured texture field based on energy diffusion simulation, and performs adaptive clustering based on texture feature correlation for multiple detail level units. A virtual energy field is constructed for the structured texture field, treating each voxel as an energy node, with voxel values ​​mapped to the node's initial energy. Energy transfer rules are defined between adjacent nodes, inversely proportional to the difference in voxel values ​​between nodes and directly proportional to the node's topological strength value. The energy transfer process is iteratively executed until the energy distribution reaches a stable state. Nodes are sorted according to their stable energy values, and abrupt changes in energy values ​​are selected as hierarchical segmentation thresholds. Based on these thresholds, the structured texture field is divided into multiple energy levels, each corresponding to a detail level unit, where higher-energy-level units contain voxels with higher topological strength and voxel values. For each level-of-detail (LHD) unit, its texture spectral features are extracted. These features are obtained by performing a Fast Fourier Transform on the voxel value sequence within the unit and taking the amplitude of the low-frequency components. The cosine similarity of the texture spectral features between any two LHD units is calculated. A dynamic similarity threshold is set, which is negatively correlated with the expected value of the global rendering load factor. With LHD units as vertices, connecting edges are established based on the condition that the cosine similarity between units is greater than the dynamic similarity threshold, forming an undirected graph. The largest complete subgraph is found in the undirected graph, and the set of LHD units contained in each largest complete subgraph constitutes a rendering subset.

[0050] In specific implementation, the dynamic decomposition and clustering module performs dynamic hierarchical decomposition of the structured texture field based on energy diffusion simulation, and performs adaptive clustering based on texture feature correlation for multiple detail level units. A virtual energy field is constructed for the structured texture field, treating each voxel as an energy node, with the voxel value mapped to the node's initial energy. A rule for energy transfer between adjacent nodes is defined, inversely proportional to the difference in voxel values ​​between nodes and directly proportional to the node's topological strength value. The energy transfer process is iteratively executed until the energy distribution reaches a stable state. Nodes are sorted according to their stable energy values, and abrupt changes in energy values ​​are selected as the hierarchical segmentation threshold. Based on the hierarchical segmentation threshold, the structured texture field is divided into multiple energy levels, each corresponding to a detail level unit, where higher energy level units contain voxels with higher topological strength and voxel values. In some embodiments, the formula for calculating the energy transfer weight between adjacent nodes is:

[0051]

[0052] in: Indicates from node Pass to adjacent nodes Weighting factors and These are nodes and nodes The topological strength value, Represents a node With nodes The absolute value of the difference in voxel values. In each iteration, the node... to adjacent nodes The energy value transferred is at the node Current energy and weighting factor The product of the nodes, and the nodes Receive energy from all neighboring nodes.

[0053] It is understandable that the iterative energy transfer process employs a synchronous update strategy, calculating the energy of all nodes simultaneously based on the state of the previous round. The criterion for a stable energy distribution is that the sum of the changes in energy values ​​of all nodes in two adjacent iterations is less than a preset minimum threshold. After sorting the node energy values, abrupt change points are found by analyzing the first-order difference of the sorted energy value sequence. In specific implementation, abrupt change points in energy values ​​are determined by detecting positions in the first-order difference sequence that exceed the global average difference by a certain multiple (e.g., three times). The energy values ​​corresponding to these positions are the hierarchical segmentation thresholds. The structured texture field is segmented into multiple non-overlapping energy levels based on these thresholds. Within each energy level, the energy values ​​of all nodes are in the same interval, and each energy level constitutes a detail level unit. Optionally, the set of detail level units constitutes a multi-level representation of the original structured texture field.

[0054] In some embodiments, texture spectral features are extracted for each HML (Level of Detail) unit. These features are obtained by performing a Fast Fourier Transform (FFT) on the voxel value sequence within the unit and extracting the amplitudes of the low-frequency components. The cosine similarity of the texture spectral features between any two HMLs is calculated. A dynamic similarity threshold is set, which is negatively correlated with the expected value of the global rendering load factor. An undirected graph is formed by establishing connections between HMLs as vertices, with the condition that the cosine similarity between units is greater than the dynamic similarity threshold. The largest complete subgraph is found within the undirected graph, and the set of HMLs contained in each largest complete subgraph constitutes a rendering subset. It can be understood that during texture spectral feature extraction, the voxel value sequence is generated by traversing all voxel points within the HML and arranging them in the order of the space-filling curve. After the FFT, the amplitudes of the first K low-frequency components are retained to form a feature vector, and the cosine similarity calculation is based on these two feature vectors. Dynamic similarity threshold. The calculation formula and the expected value of the global rendering load factor Related, for example ,in It is the basic threshold. It is the adjustment coefficient.

[0055] In practice, the undirected graph is constructed based on the calculated cosine similarity matrix between all pairs of detail-level units. Connecting edges exist only between unit pairs with a similarity greater than the dynamic similarity threshold. The Bron-Kerbosch algorithm is used to find the maximal complete subgraph within the undirected graph. Ultimately, each rendered subset is composed of multiple detail-level units with high texture feature correlation. Optionally, the expected value of the global rendering load factor is estimated during system initialization based on historical data or default parameters, and is used to guide the initial setting of the dynamic similarity threshold.

[0056] Example 3: The index calculation module calculates the texture variation intensity and spatial consistency index for each rendering subset. For each rendering subset, it obtains all the level-of-detail units (LHDs) it contains; calculates the standard deviation of the voxel values ​​within each LHD; performs a weighted average of the standard deviations of all LHDs, with the weight being the proportion of the voxel points contained in the LHD to the total number of voxels in the rendering subset, to obtain the internal variation intensity of the rendering subset; calculates the variance of the difference in the mean voxel values ​​among the LHDs within the rendering subset, as the inter-unit variation intensity of the rendering subset; and adds the internal variation intensity to the inter-unit variation intensity to obtain the texture variation intensity of the rendering subset. Calculate the spatial consistency index of the rendered subset, calculate the coordinates of the spatial center points of all detail units in the rendered subset; calculate the covariance matrix of these spatial center point coordinates; perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​in three principal directions; use the ratio of the minimum eigenvalue to the maximum eigenvalue as the spatial distribution dispersion of the rendered subset; calculate the Euclidean distance between each detail unit and its spatial nearest neighbor unit on the texture spectral features, and take the average of all such distances as a measure of local feature continuity of the rendered subset; multiply the spatial distribution dispersion by the local feature continuity measure to obtain the spatial consistency index of the rendered subset, the lower the spatial consistency index, the better the spatial consistency.

[0057] In specific implementation, the index calculation module calculates the texture variation intensity and spatial consistency index for each rendering subset. For each rendering subset, it obtains all its level-of-detail (HLD) units; calculates the standard deviation of voxel values ​​within each HLD unit; and performs a weighted average of the standard deviations of all HLD units, with the weight being the proportion of voxel points contained in the HLD unit to the total number of voxel points in the rendering subset, to obtain the internal variation intensity of the rendering subset. It also calculates the variance of the difference in the mean voxel values ​​among the HLD units within the rendering subset, as the inter-unit variation intensity of the rendering subset. Finally, it adds the internal variation intensity to the inter-unit variation intensity to obtain the texture variation intensity of the rendering subset. In some embodiments, the standard deviation of voxel values ​​within each HLD unit is calculated using an unbiased estimation formula, and the number of voxel points contained in the HLD unit is directly obtained from the unit's data structure. The weighted averaging process involves multiplying the standard deviation of each HLD unit by its corresponding weight coefficient and then summing the results. To calculate the variance of the difference in voxel values ​​among the units of detail within a rendering subset, we first need to calculate the mean voxel value of each unit of detail in the rendering subset. Then, we calculate the absolute difference between each pair of these means. Finally, we calculate the variance of these absolute difference sequences. The formula for calculating the variance is:

[0058]

[0059] in: This indicates the intensity of variation between cells in the rendered subset. This indicates the number of detail level units contained in the rendered subset. and They represent the first The and the first The average voxel value of each detail unit. This formula quantifies the degree of fluctuation in average brightness between different units within a rendered subset.

[0060] It can be understood that the texture variation intensity of a rendered subset is the direct sum of the internal variation intensity and the inter-unit variation intensity. The internal variation intensity reflects the richness of detail within each detail level unit, while the inter-unit variation intensity reflects the overall contrast difference between different units. In specific implementation, the spatial consistency index of the rendered subset is calculated, the coordinates of the spatial center points of all detail level units in the rendered subset are calculated, and the covariance matrix of these spatial center point coordinates is calculated. Eigenvalue decomposition is performed on the covariance matrix to obtain eigenvalues ​​in three principal directions. The ratio of the minimum eigenvalue to the maximum eigenvalue is used as the spatial distribution dispersion of the rendered subset. The Euclidean distance between each detail level unit and its spatial nearest neighbor unit in the texture spectral features is calculated, and the average of all such distances is taken as a measure of the local feature continuity of the rendered subset. The spatial distribution dispersion is multiplied by the local feature continuity measure to obtain the spatial consistency index of the rendered subset. The lower the spatial consistency index, the better the spatial consistency. In some embodiments, the coordinates of the spatial center point of the detail level unit are obtained by calculating the arithmetic mean of the three-dimensional coordinates of all voxel points within the unit. The covariance matrix is ​​calculated based on the set of these center point coordinates. The eigenvalue decomposition uses the Jacobi iteration method to obtain three real eigenvalues ​​and sort them in descending order.

[0061] Optionally, the search for the spatial nearest neighbor unit is achieved by calculating the Euclidean distance between each pair of center points of all detail level units in 3D space, finding the unit with the smallest distance for each unit as its spatial nearest neighbor unit. It can be understood that the calculation of the Euclidean distance of the texture spectrum features is based on the texture spectrum feature vector extracted in Example 2; the smaller the distance value, the more similar the two units are in texture patterns. In a specific implementation, the calculation of the local feature continuity metric traverses each detail level unit in the rendering subset, finds its spatial nearest neighbor unit, calculates the Euclidean distance between their texture spectrum feature vectors, and finally takes the arithmetic mean of the distance values ​​of all units. The calculation of the spatial consistency index directly multiplies the spatial distribution dispersion by the local feature continuity metric. This product comprehensively reflects the degree of clustering of the rendering subset in spatial location and the local smoothness of texture features. Optionally, the spatial distribution dispersion ranges from 0 to 1. When all center points are collinear or coplanar, the ratio approaches 0, indicating a highly concentrated spatial distribution; when center points are uniformly distributed in all directions, the ratio approaches 1, indicating a dispersed spatial distribution.

[0062] Example 4: The index calculation module calculates the global rendering load factor based on the texture change intensity and spatial consistency index of all rendering subsets. It iterates through all rendering subsets, obtaining the texture change intensity and spatial consistency index for each subset; for each subset, it calculates the product of its texture change intensity and spatial consistency index as the subset rendering complexity; it multiplies the subset rendering complexity by the total number of voxels contained in the subset to obtain the estimated load of the subset; it sums the estimated loads of all rendering subsets to obtain the total estimated load; it calculates the mean of the spatial consistency index of all rendering subsets as the average spatial consistency; and it divides the total estimated load by the product of the average spatial consistency and a preset baseline factor to obtain the global rendering load factor. The strategy determination module determines the texture compression strategy based on the global rendering load factor and preset rendering parameter thresholds. It compares the global rendering load factor with preset high load thresholds and medium load thresholds. If the global rendering load factor is greater than the high load threshold, a multi-level compression strategy based on wavelet transform is adopted. The compression level of the multi-level compression strategy is dynamically determined by the proportion by which the global rendering load factor exceeds the high load threshold. If the global rendering load factor is between the medium load threshold and the high load threshold, a hybrid compression strategy based on block truncation coding is adopted. The block size and quantization bits during compression are obtained by linear interpolation of the global rendering load factor. If the global rendering load factor is less than the medium load threshold, a lossless compression strategy based on predictive coding is adopted.

[0063] In specific implementation, the index calculation module calculates the global rendering load factor based on the texture change intensity and spatial consistency index of all rendering subsets. It iterates through all rendering subsets, obtaining the texture change intensity and spatial consistency index for each subset. For each rendering subset, it calculates the product of its texture change intensity and spatial consistency index as the subset rendering complexity. It then multiplies the subset rendering complexity by the total number of voxels contained in the subset to obtain the estimated load of the subset. Finally, it sums the estimated loads of all rendering subsets to obtain the total estimated load. The average spatial consistency index of all rendering subsets is calculated as the average spatial consistency. The total estimated load is divided by the product of the average spatial consistency and a preset benchmark factor to obtain the global rendering load factor. In some embodiments, the texture change intensity and spatial consistency index of the rendering subset are directly obtained from the calculation process in Example 3. The product calculation of the subset rendering complexity reflects the rendering difficulty jointly constituted by the texture complexity and spatial distribution consistency within the subset. The total number of voxels contained in the rendering subset is directly accumulated from the data structure of the rendering subset. Calculating the total estimated load involves arithmetically summing the estimated loads of all rendered subsets. Calculating the average spatial consistency involves arithmetically averaging the spatial consistency metrics of all rendered subsets. The preset baseline factor is a constant used to adjust the range of the global rendering load factor; its value is related to the target frame rate or historical performance data of the display device. Global Rendering Load Factor The calculation formula is expressed as follows:

[0064]

[0065] in: Indicates the global rendering load factor. This represents the total estimated load for all rendered subsets. This represents the average spatial consistency metric across all rendered subsets, i.e., the average spatial consistency. This represents the preset benchmark factor. It can be understood that the numerator in the formula... Taking into account the complexity and data volume of all subsets, the denominator achieves average spatial consistency. Normalization is performed so that when the rendered subset is more spatially clustered and its features are more continuous, the global rendering load factor will decrease accordingly, based on a preset baseline factor. This is used to adjust the final factor value to a typical order of magnitude that facilitates subsequent threshold comparisons. The strategy determination module determines the texture compression strategy based on the global rendering load factor and preset rendering parameter thresholds, comparing the global rendering load factor with preset high and medium load thresholds. If the global rendering load factor is greater than the high load threshold, a multi-level compression strategy based on wavelet transform is adopted, with the compression level dynamically determined by the proportion by which the global rendering load factor exceeds the high load threshold. If the global rendering load factor is between the medium and high load thresholds, a hybrid compression strategy based on block truncation coding is adopted, with the block size and quantization bits obtained by linear interpolation of the global rendering load factor value. If the global rendering load factor is less than the medium load threshold, a lossless compression strategy based on predictive coding is adopted. In specific implementation, the preset high and medium load thresholds are pre-set based on WebGL rendering pipeline performance tests and empirical values ​​to define high, medium, and low rendering load scenarios. The formula for calculating the proportion by which the global rendering load factor exceeds the high load threshold is... ,in To exceed the proportion, To determine the high load threshold, a specific compression level is determined based on this ratio through a preset mapping relationship. For example, a ratio R in the range of 0-0.2 maps to compression level 1, and one in the range of 0.2-0.5 maps to compression level 2. For hybrid compression strategies based on block truncation coding, the block size... and quantization bits The boundary is determined by linear interpolation between preset upper and lower bounds, and the calculation formulas are as follows: and ,in For medium load threshold, and These are the preset minimum and maximum block sizes, respectively. and These represent the preset minimum and maximum quantization bits, respectively. The trigger logic for the three optional texture compression strategies is shown in Table 1.

[0066] Table 1: Correspondence between Global Rendering Load Factor and Texture Compression Strategy

[0067]

[0068] In some embodiments, the high load threshold With medium load threshold The numerical relationship is The specific value is configured during system initialization based on the rendering capabilities of the target hardware platform. It can be understood that when the global rendering load factor is high, it indicates complex data and high rendering pressure; therefore, a multi-level compression strategy based on wavelet transform with a higher compression ratio is adopted to significantly reduce the data volume. When the load factor is moderate, a hybrid compression strategy based on block truncation coding, balancing compression ratio and image quality, is used. When the load factor is low, rendering pressure is low, and a lossless compression strategy based on predictive coding is used to preserve data integrity.

[0069] See Figure 4 In the global load factor sensitivity analysis, the influence of the texture variation intensity adjustment coefficient and the spatial consistency index adjustment coefficient on the global rendering load factor was quantified in the form of a heatmap. The vertical axis of the heatmap represents the texture variation intensity adjustment coefficient (range 0.80-1.20), the horizontal axis represents the spatial consistency index adjustment coefficient (range 0.80-1.20), and the color gradient corresponds to the global rendering load factor value (range 34-48). The black star marks the global load factor under the original parameter combination (40.0). The heatmap distribution shows that when the texture variation intensity adjustment coefficient decreases (approaching 0.80) or the spatial consistency index adjustment coefficient decreases, the global load factor shows a significant upward trend (corresponding to the red area in the heatmap); while as both coefficients increase towards 1.20, the load factor gradually decreases (corresponding to the blue area in the heatmap). This pattern aligns perfectly with the calculation formula for the global rendering load factor (the coupling relationship between total estimated load, average spatial consistency, and baseline factor): reducing texture variation intensity directly increases subset rendering complexity, while reducing spatial consistency reduces the weight of the normalization denominator; the combined effect of these two factors amplifies the load factor. The original parameters (all coefficients 1.00) fall within the middle range of the load factor, confirming that the system is in a balanced rendering load state under the initial parameter configuration. Regarding parameter configuration, the figure provides a quantitative basis for system parameter tuning: to reduce the global load factor to adapt to high rendering pressure scenarios, the texture variation intensity adjustment coefficient can be increased first; to test the adaptability of compression strategies under extreme loads, both types of coefficients can be simultaneously reduced to trigger the wavelet compression strategy corresponding to the high load threshold.

[0070] Example 5: The compression rendering module performs lossy compression coding on the structured texture field according to the texture compression strategy. When a multi-level compression strategy based on wavelet transform is adopted, a three-dimensional discrete wavelet transform is performed on the structured texture field to obtain multi-level sub-band coefficients; according to the compression level, a corresponding number of low-frequency sub-band coefficients are retained, high-frequency sub-band coefficients are threshold-quantized, high-frequency coefficients below the threshold are discarded, and entropy coding is performed on the quantized coefficients. When a hybrid compression strategy based on block truncation coding is adopted, the structured texture field is divided into multiple texture blocks of a size determined by the global rendering load factor. The mean and range of voxel values ​​are calculated for each texture block, and the voxel values ​​within the block are recalibrated using the quantization bit depth. When a lossless compression strategy based on predictive coding is adopted, a three-dimensional predictive filter is used to process the structured texture field, and arithmetic coding is performed on the prediction residuals. The steps for performing a three-dimensional discrete wavelet transform on the structured texture field to obtain multi-level subband coefficients include: selecting a preset wavelet basis function, which has compact support and regularity; performing a one-dimensional wavelet transform on the structured texture field in three spatial dimensions sequentially, with each dimension's transformation including low-pass filtering and high-pass filtering of the voxel data sequence, and downsampling the filtering results; combining the transformation results in the three dimensions using tensor products to generate a three-dimensional wavelet coefficient matrix; and repeating the transformation process to decompose the low-frequency subband into multiple levels until a preset decomposition level is reached, resulting in a set of three-dimensional subband coefficients containing multiple resolution levels.

[0071] In practical implementation, the compression rendering module performs lossy compression encoding on the structured texture field according to the texture compression strategy. When a multi-level compression strategy based on wavelet transform is adopted, a three-dimensional discrete wavelet transform is performed on the structured texture field to obtain multi-level sub-band coefficients. According to the compression level, a corresponding number of low-frequency sub-band coefficients are retained, and the high-frequency sub-band coefficients are threshold-quantized, discarding high-frequency coefficients below the threshold. The quantized coefficients are then entropy-encoded. When a hybrid compression strategy based on block truncation encoding is adopted, the structured texture field is divided into multiple texture blocks of a size determined by the global rendering load factor. The mean and range of voxel values ​​are calculated for each texture block, and the voxel values ​​within the block are recalibrated using the quantization bit depth. When a lossless compression strategy based on predictive encoding is adopted, a three-dimensional predictive filter is used to process the structured texture field, and the prediction residuals are arithmetically encoded. The steps for performing a three-dimensional discrete wavelet transform on a structured texture field to obtain multi-level subband coefficients include: selecting a preset wavelet basis function, which has compact support and regularity; performing a one-dimensional wavelet transform on the structured texture field in three spatial dimensions sequentially, with each dimension's transformation including low-pass filtering and high-pass filtering of the voxel data sequence, and downsampling the filtering results; combining the transformation results in the three dimensions using tensor products to generate a three-dimensional wavelet coefficient matrix; and repeating the transformation process to decompose the low-frequency subband into multiple levels until a preset decomposition level is reached, resulting in a set of three-dimensional subband coefficients containing multiple resolution levels.

[0072] In some embodiments, the three-dimensional discrete wavelet transform performed by the multi-level compression strategy based on wavelet transform selects Daubechies wavelet or Haar wavelet as the wavelet basis function. Low-pass filtering and high-pass filtering are implemented through convolution operations, and downsampling operations retain even-indexed samples of the filtered sequence. During the multi-level decomposition process, the low-frequency subband obtained from the previous level is used as input for a new round of three-dimensional discrete wavelet transform, thereby forming a pyramid-shaped multi-resolution coefficient structure. The threshold quantization processing of the high-frequency subband coefficients sets a global threshold according to the compression level. The processing rules are as follows:

[0073]

[0074] in: Indicates the position in the high-frequency subband The original coefficient value at that location, This represents the quantized coefficient value. This represents a threshold dynamically calculated based on the compression level. This formula means that if the absolute value is less than the threshold... High-frequency coefficients are considered unimportant information and are set to zero and discarded. It's understandable that the higher the compression level, the higher the threshold. The larger the value, the more high-frequency coefficients are discarded, and the higher the data compression rate. The entropy coding stage uses a combination of run-length coding and Huffman coding to compress the quantized non-zero coefficients and their positional information.

[0075] It is understandable that the size of texture blocks in a hybrid compression strategy based on block truncation coding is three-dimensional, for example... or The specific size of each voxel is determined between a preset minimum and maximum size based on the global rendering load factor using the linear interpolation formula described in Example 4. For each divided texture block, the arithmetic mean of all voxel values ​​within the block is calculated as the mean, and the maximum and minimum voxel values ​​are identified to calculate the range. The formula for recalibrating each voxel value within the block using quantization bits is as follows: ,in It is the original voxel value. It is the block average. It is the range of voxel values ​​within the block (maximum value minus minimum value). It is the number of quantization bits. It is the recalibrated integer value. In some embodiments, the quantization bits... Also determined by linear interpolation of the global rendering load factor, typically ranging from 4 to 8 bits. Optional, a recalibrated integer value. The average value of the block ,scope The information is stored or transmitted together for data reconstruction before rendering.

[0076] In practical implementation, the lossless compression strategy based on predictive coding employs a three-dimensional linear predictive filter. The filter predicts the current value based on several spatial neighborhood voxel values ​​of the current voxel point, and the difference between the predicted value and the true value is the prediction residual. The coefficients of the three-dimensional predictive filter can be fixed to preset values ​​or obtained through adaptive training by analyzing the local statistical characteristics of the structured texture field. The prediction residual sequence usually has a smaller numerical range and higher concentration than the original voxel value sequence, so arithmetic encoding of it can achieve a high lossless compression ratio. It can be understood that regardless of which of the above three compression strategies is used, the compressed texture data will be reorganized and input into the WebGL rendering pipeline, where the shader program in the pipeline will perform decoding and volume rendering calculations. Optionally, when using a multi-level compression strategy based on wavelet transform, in order to adapt to the access characteristics of GPU texture memory, the multi-level subband coefficients may need to be rearranged and encapsulated in spatial order after entropy encoding. In some embodiments, the preset decomposition level of the three-dimensional discrete wavelet transform is determined by both the original size of the structured texture field and the compression level to ensure that the lowest frequency subband still retains an acceptable spatial resolution.

[0077] See Figure 5 In a WebGL-based radar volume rendering performance optimization system, the distribution characteristics of rendering efficiency (represented by normalized efficiency values) under different combinations of 3D texture block sizes (8×8×8, 16×16×16, etc.) and quantization bit depths (4 to 8 bits) are presented. Specifically, the vertical axis in the figure corresponds to the 3D size of the texture block, the horizontal axis represents the quantization bit depth, and the color and value of the color block represent the normalized rendering efficiency under the corresponding parameter combination: the higher the value, the better the rendering performance under that parameter configuration. From the distribution pattern, when the texture block size is small (e.g., 8×8×8) and the quantization bit depth is low (e.g., 4 bits), the normalized efficiency value is close to 0.89, which is at a relatively high level; while as the texture block size increases (e.g., 64×64×64) and the quantization bit depth increases (e.g., 8 bits), the efficiency value gradually decreases to 0.62. This result is consistent with the logic of the hybrid compression strategy based on block truncation coding: increasing the texture block size and quantization bit depth will improve data accuracy, but at the same time increase the rendering computation load, thereby reducing rendering efficiency. This figure can provide a direct basis for the selection of texture block size and quantization bit depth parameters under different global rendering load factors.

[0078] 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.

[0079] 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. A radar volume rendering performance optimization system based on WebGL, characterized in that, include: The texture acquisition and reconstruction module is used to acquire the original three-dimensional texture field of radar volume data and perform topological reconstruction based on voxel structure features on the original three-dimensional texture field to generate a structured texture field. The dynamic decomposition and clustering module is used to perform dynamic hierarchical decomposition based on energy diffusion simulation on the structured texture field to obtain multiple detail level units, and to perform adaptive clustering based on texture feature correlation on the multiple detail level units to form several rendering subsets. The metrics calculation module is used to calculate the texture change intensity and spatial consistency metrics for each rendering subset, and to calculate the global rendering load factor based on the texture change intensity and spatial consistency metrics of all rendering subsets. The strategy determination module is used to determine a texture compression strategy based on the global rendering load factor and a preset rendering parameter threshold. The compression rendering module is used to perform lossy compression encoding on the structured texture field according to the texture compression strategy, and input the compressed texture data into the WebGL rendering pipeline for volume rendering.

2. The radar volume rendering performance optimization system based on WebGL as described in claim 1, characterized in that, The method for acquiring the original three-dimensional texture field of radar volume data and performing topological reconstruction based on voxel structure features to generate a structured texture field includes: The spatial gradient vectors of each voxel point within the original 3D texture field are calculated to form a gradient field. Discrete divergence calculation is performed on the gradient field to obtain the structure tensor of each voxel point. Local anisotropy descriptors for voxel points are constructed using the eigenvalues ​​of the structure tensor. Region growing is performed on the voxel points based on the local anisotropy descriptors, merging voxel points with continuous characteristics into homogeneous regions. The boundary curvature and the uniformity of the internal voxel value distribution of each homogeneous region are calculated, and the boundary curvature and the uniformity are weighted and fused to generate the topological intensity value of the homogeneous region. The voxel points within the region are interpolated and resampled using the topological intensity value of each homogeneous region as a weight to generate a structured texture field.

3. The radar volume rendering performance optimization system based on WebGL as described in claim 2, characterized in that, Performing a dynamic hierarchical decomposition of the structured texture field based on energy diffusion simulation includes: A virtual energy field is constructed for the structured texture field, with each voxel point regarded as an energy node, and the voxel value mapped to the initial energy of the node; the energy transfer rule between adjacent nodes is defined, which is inversely proportional to the difference in voxel values ​​between nodes and directly proportional to the topological strength value of the node; the energy transfer process is iteratively executed until the energy distribution reaches a stable state; the nodes are sorted according to the energy values ​​of each node after stabilization, and the abrupt change point of the energy value is selected as the hierarchical segmentation threshold; the structured texture field is divided into multiple energy levels according to the hierarchical segmentation threshold, and each energy level corresponds to a detail level unit.

4. The WebGL-based radar volume rendering performance optimization system as described in claim 3, characterized in that, Performing adaptive clustering based on texture feature correlation on the plurality of detail level units includes: For each level-of-detail (LHD) unit, its texture spectral features are extracted. These features are obtained by performing a Fast Fourier Transform on the voxel value sequence within the unit and taking the amplitude of the low-frequency components. The cosine similarity of the texture spectral features between any two LHD units is calculated. A dynamic similarity threshold is set, which is negatively correlated with the expected value of the global rendering load factor. Connection edges are established with LHD units as vertices, and the cosine similarity between units is greater than the dynamic similarity threshold, forming an undirected graph. The largest complete subgraph is found in the undirected graph, and the set of LHD units contained in each largest complete subgraph constitutes a rendering subset.

5. The WebGL-based radar volume rendering performance optimization system as described in claim 4, characterized in that, Methods for calculating the intensity of texture variations in a rendered subset include: For the rendered subset, obtain all its level-of-detail units; calculate the standard deviation of voxel values ​​within each level-of-detail unit; take a weighted average of the standard deviations of all level-of-detail units, with the weight being the proportion of voxel points contained in the level-of-detail unit to the total number of voxel points in the rendered subset, to obtain the internal variation intensity of the rendered subset; calculate the variance of the difference in the mean voxel values ​​among the level-of-detail units within the rendered subset, as the inter-unit variation intensity of the rendered subset; add the internal variation intensity to the inter-unit variation intensity to obtain the texture variation intensity of the rendered subset.

6. The radar volume rendering performance optimization system based on WebGL as described in claim 5, characterized in that, Methods for calculating the spatial consistency metric of a rendered subset include: Calculate the spatial center coordinates of all detail level units in the rendered subset; calculate the covariance matrix of these spatial center coordinates; perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​in three principal directions; use the ratio of the minimum eigenvalue to the maximum eigenvalue as the spatial distribution dispersion of the rendered subset; calculate the Euclidean distance between each detail level unit and its spatial nearest neighbor unit on the texture spectral features, and take the average of all such distances as a measure of local feature continuity of the rendered subset; multiply the spatial distribution dispersion by the local feature continuity measure to obtain the spatial consistency index of the rendered subset, where a lower spatial consistency index indicates better spatial consistency.

7. The WebGL-based radar volume rendering performance optimization system as described in claim 6, characterized in that, Methods for calculating the global rendering load factor include: Traverse all rendering subsets and obtain the texture variation intensity and spatial consistency index for each subset. For each subset, calculate the product of its texture variation intensity and spatial consistency index as the subset rendering complexity. Multiply the subset rendering complexity by the total number of voxels contained in the subset to obtain the estimated load of the subset. Sum the estimated loads of all rendering subsets to obtain the total estimated load. Calculate the mean of the spatial consistency index of all rendering subsets as the average spatial consistency. Divide the total estimated load by the product of the average spatial consistency and a preset baseline factor to obtain the global rendering load factor.

8. The WebGL-based radar volume rendering performance optimization system as described in claim 7, characterized in that, The texture compression strategy determined based on the global rendering load factor and preset rendering parameter thresholds includes: The global rendering load factor is compared with preset high load thresholds and medium load thresholds. If the global rendering load factor is greater than the high load threshold, a multi-level compression strategy based on wavelet transform is adopted, and the compression level of the multi-level compression strategy is dynamically determined by the proportion by which the global rendering load factor exceeds the high load threshold. If the global rendering load factor is between the medium load threshold and the high load threshold, a hybrid compression strategy based on block truncation coding is adopted, and the block size and quantization bits during compression are obtained by linear interpolation of the global rendering load factor. If the global rendering load factor is less than the medium load threshold, a lossless compression strategy based on predictive coding is adopted.

9. The WebGL-based radar volume rendering performance optimization system as described in claim 8, characterized in that, Performing lossy compression encoding on the structured texture field according to the texture compression strategy includes: When a multi-level compression strategy based on wavelet transform is adopted, a three-dimensional discrete wavelet transform is performed on the structured texture field to obtain multi-level sub-band coefficients. According to the compression level, a corresponding number of low-frequency sub-band coefficients are retained, high-frequency sub-band coefficients are threshold quantized, high-frequency coefficients below the threshold are discarded, and entropy encoding is performed on the quantized coefficients. When a hybrid compression strategy based on block truncation coding is adopted, the structured texture field is divided into multiple texture blocks of a size determined by the global rendering load factor. The mean and range of voxel values ​​are calculated for each texture block, and the voxel values ​​within the block are recalibrated using the quantization bit depth. When a lossless compression strategy based on predictive coding is adopted, a three-dimensional predictive filter is used to process the structured texture field, and arithmetic coding is performed on the prediction residuals.

10. The WebGL-based radar volume rendering performance optimization system as described in claim 9, characterized in that, When employing a multi-level compression strategy based on wavelet transform, the steps of performing a three-dimensional discrete wavelet transform on the structured texture field to obtain multi-level sub-band coefficients include: Select a preset wavelet basis function, which has compact support and regularity; The structured texture field is subjected to one-dimensional wavelet transform in three spatial dimensions. The transform in each dimension includes low-pass filtering and high-pass filtering of the voxel data sequence, and downsampling of the filtering results. The transformation results in the three dimensions are combined by tensor product to generate a three-dimensional wavelet coefficient matrix. The repeated transformation process decomposes the low-frequency subband into multiple levels until the preset decomposition level is reached, resulting in a three-dimensional subband coefficient set containing multiple resolution levels.

Images