Method and system for rendering three-dimensional images of industrial products based on virtual reality

Abstract

The invention relates to the field of image rendering, and in particular to a method and system for rendering a three-dimensional image of an industrial product based on virtual reality. The system comprises a data acquisition module, a key point extraction module, a node graph construction module, a node graph optimization module, a plane graph reconstruction module and a three-plane coding rendering module. The invention adopts a three-plane coding process based on distortion correction to simplify and optimize the image of the industrial product, adopts distortion correction to reconstruct the simplified image, adopts three-plane coding process, performs three-plane construction and optimization and three-dimensional reconstruction on the reconstructed simplified image, reduces rendering node parameters, and ultimately increases rendering efficiency; the invention creatively adopts an optimization method based on graph embedding to obtain a distinct relationship between a graph topological structure and nodes, optimizes the connection of nodes by using Mahalanobis distance, reduces the number of nodes, and based on this, realizes reconstruction optimization of a plane graph to be rendered, renders the optimized reconstructed graph, and has higher accuracy and faster efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of image rendering, and in particular relates to a method and system for rendering three-dimensional images of industrial products based on virtual reality. Background Art

[0002] With the development of virtual reality technology, VR-based industrial product three-dimensional image rendering methods and systems have become increasingly important in the fields of industrial design, product display, assembly simulation, etc. It can provide users with an immersive interactive experience and display the details of industrial products through realistic three-dimensional rendering; however, the existing technology is too dependent on three-dimensional images when rendering three-dimensional images, and the existing three-dimensional rendering technology is not mature enough, resulting in low efficiency in three-dimensional image rendering. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for rendering a three-dimensional image of an industrial product based on virtual reality. In view of the fact that the prior art is too dependent on three-dimensional images when rendering three-dimensional images, and the efficiency of three-dimensional image rendering is low, the present invention creatively adopts a three-plane coding process based on distortion correction, and through innovative node graph optimization, simplifies and optimizes the image of the industrial product to be rendered, and reconstructs the simplified image based on simplification to maintain the original linear characteristics, and on the basis of preventing the distortion of the product rendering image, adopts three-plane coding processing, constructs and optimizes the reconstructed simplified image in three planes, and finally obtains a three-dimensional rendering image through volume rendering decoding, reduces rendering node parameters, and increases rendering efficiency through plane-to-three-dimensional conversion; the present invention creatively adopts an optimization method based on graph embedding to obtain a distinct relationship between the graph topology and nodes, optimizes the connection of nodes by Mahalanobis distance, reduces the number of nodes, and on this basis realizes the reconstruction and optimization of the plane image to be rendered, and renders the optimized reconstructed image with higher accuracy and faster efficiency.

[0004] The technical solution adopted by the present invention is as follows: a method for rendering a three-dimensional image of an industrial product based on virtual reality, specifically comprising the following steps:

[0005] Step S1: image acquisition, constructing a virtual image of the industrial product to be rendered by using an image drawing tool, and acquiring a plan view of the virtual image of the industrial product to be rendered, i.e., a picture of the product to be rendered, by using a camera;

[0006] Step S2: key point extraction, using SIFT algorithm to extract key points and local features of key points from the rendered product image, specifically including the following steps;

[0007] Step S21: Scale space extreme value detection, using Gaussian difference pyramid to construct scale space, and detecting extreme value points in the product image to be rendered layer by layer, that is, finding pixels with maximum or minimum values ​​in the scale space;

[0008] Step S22: key point positioning, accurately positioning the key points around the extreme points in the scale space by fitting a quadratic function;

[0009] Step S23: Orienting the key points, calculating the gradient direction and magnitude for the neighborhood around each key point, and determining the main direction so that the key point has rotation invariance;

[0010] Step S24: extracting local features of key points, constructing a gradient histogram in an n×n neighborhood around the key points, and performing feature extraction on the gradient histogram to obtain local features of all key points;

[0011] Step S3: Node graph construction: construct a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, where each node in the node graph is a key point;

[0012] Step S4: node graph optimization, performing graph embedding-based optimization on the node graph to obtain an optimized node graph and optimized local features;

[0013] Step S5: plane graph reconstruction, by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, a plane diagram is obtained;

[0014] Step S6: Based on the three-plane encoding processing of distortion correction, the product image to be rendered and the plane diagram are processed by single-view three-plane encoder respectively, and the plane diagram is subjected to distortion correction and fusion decoding based on the product image to be rendered to obtain a three-dimensional rendering image;

[0015] Step S7: loss optimization, performing loss calculation and optimization on the three-plane encoding process based on distortion correction according to the rendering effect of the three-dimensional rendering image.

[0016] Further, step S4 specifically includes the following steps:

[0017] Step S41: Initialize each node in the node graph , The set of adjacent nodes is defined as ;

[0018] Step S42: Feature aggregation, node The corresponding local features are aggregated and activated with the local features of its adjacent nodes to obtain the aggregated features:

[0019] ;

[0020] In the formula, represents the aggregation feature, Representation Node All adjacent nodes of Features, represents the feature aggregation operation, Representation Node Features, represents the feature connection operation, represents the learnable weight matrix, Represents the Sigmoid activation function;

[0021] Step S43: Embedding feature calculation:

[0022] ;

[0023] In the formula, Represents a splicing operation, represents the learnable weight matrix, represents the embedded features;

[0024] Step S44: Calculate the Mahalanobis distance, calculate the Mahalanobis distance between the embedded features of all nodes, connect two nodes whose Mahalanobis distance is less than a preset threshold to obtain an optimized node graph, and average the local features of the key points corresponding to the two nodes to obtain the optimized local features.

[0025] Further, step S6 specifically includes the following steps:

[0026] Step S61: single-view three-plane encoder processing, performing single-view three-plane encoder processing on the product image to be rendered and the planar diagram respectively to obtain an original three-plane representation and a simplified three-plane representation;

[0027] Step S62: distortion correction optimization, correcting the distortion of the simplified three-plane representation through the original three-plane representation to obtain a corrected simplified three-plane representation;

[0028] Step S63: three-plane fusion, weighted fusion of the original three-plane representation and the corrected simplified three-plane representation to obtain an optimized simplified three-plane;

[0029] Step S64: three-plane decoding, performing volume rendering decoding on the optimized and simplified three-planes to obtain a three-dimensional optimized RGB image, that is, a three-dimensional rendering image.

[0030] Further, in step S61, the single-view three-plane encoder processing specifically includes the following steps:

[0031] Step S611: low-resolution feature extraction, using dilated convolution to extract features from the plane diagram to obtain low-resolution features;

[0032] Step S612: global inductive bias processing, using the VIT model and the convolutional network to perform global inductive bias processing on the low-resolution features to obtain optimized low-resolution features;

[0033] Step S613: extracting high-resolution features, using an encoder to extract features from the plane diagram to obtain high-resolution features;

[0034] Step S614: feature fusion decoding, fusing the optimized low-resolution features with the high-resolution features, and performing global inductive bias processing to obtain a three-plane representation.

[0035] Further, step S62 specifically includes the following steps:

[0036] Step S621: Calculate the camera intrinsic parameters, using the geometric shape of the product image to be rendered and the pixel coordinates of the nodes in the original three-plane representation to estimate the camera intrinsic parameters through a camera calibration algorithm;

[0037] Step S622: constructing a radial distortion model and establishing a mathematical correction formula for distortion correction, including radial distortion coefficients and distortion centers;

[0038] Step S623: pixel coordinate conversion, for each pixel point in the simplified three-plane representation, the pixel coordinates in the simplified three-plane representation are converted into corrected coordinates according to the distortion model and the intrinsic parameters of the camera to obtain the corrected simplified three-plane representation.

[0039] The virtual reality-based industrial product three-dimensional image rendering system provided by the present invention includes a data acquisition module, a key point extraction module, a node graph construction module, a node graph optimization module, a plane graph reconstruction module and a three-plane coding rendering module;

[0040] The data acquisition module performs virtual image drawing and plane image acquisition on the industrial product to be rendered to obtain a product image to be rendered, and sends the product image to be rendered to the key point extraction module;

[0041] The key point extraction module uses the SIFT algorithm to extract key points and local features of key points from the rendered product image, and sends the key points and local features of key points to the node graph construction module and the node graph optimization module;

[0042] The node graph construction module constructs a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, and sends the node graph to the node graph optimization module;

[0043] The node graph optimization module performs a graph embedding-based optimization on the node graph to obtain an optimized node graph and optimized local features, and sends them to the plane graph reconstruction module;

[0044] The plane graph reconstruction module obtains a plane diagram by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, and sends the plane diagram to the three-plane coding and rendering module;

[0045] The three-plane encoding rendering module performs single-view three-plane encoder processing on the product image to be rendered and the plane diagram respectively, performs distortion correction and fusion decoding on the plane diagram based on the product image to be rendered to obtain a three-dimensional rendering image, and calculates and optimizes the loss of the three-plane encoding processing based on the distortion correction according to the rendering effect of the three-dimensional rendering image.

[0046] The beneficial results achieved by the present invention using the above scheme are as follows:

[0047] (1) In view of the fact that the prior art relies too much on three-dimensional images when rendering three-dimensional images, and the efficiency of three-dimensional image rendering is low, the present invention adopts a three-plane encoding process based on distortion correction, simplifies and optimizes the image of the industrial product to be rendered through innovative node graph optimization, and reconstructs the simplified image based on simplification to maintain the original linear characteristics. On the basis of preventing the distortion of the product rendering image, three-plane encoding processing is adopted to construct and optimize the reconstructed simplified image in three planes, and finally a three-dimensional rendering image is obtained through volume rendering decoding, thereby reducing rendering node parameters and increasing rendering efficiency through plane-to-three-dimensional conversion;

[0048] (2) The present invention adopts an optimization method based on graph embedding to obtain a distinct relationship between the graph topology and nodes, optimizes the connection of nodes using Mahalanobis distance, reduces the number of nodes, and based on this, realizes the reconstruction optimization of the plane graph to be rendered, and renders the optimized reconstructed graph with higher accuracy and faster efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram of the process of the virtual reality-based industrial product three-dimensional image rendering method provided by the present invention;

[0050] Figure 2 A schematic diagram of the modules of the virtual reality-based industrial product three-dimensional image rendering system provided by the present invention.

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0053] Example 1, see Figure 1 The method for rendering a three-dimensional image of an industrial product based on virtual reality provided by the present invention specifically comprises the following steps:

[0054] Step S1: image acquisition, constructing a virtual image of the industrial product to be rendered by using an image drawing tool, and acquiring a plan view of the virtual image of the industrial product to be rendered, i.e., a picture of the product to be rendered, by using a camera;

[0055] Step S2: key point extraction, using SIFT algorithm to extract key points and local features of key points from the rendered product image;

[0056] Step S3: Node graph construction: construct a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, where each node in the node graph is a key point;

[0057] Step S4: node graph optimization, performing a graph embedding-based optimization on the node graph to obtain an optimized node graph and optimized local features;

[0058] Step S5: plane graph reconstruction, by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, a plane diagram is obtained;

[0059] Step S6: Based on the three-plane encoding processing of distortion correction, the product image to be rendered and the plane diagram are processed by single-view three-plane encoder respectively, and the plane diagram is subjected to distortion correction and fusion decoding based on the product image to be rendered to obtain a three-dimensional rendering image;

[0060] Step S7: loss optimization, performing loss calculation and optimization on the three-plane encoding process based on distortion correction according to the rendering effect of the three-dimensional rendering image.

[0061] By performing the above operations, the prior art over-relies on three-dimensional images when rendering three-dimensional images, and the efficiency of three-dimensional image rendering is low. The present invention creatively adopts a three-plane encoding process based on distortion correction, and simplifies and optimizes the image of the industrial product to be rendered through innovative node graph optimization, and reconstructs the simplified image based on simplification to maintain the original linear characteristics, and adopts three-plane encoding process on the basis of preventing the distortion of the product rendering image, and constructs and optimizes the reconstructed simplified image in three planes, and finally obtains the three-dimensional rendering image through volume rendering decoding, reduces the rendering node parameters, and increases the rendering efficiency through the plane-to-three-dimensional conversion.

[0062] Embodiment 2: This embodiment is based on the above embodiment, and step S4 specifically includes the following steps:

[0063] Step S41: Initialize each node in the node graph , The set of adjacent nodes is defined as ;

[0064] Step S42: Feature aggregation, node The corresponding local features are aggregated and activated with the local features of its adjacent nodes to obtain the aggregated features:

[0065] ;

[0066] In the formula, represents the aggregation feature, Representation Node All adjacent nodes of Features, represents the feature aggregation operation, Representation Node Features, represents the feature connection operation, represents the learnable weight matrix, Represents the Sigmoid activation function;

[0067] Step S43: Embedding feature calculation:

[0068] ;

[0069] In the formula, Represents a splicing operation, represents the learnable weight matrix, represents the embedded features;

[0070] Step S44: Calculate the Mahalanobis distance, calculate the Mahalanobis distance between the embedded features of all nodes, connect two nodes whose Mahalanobis distance is less than a preset threshold to obtain an optimized node graph, and average the local features of the key points corresponding to the two nodes to obtain the optimized local features.

[0071] By performing the above operations, the present invention creatively adopts an optimization method based on graph embedding to obtain a distinct relationship between the graph topology and nodes, optimizes the connection of nodes using Mahalanobis distance, reduces the number of nodes, and based on this, realizes the reconstruction optimization of the plane map to be rendered, and renders the optimized reconstructed map with higher accuracy and faster efficiency.

[0072] Embodiment 3: This embodiment is based on the above embodiment, and step S6 specifically includes the following steps:

[0073] Step S61: single-view three-plane encoder processing, performing single-view three-plane encoder processing on the product image to be rendered and the planar diagram respectively to obtain an original three-plane representation and a simplified three-plane representation;

[0074] Step S62: distortion correction optimization, correcting the distortion of the simplified three-plane representation through the original three-plane representation to obtain a corrected simplified three-plane representation;

[0075] Step S63: three-plane fusion, weighted fusion of the original three-plane representation and the corrected simplified three-plane representation to obtain an optimized simplified three-plane;

[0076] Step S64: three-plane decoding, performing volume rendering decoding on the optimized and simplified three-planes to obtain a three-dimensional optimized RGB image, that is, a three-dimensional rendering image.

[0077] Embodiment 4: This embodiment is based on the above embodiment, and step S61 specifically includes the following steps:

[0078] Step S611: low-resolution feature extraction, using dilated convolution to extract features from the plane diagram to obtain low-resolution features;

[0079] Step S612: global inductive bias processing, using the VIT model and the convolutional network to perform global inductive bias processing on the low-resolution features to obtain optimized low-resolution features;

[0080] Step S613: extracting high-resolution features, using an encoder to extract features from the plane diagram to obtain high-resolution features;

[0081] Step S614: feature fusion decoding, fusing the optimized low-resolution features with the high-resolution features, and performing global inductive bias processing to obtain a three-plane representation.

[0082] Embodiment 5: This embodiment is based on the above embodiment, and step S62 specifically includes the following steps:

[0083] Step S621: Calculate the camera intrinsic parameters, using the geometric shape of the product image to be rendered and the pixel coordinates of the nodes in the original three-plane representation to estimate the camera intrinsic parameters through a camera calibration algorithm;

[0084] Step S622: constructing a radial distortion model and establishing a mathematical correction formula for distortion correction, including radial distortion coefficients and distortion centers;

[0085] Step S623: pixel coordinate conversion, for each pixel point in the simplified three-plane representation, the pixel coordinates in the simplified three-plane representation are converted into corrected coordinates according to the distortion model and the intrinsic parameters of the camera to obtain the corrected simplified three-plane representation.

[0086] Example 6, see Figure 2 ,This embodiment is based on the above embodiment. The virtual reality-based industrial product three-dimensional image rendering system includes a data acquisition module, a key point extraction module, a node graph construction module, a node graph optimization module, a plane graph reconstruction module and a three-plane coding rendering module;

[0087] The data acquisition module performs virtual image drawing and plane image acquisition on the industrial product to be rendered to obtain a product image to be rendered, and sends the product image to be rendered to the key point extraction module;

[0088] The key point extraction module uses the SIFT algorithm to extract key points and local features of key points from the rendered product image, and sends the key points and local features of key points to the node graph construction module and the node graph optimization module;

[0089] The node graph construction module constructs a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, and sends the node graph to the node graph optimization module;

[0090] The node graph optimization module optimizes the node graph based on graph embedding to obtain an optimized node graph and optimized local features, and sends them to the plane graph reconstruction module;

[0091] The plane graph reconstruction module obtains a plane diagram by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, and sends the plane diagram to the three-plane coding and rendering module;

[0092] The three-plane encoding rendering module performs single-view three-plane encoder processing on the product image to be rendered and the plane diagram respectively, performs distortion correction and fusion decoding on the plane diagram based on the product image to be rendered to obtain a three-dimensional rendering image, and calculates and optimizes the loss of the three-plane encoding processing based on the distortion correction according to the rendering effect of the three-dimensional rendering image.

[0093] Embodiment 7: This embodiment is based on the above embodiment, and the optimization implementation of step S62 is as follows:

[0094] Distortion model:

[0095] Based on the characteristics of radial distortion, a polynomial function is used to model radial distortion.

[0096] Parameter estimates:

[0097] Using known industrial product images, detect feature points and use these feature points to estimate the camera’s distortion parameters;

[0098] Determine the center of distortion:

[0099] Determine the center of distortion, which is the position of the camera's principal point;

[0100] Distortion correction formula:

[0101] According to the selected radial distortion model, a mathematical model of distortion correction is established, and the correction formula includes radial distortion coefficient and distortion center;

[0102] Pixel coordinate conversion:

[0103] For each pixel point, the pixel coordinates are converted into corrected coordinates according to the distortion model and distortion parameters;

[0104] Interpolation:

[0105] For the corrected coordinates, bilinear interpolation is performed to obtain the corrected pixel values;

[0106] Boundary processing:

[0107] Process the boundary pixels of the corrected image and use mirroring and filling methods to avoid blank areas;

[0108] Apply to the entire image:

[0109] Apply radial distortion correction to each pixel in the image to obtain the correction effect for the entire image;

[0110] Optimization and Evaluation:

[0111] The distortion parameters are optimized by minimizing the reprojection error, and whether the correction effect meets expectations is evaluated.

[0112] Embodiment 8: This embodiment is based on the above embodiment, and the optimization of SIFT algorithm is implemented as follows:

[0113] Scale space extrema detection:

[0114] Use Gaussian difference pyramid to construct scale space and detect extreme points in the product plane image layer by layer, that is, find the pixel points with maximum or minimum value in the scale space;

[0115] Key point positioning:

[0116] Around the extreme points in the scale space, the positions of key points are accurately located by fitting quadratic functions. The positions in industrial products usually reflect the corner points in the dimensional image.

[0117] Key point screening:

[0118] Unstable or unrepresentative key points are screened out based on stability criteria, namely gradient amplitude, gradient direction, and local image contrast;

[0119] Key point orientation:

[0120] The gradient direction and magnitude are calculated for the neighborhood around each key point, and then the main direction is determined to make the key point rotation invariant;

[0121] Key points description:

[0122] A gradient histogram is constructed in the neighborhood around the key point to describe the local features around the key point. The SIFT descriptor is a 128-dimensional vector used to describe the appearance characteristics of the key point.

[0123] Normalize the eigenvector:

[0124] Normalize the SIFT descriptor to reduce the impact of illumination changes and noise;

[0125] Keypoint matching:

[0126] The distance metric between descriptors, Euclidean distance, is used to match keypoints in different images.

[0127] Embodiment 9: Based on the above embodiment, this embodiment performs loss calculation and optimization on three-plane encoding processing based on distortion correction as follows:

[0128] Data preparation:

[0129] Input: single view RGB image of product;

[0130] Output: target representation of the 3D model, i.e. 3D rendering;

[0131] Loss function:

[0132] Geometric loss: used to optimize the generation quality of 3D structures and calculate IoU loss;

[0133] Texture Loss:

[0134] Use perceptual loss to optimize color consistency;

[0135] Projection consistency loss:

[0136] Ensure that the generated 3D model is consistent with the real image when projected from different viewing angles.

[0137] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0138] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0139] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for rendering a three-dimensional image of an industrial product based on virtual reality, characterized in that: The specific steps include: Step S1: image acquisition, constructing a virtual image of the industrial product to be rendered by using an image drawing tool, and acquiring a plan view of the virtual image of the industrial product to be rendered, i.e., a picture of the product to be rendered, by using a camera; Step S2: Key point extraction: Use SIFT algorithm to extract key points and local features of key points in the rendered product image. The steps include: Step S21: Scale space extreme value detection, using Gaussian difference pyramid to construct scale space, and detecting extreme value points in the product image to be rendered layer by layer, that is, finding pixels with maximum or minimum values ​​in the scale space; Step S22: key point positioning, accurately positioning the key points around the extreme points in the scale space by fitting a quadratic function; Step S23: Orienting the key points, calculating the gradient direction and magnitude for the neighborhood around each key point, and determining the main direction so that the key point has rotation invariance; Step S24: extracting local features of key points, constructing a gradient histogram in an n×n neighborhood around the key points, and performing feature extraction on the gradient histogram to obtain local features of all key points; Step S3: Node graph construction: construct a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, where each node in the node graph is a key point; Step S4: node graph optimization, performing a graph embedding-based optimization on the node graph to obtain an optimized node graph and optimized local features; Step S5: plane graph reconstruction, by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, a plane diagram is obtained; Step S6: Based on the three-plane encoding processing of distortion correction, the product image to be rendered and the plane diagram are respectively processed by a single-view three-plane encoder, and the plane diagram is subjected to distortion correction and fusion decoding based on the product image to be rendered to obtain a three-dimensional rendering image; Step S7: loss optimization, performing loss calculation and optimization on the three-plane encoding process based on distortion correction according to the rendering effect of the three-dimensional rendering image.

2. The method for rendering a three-dimensional image of an industrial product based on virtual reality according to claim 1, characterized in that: Step S4 The specific steps include: Step S41: Initialize each node in the node graph , The set of adjacent nodes is defined as ; Step S42: Feature aggregation, node The corresponding local features are aggregated and activated with the local features of its adjacent nodes to obtain the aggregated features: ； In the formula, represents the aggregation feature, Representation Node All adjacent nodes of Features, represents the feature aggregation operation, Representation Node Features, represents the feature connection operation, represents the learnable weight matrix, Represents the Sigmoid activation function; Step S43: Embedding feature calculation: ； In the formula, Represents a splicing operation, represents the learnable weight matrix, represents the embedded features; Step S44: Calculate the Mahalanobis distance, calculate the Mahalanobis distance between the embedded features of all nodes, connect two nodes whose Mahalanobis distance is less than a preset threshold to obtain an optimized node graph, and average the local features of the key points corresponding to the two nodes to obtain the optimized local features.

3. The method for rendering a three-dimensional image of an industrial product based on virtual reality according to claim 2, characterized in that: Step S6 specifically includes the following steps: Step S61: single-view three-plane encoder processing, performing single-view three-plane encoder processing on the product image to be rendered and the planar diagram respectively to obtain an original three-plane representation and a simplified three-plane representation; Step S62: distortion correction optimization, correcting the distortion of the simplified three-plane representation through the original three-plane representation to obtain a corrected simplified three-plane representation; Step S63: three-plane fusion, weighted fusion of the original three-plane representation and the corrected simplified three-plane representation to obtain an optimized simplified three-plane; Step S64: three-plane decoding, performing volume rendering decoding on the optimized and simplified three-planes to obtain a three-dimensional optimized RGB image, that is, a three-dimensional rendering image.

4. The method for rendering a three-dimensional image of an industrial product based on virtual reality according to claim 3, characterized in that: In step S61, the single-view three-plane encoder process specifically includes the following steps: Step S611: low-resolution feature extraction, using dilated convolution to extract features from the plane diagram to obtain low-resolution features; Step S612: global inductive bias processing, using the VIT model and the convolutional network to perform global inductive bias processing on the low-resolution features to obtain optimized low-resolution features; Step S613: extracting high-resolution features, using an encoder to extract features from the plane diagram to obtain high-resolution features; Step S614: feature fusion decoding, fusing the optimized low-resolution features with the high-resolution features, and performing global inductive bias processing to obtain a three-plane representation.

5. The method for rendering a three-dimensional image of an industrial product based on virtual reality according to claim 4, characterized in that: In step S62, the distortion correction optimization specifically includes the following steps: Step S621: Calculate the camera intrinsic parameters, using the geometric shape of the product image to be rendered and the pixel coordinates of the nodes in the original three-plane representation to estimate the camera intrinsic parameters through a camera calibration algorithm; Step S622: constructing a radial distortion model and establishing a mathematical correction formula for distortion correction, including radial distortion coefficients and distortion centers; Step S623: pixel coordinate conversion, for each pixel point in the simplified three-plane representation, the pixel coordinates in the simplified three-plane representation are converted into corrected coordinates according to the distortion model and the intrinsic parameters of the camera to obtain the corrected simplified three-plane representation.

6. A virtual reality-based industrial product three-dimensional image rendering system, used to implement the virtual reality-based industrial product three-dimensional image rendering method according to any one of claims 1 to 5, characterized in that: It includes data acquisition module, key point extraction module, node graph construction module, node graph optimization module, plane graph reconstruction module and three-plane coding rendering module; The data acquisition module performs virtual image drawing and plane image acquisition on the industrial product to be rendered to obtain a product image to be rendered, and sends the product image to be rendered to the key point extraction module; The key point extraction module uses the SIFT algorithm to extract key points and local features of key points from the rendered product image, and sends the key points and local features of key points to the node graph construction module and the node graph optimization module; The node graph construction module constructs a node graph according to all key points of the product image to be rendered and the connectivity relationship between the key points, and sends the node graph to the node graph optimization module; The node graph optimization module optimizes the node graph based on graph embedding to obtain an optimized node graph and optimized local features, and sends them to the plane graph reconstruction module; The plane graph reconstruction module obtains a plane diagram by activating and outputting the optimized node graph and the optimized local features and reconstructing the graph, and sends the plane diagram to the three-plane coding and rendering module; The three-plane encoding rendering module performs single-view three-plane encoder processing on the product image to be rendered and the plane diagram respectively, performs distortion correction and fusion decoding on the plane diagram based on the product image to be rendered to obtain a three-dimensional rendering image, and calculates and optimizes the loss of the three-plane encoding processing based on the distortion correction according to the rendering effect of the three-dimensional rendering image.

Images