Data visualization display method and system based on business model

Abstract

The embodiment of the invention discloses a data visualization display method and system based on a business model, and belongs to the technical field of digital cities. According to the embodiment of the invention, the three-dimensional surveying and mapping data including various types of data such as terrains, ground features and textures can be effectively processed, the visual element vectors are mined through the first model visual mapping network, accurate input is provided for subsequent rendering, and the visualization efficiency is improved. And the model visual rendering network performs rendering operation on the basis of the mined vectors to realize visual display of the target digital city area. The two networks are jointly debugged based on the first training error and the second training error, the network performance can be accurately optimized, the accuracy and fidelity of the visualization effect can be improved, the spatial features of the digital city can be better presented, and the requirements of multiple aspects such as city planning and management can be met.

Description

Technical Field

[0001] The embodiments of the present invention belong to the technical field of digital cities, and particularly relate to a method and system for data visualization display based on a business model. Background Art

[0002] With the development of digital cities, the demand for three-dimensional model visualization display is increasing day by day. When traditional technologies process three-dimensional surveying and mapping data, they face many problems. On the one hand, it is difficult to effectively integrate and utilize complex terrain, ground object, and texture data, resulting in low efficiency in processing the basic data for visualization. On the other hand, there is a lack of effective mapping and rendering mechanisms in the process of converting data into visualization, and the visualization effect is often not accurate and realistic enough. At the same time, there is no debugging method that can accurately optimize the relevant network to improve the visualization effect, and it cannot meet the requirements for accurate visualization of digital cities in fields such as urban planning and management. Summary of the Invention

[0003] The embodiments of the present invention provide a method and system for data visualization display based on a business model, which can solve or partially solve the technical problems involved in the above background art.

[0004] The embodiments of the present invention provide a method for data visualization display based on a business model, which is applied to a data visualization display system. The method includes: obtaining to-be-processed three-dimensional surveying and mapping data, where the to-be-processed three-dimensional surveying and mapping data is used to describe the multi-modal spatial representation vector of a target digital city area, and the to-be-processed three-dimensional surveying and mapping data includes terrain data, ground object data, and texture data; performing visual element vector mining on the to-be-processed three-dimensional surveying and mapping data through a first model visual mapping network to obtain the visual element vector of the to-be-processed three-dimensional surveying and mapping data; performing a visual rendering operation on the visual element vector of the to-be-processed three-dimensional surveying and mapping data through a model visual rendering network to obtain a target visual rendering output result; the target visual rendering output result is used to realize the model visualization display of the target digital city area; wherein, the model visual rendering network and the first model visual mapping network are jointly debugged based on a first training error and a second training error.

[0005] The embodiments of the present invention provide a data visualization display system, including at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the above method.

[0006] The embodiments of the present invention provide a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] The embodiments of the present invention can effectively process 3D surveying and mapping data containing multiple types of data such as terrain, features, and textures, mine visual element vectors through the first model visual mapping network, provide accurate input for subsequent rendering, and improve visualization efficiency. The model visual rendering network performs rendering operations based on the mined vectors to achieve the visual display of the target digital city area. By jointly debugging the two networks based on the first training error and the second training error, the network performance can be accurately optimized, the accuracy and fidelity of the visualization effect can be improved, the spatial characteristics of the digital city can be better presented, and the requirements in multiple aspects such as urban planning and management can be met. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flowchart of a data visualization display method based on a service model provided by an embodiment of the present invention.

[0009] Figure 2 It is a schematic structural diagram of a data visualization display system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] Figure 1 A data visualization display method based on a service model is shown, which is applied to a data visualization display system. The method includes the following steps 101 to 103.

[0011] Step 101: Obtain the to-be-processed 3D surveying and mapping data, where the to-be-processed 3D surveying and mapping data is used to describe the multi-modal spatial representation vector of the target digital city area, and the to-be-processed 3D surveying and mapping data includes terrain data, feature data, and texture data.

[0012] Step 102: Mine visual element vectors of the to-be-processed 3D surveying and mapping data through the first model visual mapping network to obtain the visual element vectors of the to-be-processed 3D surveying and mapping data.

[0013] Step 103: Perform visual rendering operations on the visual element vectors of the to-be-processed 3D surveying and mapping data through the model visual rendering network to obtain a target visual rendering output result; the target visual rendering output result is used to achieve the model visual display of the target digital city area.

[0014] Among them, the model visual rendering network and the first model visual mapping network are jointly debugged based on the first training error and the second training error.

[0015] Based on the above steps 101 - 103, the debugging steps of the first model visual mapping network and the model visual rendering network include: obtaining X visual rendering output samples and Y 3D surveying and mapping data samples included in the current debugging sample set, and performing visual feature vector mining on each 3D surveying and mapping data sample through the first model visual mapping network to obtain the first visual feature vector of each 3D surveying and mapping data sample, and performing visual feature vector mining on each visual rendering output sample to obtain the first visual sample vector of each visual rendering output sample. The 3D surveying and mapping data samples are used to describe the multi-modal spatial representation vectors of the sample digital city area, and both X and Y are positive integers; determining the first training error based on the first visual feature vectors of the Y 3D surveying and mapping data samples and the first visual sample vectors of the X visual rendering output samples; performing visual rendering operations on the first visual feature vectors of the Y 3D surveying and mapping data samples through the model visual rendering network to obtain the visual rendering training results corresponding to the Y 3D surveying and mapping data samples respectively, and determining the second training error based on the visual rendering training results corresponding to the Y 3D surveying and mapping data samples respectively; and jointly debugging the first model visual mapping network and the model visual rendering network based on the first training error and the second training error.

[0016] As the execution entity of the embodiment of the present invention, the data visualization display system first obtains the 3D surveying and mapping data to be processed in step 101. These 3D surveying and mapping data to be processed are important information sources for describing the multi-modal spatial representation vectors of the target digital city area, covering various aspects of data such as terrain data, ground object data, and texture data. The terrain data can accurately represent the terrain undulation of the target digital city area. For example, the height of the hills in the city may be reflected by specific numerical values, such as hills with altitudes ranging from 100 meters to 500 meters, and their terrain features such as slope and aspect are also included. The ground object data includes various artificial structures in the city, such as the 3D shape, size, and location of buildings. For example, a commercial center building may cover an area of tens of thousands of square meters and have a height of dozens of stories; the data of roads includes their width, direction, etc. For example, the width of the main urban roads may be between 50 meters and 100 meters, and the width of the branch roads may be between 10 meters and 20 meters, etc. The texture data endows the ground objects with a realistic appearance. For example, the material texture of the building wall may be marble texture or brick texture, and the road surface texture may be asphalt road texture or cement road texture, etc.

[0017] Next, enter step 102. The data visualization display system uses the first model visual mapping network to mine visual element vectors from the three-dimensional surveying and mapping data to be processed. Visual element vector mining is a complex and crucial process. Through the first model visual mapping network, visual element vectors can be extracted from the three-dimensional surveying and mapping data containing various data such as terrain, features, and textures to be processed. Taking the buildings in a city as an example, the first model visual mapping network can mine the element vectors related to visualization from the comprehensive information such as the three-dimensional shape data of the buildings and the texture data on their surfaces. These vectors may contain the key features of the buildings in terms of visual presentation, such as the contour feature vectors and color feature vectors of the buildings from a specific perspective. Through such a mining process, the original complex three-dimensional surveying and mapping data can be transformed into a form of visual element vectors that is more convenient for subsequent visualization processing, thus laying a foundation for realizing the model visualization display of the target digital city area.

[0018] In step 103, the data visualization display system uses the model visual rendering network to perform visual rendering operations on the visual element vectors of the three-dimensional surveying and mapping data to be processed obtained previously. The visual rendering operation is a key step in transforming the mined visual element vectors into the target visual rendering output results that can be directly presented on the display device. In this process, the model visual rendering network will perform operations such as lighting effect calculation, color adjustment, and spatial layout optimization according to various feature information in the visual element vectors. For example, for the park area in a city, according to the visual element vectors of the grassland and lake in its terrain data, a realistic green grassland and blue lake are rendered, and the shadow effects on the grassland and lake are calculated according to the lighting direction, and the lake water surface is rendered according to the reflected light of the surrounding buildings, making the lake water surface look sparkling. The final target visual rendering output results can realize the model visualization display of the target digital city area, allowing users to intuitively see the three-dimensional model of the digital city area.

[0019] For the debugging process of the first model visual mapping network and the model visual rendering network, this is an important link to ensure the accuracy and effectiveness of the entire visualization system. During the debugging process, the data visualization display system first obtains the current debugging sample set. This debugging sample set contains X visual rendering output samples and Y three-dimensional surveying and mapping data samples, where both X and Y are positive integers. For example, X can take the value of 100 and Y takes the value of 50. For each three-dimensional surveying and mapping data sample, its multi-modal space representation vector used to describe the sample digital city area, similar to the three-dimensional surveying and mapping data of the target digital city area mentioned above, contains data in multiple aspects such as terrain, features, and textures.

[0020] The data visualization display system performs visual feature vector mining on each 3D mapping data sample through the first model visual mapping network to obtain the first visual feature vector of each 3D mapping data sample. At the same time, visual feature vector mining is also performed on each visual rendering output sample to obtain the first visual sample vector of each visual rendering output sample. For example, for a visual rendering output sample that is the visualization result of a certain historical block in a city, the first visual sample vector mined may include the color distribution feature vector of the block buildings, the architectural style feature vector, etc.

[0021] Based on the first visual feature vectors of these Y 3D mapping data samples and the first visual sample vectors of these X visual rendering output samples, the first training error is determined. The determination of this first training error may involve the measurement of the difference between the two vectors. For example, comparing the degree of difference between the two in key visual feature characteristics. If there is a large deviation between the first visual feature vector of the 3D mapping data sample and the first visual sample vector of the visual rendering output sample in a certain visual feature characteristic, such as the representation of building height, it will be reflected in the first training error.

[0022] Then, the data visualization display system performs visual rendering operations on the first visual feature vectors of the Y 3D mapping data samples through the model visual rendering network to obtain the visual rendering training results corresponding to the Y 3D mapping data samples respectively. For example, for a 3D mapping data sample that is an area containing a large bridge, the visual rendering training result obtained after the visual rendering operation is the visualized presentation of the bridge area after rendering, including the color of the bridge, the texture of the material, and the integration effect with the surrounding environment. Based on the visual rendering training results corresponding to the Y 3D mapping data samples respectively, the second training error is determined. The determination of this second training error may be to compare the difference between the visual rendering training result and the expected result. For example, if the color of the bridge in the visual rendering training result deviates greatly from the actual color, the value of the second training error will increase.

[0023] Finally, based on the first training error and the second training error, joint debugging is performed on the first model visual mapping network and the model visual rendering network. This joint debugging can make the two networks coordinate and optimize each other. For example, if the first training error is mainly caused by inaccurate feature extraction of a certain type of ground object during the visual feature vector mining process, then the relevant parameters in the first model visual mapping network can be adjusted; if the second training error is due to improper handling of the lighting effect during the visual rendering operation, the lighting calculation module in the model visual rendering network can be adjusted. Through such continuous joint debugging process, the performance of the entire visualization system is gradually improved, so that the final target visual rendering output result can more accurately and vividly realize the model visualization display of the target digital city area.

[0024] During the implementation of the entire invention, through the orderly execution of each step, the data visualization display system continuously optimizes and improves the model visualization display effect of the target digital city area, from data acquisition, mining of visual element vectors, performing visual rendering operations to joint debugging of the network, providing users with a high-quality digital city three-dimensional model visualization experience. This process involves precise processing of various data, construction and optimization of complex networks, and close cooperation between each link, thus realizing the conversion from the original three-dimensional surveying and mapping data to high-quality visual presentation.

[0025] It can be understood that the data visualization display system is the core execution entity of the embodiment of the present invention, and the first model visual mapping network, the model visual rendering network, and the joint debugging process of the two constitute the key part of the invention.

[0026] The first model visual mapping network undertakes the important task of mining visual element vectors from the three-dimensional surveying and mapping data to be processed throughout the invention. This network can deeply analyze the three-dimensional surveying and mapping data to be processed, which includes terrain data, feature data, and texture data. Taking a certain area in the city as an example, this area contains complex terrain undulations, such as mountains with an altitude ranging from 50 meters to 300 meters, and there are also many different types of features, such as buildings with heights ranging from 10 floors to 50 floors, and roads with different widths and orientations. The multi-modal spatial representation vectors of the target digital city area described by the three-dimensional surveying and mapping data to be processed contain comprehensive information in these aspects of terrain, features, texture, etc. For example, for a building, the multi-modal spatial representation vector may contain numerical features such as its three-dimensional coordinates (x, y, z), building height, and building area. For a road, it may contain numerical features such as the starting and ending coordinates and road width. For the terrain, it may contain elevation values at different locations. These multi-modal spatial representation vectors are the original and comprehensive descriptions of the urban area spatial information.

[0027] The first model visual mapping network processes this complex multi-modal spatial representation vector through its own structure and algorithm, and mines out the visual element vectors. The visual element vector is a vector representation that is of great significance for visualization. For example, for a specific building, the visual element vector may contain the contour feature vector of the building when observed from a certain perspective, and its value can represent the shape parameters of the contour, such as the curvature of the corner; it may also contain the surface color distribution feature vector of the building, which numerically represents the proportion of different color areas; and the relative position relationship feature vector between the building and the surrounding environment, which numerically represents the distance ratio to adjacent roads and other buildings.

[0028] The model visual rendering network is responsible for performing visual rendering operations on the visual element vectors mined by the first model visual mapping network. This network will perform a series of rendering processes based on various characteristic values in the visual element vectors. For example, for the building contour feature vector in the visual element vector, the model visual rendering network will determine the shape presentation of the building in the visualization result according to the shape parameters of the contour; for the color distribution feature vector, it will accurately assign colors to the building surface according to the proportion values of different color regions; for the relative position relationship feature vector, it will reasonably arrange the position layout of the building in the visualization scene of the entire urban area. During the rendering process, the model visual rendering network will also consider factors such as lighting effects and material textures to generate a more realistic visualization effect. For example, calculate the position and size of the shadow according to the direction of the sun and the position relationship of the building, and determine the texture of the building surface material according to the function of the building (such as residential or commercial) (for example, a residential building may have a softer material texture, and a commercial building may have a more modern material texture), so as to obtain the target visual rendering output result. The target visual rendering output result is the final visualized image or scene presented to the user, which can accurately and realistically display the 3D model of the target digital urban area, enabling users to intuitively observe the terrain, landform, and distribution of ground objects in the urban area.

[0029] The joint debugging process of these two networks is the key to ensuring the accuracy and effectiveness of the entire visualization system. During the joint debugging process, first obtain a debugging sample set, which contains X visual rendering output samples and Y 3D surveying and mapping data samples (for example, X = 80, Y = 60). For each 3D surveying and mapping data sample (these samples also contain multi-modal space representation vector information as mentioned before, but for the sample digital urban area), the first model visual mapping network performs visual element vector mining to obtain the first visual element vector. At the same time, for each visual rendering output sample, visual element vector mining is also performed to obtain the first visual sample vector.

[0030] For example, for a building in a 3D mapping data sample, the building height value in its multi-modal space representation vector is 100 meters, and the building area is 5,000 square meters. After being mined by the first model visual mapping network, the numerical value of the contour feature vector in the first visual feature vector represents that the curvature of a certain corner of the contour is 0.3, and the color distribution feature vector represents that the proportion of a certain color is 30%. For a similar building in a visual rendering output sample, the numerical value of the contour feature vector in the mined first visual sample vector represents that the corner curvature is 0.35, and the color distribution feature vector represents that the proportion of this color is 28%. By comparing the first visual feature vectors of these Y 3D mapping data samples with the first visual sample vectors of X visual rendering output samples, the first training error is determined. The determination of the first training error may be based on various difference measurement methods, such as calculating the sum of squared differences of the numerical values of the contour feature vectors, the sum of absolute differences of the numerical values of the color distribution feature vectors, etc., and the numerical value of the first training error is comprehensively obtained based on these calculation results.

[0031] Next, the model visual rendering network performs visual rendering operations on the first visual feature vectors of Y 3D mapping data samples to obtain visual rendering training results. For example, for the building contour feature vector and color distribution feature vector in the visual feature vector of one of the 3D mapping data samples, the shape and color of the building rendered by the model visual rendering network may be different from the expected ones. The second training error is determined based on the difference between these visual rendering training results and the expected results. For example, if the rendered building height looks 5 meters shorter than the actual 100 meters, or the deviation of the color from the expected color reaches a certain degree, these will all be reflected in the calculation of the second training error.

[0032] Finally, based on the first training error and the second training error, the first model visual mapping network and the model visual rendering network are jointly debugged. If the first training error is mainly due to inaccurate mining of the contour feature vector (such as the error of the corner curvature), then the parameters related to contour feature mining in the first model visual mapping network are adjusted; if the second training error is due to improper color processing during the rendering process (such as the deviation of the color proportion), then the modules related to color rendering in the model visual rendering network are adjusted. Through such a joint debugging process, the performance of the two networks is continuously optimized, so that the entire visualization system can more accurately obtain a high-quality target visual rendering output result from the multi-modal space representation vector through visual feature vector mining and visual rendering operations, and achieve the precise visualization display of the target digital city area.

[0033] Based on the above content, it can be seen that the first training error and the second training error play a crucial role, and they are the key indicators for measuring and optimizing the performance of the first model visual mapping network and the model visual rendering network.

[0034] The first training error is determined based on the first visual element vector of the 3D mapping data sample and the first visual sample vector of the visual rendering output sample. For each 3D mapping data sample, which contains a multimodal spatial representation vector describing the sample digital city area, the first model visual mapping network mines the visual element vector therefrom to obtain the first visual element vector. For example, for a 3D mapping data sample containing buildings, the first visual element vector may include a shape feature vector of the buildings (such as a series of numerical values representing the complexity, symmetry, etc. of the building outline), a texture feature vector (such as numerical values representing the roughness of the texture, the color distribution ratio of the texture, etc.). For the visual rendering output sample, the first visual sample vector is obtained through the mining of the visual element vector by the first model visual mapping network. The calculation of the first training error may involve various ways to measure the difference between these two sets of vectors. For example, for the shape feature vector, the sum of the squares of the differences between the corresponding numerical values can be calculated; for the texture feature vector, the sum of the absolute values of the differences in the color distribution ratio numerical values can be calculated, etc. By combining the differences in these different parts, the value of the first training error is obtained.

[0035] In traditional data visualization technologies, there is often a lack of an effective evaluation of the accuracy of the process of mining visual element vectors from the original data. The embodiments of the present invention introduce the first training error, which can accurately quantify the deviation between the first model visual mapping network and the ideal result when mining visual element vectors. This enables the system to optimize the first model visual mapping network in a targeted manner. For example, if the first training error shows a large deviation in terms of the shape feature vector, it may mean that there are deficiencies in the mining algorithm of the first model visual mapping network for processing shape information such as building outlines, and relevant parameters in the network need to be adjusted or the algorithm needs to be improved. This error quantification of the mining process is an innovative idea, which helps to improve the accuracy of visual element vector mining and thus lays a foundation for subsequent high-quality visual rendering.

[0036] The second training error is determined based on the visual rendering training result obtained by the model visual rendering network performing a visual rendering operation on the first visual feature vector of the 3D mapping data sample. When the model visual rendering network renders the first visual feature vector, a visual rendering training result will be generated, which is a visual presentation of the sample digital city area. For example, for the first visual feature vector of a 3D mapping data sample containing roads and surrounding buildings, the visual rendering training result will show the presentation effects in the visual scene such as the width, color, and material texture of the roads, as well as the height, color, and relative position relationship with the roads of the buildings. The calculation of the second training error is achieved by comparing the differences between the visual rendering training result and the expected result. The expected result can be based on known accurate information or a pre-set ideal visual effect. For example, if in the visual rendering training result, the height of a building appears to be shorter than the actual height by a certain value, or the color of the road has a significant deviation from the expected color, these will be quantified into the second training error. Various methods can be used to calculate this difference, such as calculating the sum of squared differences between each visual element (such as building height, color, etc.) and the expected value.

[0037] In previous data visualization technologies, the evaluation of rendering results was often relatively rough, lacking precise quantitative error analysis. The second training error in the embodiments of the present invention can carefully measure the gap between the rendering effect of the model visual rendering network and the expectation. This enables the system to adjust relevant parameters in the model visual rendering network, such as the lighting calculation module, color assignment module, etc., according to the feedback of the second training error. For example, if the second training error shows that there are significant problems in the lighting effect of the building, the part of the model visual rendering network related to lighting calculation can be adjusted to improve the accuracy and realism of the rendering. This method of performing precise quantitative error analysis on the rendering result and using it to optimize the rendering network is the innovation of the embodiments of the present invention, which helps to improve the rendering quality of the entire visualization system and make the final target visual rendering output result more in line with actual requirements.

[0038] Furthermore, the embodiments of the present invention realize the joint debugging of the first model visual mapping network and the model visual rendering network by simultaneously introducing the first training error and the second training error. This joint debugging is an innovative method. Previous technologies often optimized the mapping or rendering link separately and it was difficult to achieve the collaborative optimization of the entire visualization process. For example, in traditional technologies, only the optimization of the rendering result may be concerned, while ignoring the impact of the process from the original data to the visual feature vector mining on the final visual effect, or vice versa.

[0039] By adjusting the first model visual mapping network based on the first training error, adjusting the model visual rendering network based on the second training error, and enabling the two networks to influence each other and optimize collaboratively during the joint debugging process, the accuracy and quality of the entire visualization system from the original 3D surveying and mapping data to the final target visual rendering output result can be significantly improved. This idea of joint debugging is a new exploration in the field of data visualization and provides a more effective solution for improving the model visualization display effect in the digital city area.

[0040] The introduction of the two training errors helps to comprehensively improve the performance of the visualization system. From the perspective of the data processing flow, the first training error ensures the accuracy of visual feature vector mining and provides high-quality input for subsequent rendering; the second training error guarantees the accuracy and fidelity of the rendering result. The combination of the two enables the entire visualization system to more accurately present the 3D model of the target digital city area. For example, in application scenarios such as digital city planning and management, it can provide users with more accurate and intuitive urban spatial information, which is an effect difficult to achieve by traditional data visualization techniques, reflecting the novelty of the embodiment of the present invention in improving visualization quality.

[0041] In an optional embodiment, determining the first training error according to the first visual feature vectors of the Y 3D surveying and mapping data samples and the first visual sample vectors of the X visual rendering output samples includes: determining a first visual vector chain and a second visual vector chain, where the first visual vector chain includes U second visual sample vectors, the second visual vector chain includes V second visual feature vectors, and both U and V are positive integers; performing joint training according to the first visual feature vectors of the Y 3D surveying and mapping data samples, the first visual sample vectors of the X visual rendering output samples, the U second visual sample vectors included in the first visual vector chain, and the V second visual feature vectors included in the second visual vector chain to determine the first training error.

[0042] In this embodiment, the process of the data visualization display system determining the first training error first determines the first visual vector chain and the second visual vector chain. The first visual vector chain contains U second visual sample vectors, and the second visual vector chain contains V second visual feature vectors, and both U and V are positive integers.

[0043] For the second visual example vectors in the first visual vector chain, these vectors are further derived or selected from the first visual example vectors of the visual rendering output examples. For example, the first visual example vectors of the visual rendering output examples are vector representations of the visualization results of a certain digital city area, which may include visual feature vectors in many aspects such as buildings and roads. The second visual example vectors selected from this first visual example vector may be more focused on specific visual elements, such as only focusing on the appearance feature vectors of buildings. For example, the appearance feature vectors of a certain building in the first visual example vector include a color distribution vector (represented by numerical values for the proportion of different colors, such as 30% red, 20% blue, etc.) and a shape feature vector (represented by numerical values for the curvature of the contour, such as the curvature at the corner is 0.3), etc., and the selected second visual example vector may be further refined to the feature vectors of specific decorative parts of the building facade, such as the roughness value of the decorative texture is 0.5, etc.

[0044] Similarly, the V second visual element vectors in the second visual vector chain are derived or selected from the first visual element vectors of the 3D mapping data examples. Taking the buildings in the 3D mapping data examples as an example, the first visual element vectors may include the basic structural feature vectors of the buildings (such as height, number of floors, etc., for example, the height is 100 meters and the number of floors is 30) and the geographical location feature vectors (represented by coordinate values for their positions in the city, such as x = 100, y = 200, etc.). And the second visual element vectors may be more focused on some visual-related feature vectors of the internal structure of the buildings, such as the spatial proportion vectors of the internal room layout (such as the length-width ratio of a certain room is 2:1, etc.).

[0045] Then, based on the first visual element vectors of Y 3D mapping data examples, the first visual example vectors of X visual rendering output examples, the U second visual example vectors included in the first visual vector chain, and the V second visual element vectors included in the second visual vector chain, joint training is carried out to determine the first training error. In this joint training process, it involves in-depth mining and quantification of the relationships between multiple groups of vectors.

[0046] For example, for a sample of 3D mapping data of a digital city area containing multiple buildings and roads, the value of the building height vector in the first visible element vector is 100 meters, while the corresponding value of the building height in the first visible sample vector of the corresponding visible rendering output sample appears to be 95 meters. At the same time, a certain second visible sample vector in the first visible vector chain can represent the display effect value of the building top decoration part in visualization (for example, this value represents that the display integrity of the decoration part is 80%), while a certain second visible element vector in the second visible vector chain can represent the original value of the building top structure in the 3D mapping data (for example, this value represents that the complexity of the top structure is 0.6).

[0047] In the joint training, the mutual relationships between these different-level vectors need to be comprehensively considered to determine the first training error. Operations such as weighted summation can be performed on the numerical differences between different vectors. For example, for the difference in building height, since it has a greater impact on the overall visualization effect, a relatively high weight may be given; while for the differences in the display effect of the top decoration part and the complexity of the top structure, different weights are given according to their relative importance to the overall visualization. Through such comprehensive weighted summation or other quantization operations considering the relationships of multiple groups of vectors, the first training error is finally determined.

[0048] This way of determining the first training error has significant innovation and advantages compared with traditional methods. Traditional methods for determining training errors may simply compare some vectors or use single-level vector comparison, which is difficult to comprehensively and accurately reflect the complex relationship from 3D mapping data to visible rendering output. However, the method in this embodiment can more deeply explore the associations and differences between different-level vectors by constructing multiple groups of vector chains and performing joint training, thus more precisely determining the first training error.

[0049] For the second training error, as mentioned before, it is determined based on the visible rendering training result obtained by the model visible rendering network performing visible rendering operations on the first visible element vector of the 3D mapping data sample. Here, it can be further seen that the determination method involved in the embodiment of the first training error of the present invention is of great significance to the entire visualization system. Because the first training error reflects the accuracy in the process of mining visible element vectors from 3D mapping data to visible rendering output, and this accuracy directly affects subsequent visible rendering operations. If the first training error is large, it indicates that there are significant problems in the mining of visible element vectors, then even if the performance of the model visible rendering network itself is good, the final obtained target visible rendering output result may deviate greatly from the expectation.

[0050] For example, in the visualization of a digital city area, due to the large first training error, the mining of the visible element vectors of buildings is inaccurate. During the visual rendering operation, the model visual rendering network renders according to the inaccurate visible element vectors, which can cause serious deviations in the presentation of the height, color, appearance, etc. of the buildings in the visualization result. By accurately determining the first training error and making corresponding adjustments, the accuracy of the visible element vector mining can be improved, thereby providing more accurate input for the model visual rendering network and ultimately improving the quality of the target visual rendering output result.

[0051] This method of determining the first training error by constructing multiple groups of vector chains and performing joint training plays a crucial role in connecting the upper and lower parts in the entire data visualization display system. It not only improves the accuracy of the visible element vector mining, but also, by combining with the second training error, realizes more effective joint debugging of the first model visual mapping network and the model visual rendering network. During the joint debugging process, the first model visual mapping network is adjusted according to the first training error, and the model visual rendering network is adjusted according to the second training error, enabling the two networks to cooperate and optimize, continuously improving the accuracy and quality of the entire visualization system from the original three-dimensional surveying and mapping data to the final target visual rendering output result.

[0052] Designed in this way, the method of determining the first training error by constructing multiple groups of vector chains and performing joint training realizes a more accurate evaluation of the visible element vector mining error. Compared with traditional methods, it can consider the relationships between vectors at different levels more comprehensively and improve the accuracy of error determination. This helps to improve the quality of the visible element vector mining, provides more accurate input for the model visual rendering network, and thus, in the entire visualization system, in cooperation with the second training error, realizes effective joint debugging of the first model visual mapping network and the model visual rendering network, enhancing the accuracy and quality from the three-dimensional surveying and mapping data to the target visual rendering output result, and providing more reliable and high-quality technical support for the model visualization display of the digital city.

[0053] Under a preferred technical idea, the determination of the first visible vector chain and the second visible vector chain includes: performing visible element vector mining on the Y three-dimensional surveying and mapping data samples respectively through the second model visual mapping network to obtain the second visible element vectors of each three-dimensional surveying and mapping data sample, and performing visible element vector mining on the X visual rendering output samples respectively to obtain the second visible sample vectors of each visual rendering output sample; migrating the second visible element vectors of the Y three-dimensional surveying and mapping data samples into the second visible vector chain; migrating the second visible element vectors of the X visual rendering output samples into the first visible vector chain.

[0054] Under this preferred technical idea, determining the first visible vector chain and the second visible vector chain is a key link in the entire technical solution, and its process has a unique logic and function.

[0055] First, the second model visual mapping network is used to respectively perform visual element vector mining on Y three-dimensional surveying and mapping data samples to obtain the second visual element vectors of each three-dimensional surveying and mapping data sample. At the same time, visual element vector mining is respectively performed on X visual rendering output samples to obtain the second visual sample vectors of each visual rendering output sample. This process is similar to the operation of the previous first model visual mapping network, but has different purposes and subsequent effects.

[0056] Taking a specific digital city area as an example, for the building part in the three-dimensional surveying and mapping data sample, for example, some characteristic values of the building in the three-dimensional surveying and mapping data are: the height is 80 meters, the floor area is 1000 square meters, and the texture complexity value of the building facade is 0.4 (the 0.4 in the embodiments of the present invention is a value set to represent the texture complexity). The second visual element vectors obtained through visual element vector mining by the second model visual mapping network can convert these values into vector representations more directly related to visualization. For example, the height vector can be converted into a height ratio vector (for example, 0.6, representing the height ratio relative to other elements in the entire visualization scene) at a specific visualization perspective according to a certain mapping relationship, the floor area may be converted into a space occupancy ratio vector (for example, 0.1, representing the space occupancy ratio in the visualization scene), and the facade texture complexity value may be converted into a texture visual effect vector (for example, 0.3, representing the texture visual significance level in the visualization).

[0057] For the visual rendering output sample, for example, a scene that has been preliminarily visualized and contains the building. For example, in this visual rendering output sample, some characteristic values of the building in the visualization are: the display height ratio of the building seen from a specific perspective looks like 0.55 (different from the ideal value of 0.6 mined from the previous three-dimensional surveying and mapping data sample), the space occupancy ratio in the visualization scene looks like 0.09 (different from the ideal value of 0.1), and the texture visual significance level looks like 0.28 (different from the ideal value of 0.3). The second visual sample vectors obtained by mining through the second model visual mapping network will perform vector representations on these actual performances in the visualization.

[0058] Then, transfer the second visual element vectors of Y three-dimensional mapping data samples to the second visual vector chain. This transfer operation makes the second visual vector chain become a set of vectors related to visualization mined from the three-dimensional mapping data samples. These vectors contain multi-faceted visualization-related information about terrain, features, etc. after the original three-dimensional mapping data is processed by the second model visual mapping network. For example, in addition to the vectors related to the above-mentioned buildings, for the roads in this digital city area, there may be values such as a road width of 20 meters and a length of 500 meters in the three-dimensional mapping data. The second visual element vectors after mining and conversion may include the width ratio vector of the road in the visualization scene (e.g., 0.05) and the length ratio vector (e.g., 0.2), etc. These vectors are all transferred to the second visual vector chain.

[0059] Similarly, transfer the second visual sample vectors of X visual rendering output samples to the first visual vector chain. In this way, the first visual vector chain becomes a set of vectors mined from the visual rendering output samples. These vectors reflect the information related to the actual visual performance in the visualization output results. For example, for different combinations of buildings and roads in multiple visual rendering output samples, the vectors in the first visual vector chain will record their actual visual features in each visualization output, such as the relative height ratio vector between buildings and the relative position vector between roads and buildings.

[0060] This way of determining the first visual vector chain and the second visual vector chain is of great significance in the entire technical solution. It provides a basis for jointly training to determine the first training error based on the first visual element vectors of Y three-dimensional mapping data samples, the first visual sample vectors of X visual rendering output samples, the U second visual sample vectors included in the first visual vector chain, and the V second visual element vectors included in the second visual vector chain. The vector chains constructed in this way can more comprehensively cover the information in all aspects from the original three-dimensional mapping data to the visual rendering output.

[0061] In traditional data visualization-related technologies, the construction of vector chains is often relatively simple or lacks the method of separately mining and transferring vectors from different perspectives (three-dimensional mapping data samples and visual rendering output samples) to construct vector chains. Traditional methods may simply extract some vectors from the original data or visualization output for simple comparison, and it is difficult to comprehensively consider the visualization-related feature vectors of data at different stages like this technical solution.

[0062] This way of constructing vector chains plays a crucial role in the error determination and network optimization process of the entire visualization system. For example, when determining the first training error, the vectors in the first visual vector chain and the second visual vector chain can be co-analyzed with the first visual feature vectors of Y three-dimensional mapping data samples and the first visual sample vectors of X visual rendering output samples. Since these vector chains contain visualization-related vectors mined from different levels and perspectives, during the joint training process, the differences and correlations between different vectors can be discovered more precisely. For example, when comparing the visualization effects of buildings, not only can the differences in basic information such as building height and floor area in the first visual feature vectors and the first visual sample vectors be compared, but also through the vectors in the first visual vector chain and the second visual vector chain, the more detailed visual feature differences such as the relative height ratio and space occupancy ratio of the building in the entire visualization scene can be further compared, so as to more accurately determine the first training error.

[0063] This ability to accurately determine the first training error further affects the performance of the entire visualization system. Because the accuracy of the first training error is directly related to the optimization of the first model's visual mapping network. If the first training error can more precisely reflect the problems in the process of mining visual feature vectors from three-dimensional mapping data to visual rendering output, then the first model's visual mapping network can be adjusted more targeted. For example, if it is found through the joint analysis of vector chains that the first training error is mainly caused by problems in the process of mining and transforming the space occupancy vector of the building, the part related to space occupancy calculation in the first model's visual mapping network can be adjusted.

[0064] At the same time, this way of determining the first visual vector chain and the second visual vector chain is also related to the optimization of the model's visual rendering network. Although the second training error is mainly determined based on the visual rendering training results obtained by the model's visual rendering network performing visual rendering operations on the first visual feature vectors of three-dimensional mapping data samples, the vectors in the first visual vector chain and the second visual vector chain can indirectly affect the visual rendering operations. Because accurate vector chain construction helps to more precisely determine the first training error, and the optimization of the first training error will make the visual feature vectors provided to the model's visual rendering network more accurate, thereby improving the input quality of the model's visual rendering network, ultimately helping to enhance the rendering effect of the model's visual rendering network, and further improving the accuracy and quality of the entire visualization system from the original three-dimensional mapping data to the final target visual rendering output result.

[0065] Thus, the method of mining data samples through the second model visual mapping network and constructing the first visual vector chain and the second visual vector chain by migrating vectors comprehensively covers various visualization-related information from 3D mapping data to visual rendering output. This provides a richer information basis for accurately determining the first training error and can more accurately reflect the problems in the process of visual element vector mining compared with traditional methods. The embodiment of the present invention helps to more specifically optimize the first model visual mapping network, indirectly improve the input quality of the model visual rendering network, and ultimately enhance the accuracy and quality of the entire visualization system, providing more reliable technical support for the 3D model visualization display of digital cities.

[0066] In an alternative embodiment, the joint training based on the first visual element vectors of the Y 3D mapping data samples, the first visual sample vectors of the X visual rendering output samples, the U second visual sample vectors included in the first visual vector chain, and the V second visual element vectors included in the second visual vector chain to determine the first training error includes: (1) For the p-th 3D mapping data sample among the Y 3D mapping data samples, determine the first feature commonality value between the first visual element vector of the p-th 3D mapping data sample and each of the U second visual sample vectors, where p is a positive integer not greater than Y; (2) For the q-th visual rendering output sample among the X visual rendering output samples, determine the second feature commonality value between the first visual sample vector of the q-th visual rendering output sample and each of the V second visual element vectors, where q is a positive integer not greater than X; (3) Determine the first training error based on the first feature commonality value corresponding to each of the Y 3D mapping data samples and the second feature commonality value corresponding to each of the X visual rendering output samples.

[0067] In this alternative embodiment, the process of determining the first training error is a crucial part of the entire technical solution, which involves complex relationship analysis and quantification operations among multiple vectors. First, for the p-th three-dimensional mapping data sample among the Y three-dimensional mapping data samples, it is necessary to determine the first feature commonality value between its first visual element vector and each of the U second visual sample vectors, where p is a positive integer not greater than Y. Taking a specific three-dimensional mapping data sample of a digital city as an example, for instance, the p-th three-dimensional mapping data sample is about a certain area and contains a building. Some characteristic values of this building in the first visual element vector are as follows: the height value is 100 meters, its relative position coordinate value in the entire area is (50, 80) (the coordinates in the embodiments of the present invention are numerical values set for representing relative positions), and the texture feature value of the building facade is 0.5 (representing a certain quantified feature of the texture).

[0068] For a certain second visual sample vector among the U second visual sample vectors, it may be mined from another visual rendering output sample and contains some characteristic vector values regarding the visualization presentation of a similar building. For example, the relative proportion value of the building height in this second visual sample vector in the visualization is 0.8 (relative to other elements in the entire visualization scene), its relative position relationship value in the visualization scene is (0.4, 0.6) (representing the relative position proportion with respect to other elements), and the visual saliency value of the facade texture in the visualization is 0.4.

[0069] When determining the first feature commonality value, it is necessary to comprehensively consider the numerical features in these vectors. For example, for the building height, its commonality degree can be calculated according to certain rules. If the height value is converted into the relative proportion in its respective reference system, the relative height proportion of the 100-meter building in its own area is, for example, 0.6, which is compared with the relative proportion of 0.8 in another second visual sample vector. Based on the difference between the two and a certain weight (this weight may be determined according to the importance of the height in the overall visualization), the commonality part value in terms of height is calculated. Similar methods are also used for the relative position coordinates and texture features, etc. The commonality part values calculated in each aspect are integrated (such as by weighted summation, etc.), and finally, the first feature commonality value between the first visual element vector and this second visual sample vector is obtained. Such operations are performed on each of the U second visual sample vectors to obtain the first feature commonality values between the p-th three-dimensional mapping data sample and each second visual sample vector.

[0070] Next, for the q-th visual rendering output example among the X visual rendering output examples, it is necessary to determine the second feature commonality value between its first visual example vector and each of the V second visual feature vectors, where q is a positive integer not greater than X. For example, the q-th visual rendering output example is a visualization result of a specific urban area, and some feature values of a building in the first visual example vector are: the height ratio value of the building viewed from a specific perspective is 0.7, the relative distance ratio value to the surrounding roads is 0.3, and the color ratio value of the building's exterior color in the visualization is 0.4 (indicating the proportion of a certain color in the overall vision).

[0071] For one of the V second visual feature vectors, for example, it is mined from a 3D surveying and mapping data example and contains a vector after converting some original relevant values of the building. For example, the relative height ratio value after converting the actual height value of the building is 0.65, the ratio of the relative distance to the surrounding features after conversion is 0.25, and a certain quantization value of the exterior wall color of the building is 0.35.

[0072] When determining the second feature commonality value, various numerical features also need to be comprehensively considered. For example, for the building height ratio, according to the difference between the two values and the importance weight of height in the visualization, calculate the common part value in terms of height. For the relative distance ratio and color ratio, etc., also in a similar way, comprehensively combine the common part values calculated in each aspect (such as by weighted summation, etc.) to obtain the second feature commonality value between the first visual example vector of the q-th visual rendering output example and this second visual feature vector. Perform such operations on each of the V second visual feature vectors to obtain the second feature commonality value between the q-th visual rendering output example and each second visual feature vector.

[0073] Finally, determine the first training error based on the first feature commonality value corresponding to each of the Y 3D surveying and mapping data examples and the second feature commonality value corresponding to each of the X visual rendering output examples. In this process, various methods can be used to comprehensively combine these feature commonality values. For example, consider all the first feature commonality values and second feature commonality values as an overall data set, and calculate a certain statistical feature of this data set to represent the first training error. Or process the first feature commonality value and the second feature commonality value separately. For example, perform weighted averaging on the first feature commonality value to obtain a comprehensive value, and perform a similar operation on the second feature commonality value, and then determine the first training error through a certain functional relationship (such as the absolute value of the difference or other more complex functions) between these two comprehensive values.

[0074] This way of determining the first training error has obvious advantages compared with traditional methods. When traditional data visualization techniques determine the training error, they often lack the delicate analysis of the common features between vectors at different levels. Usually, they simply compare some surface vector values or use relatively rough methods to evaluate the error. However, the method in this embodiment can more accurately reflect the accuracy in the mining of visual element vectors during the process from 3D mapping data to visual rendering output by deeply analyzing the common features between different vectors.

[0075] For example, in the traditional method, it may only simply compare the height values of buildings, while ignoring their relative proportions in their respective reference systems and their comprehensive relationships with other features (such as location, texture, etc.). However, in this embodiment, by calculating the first common feature value and the second common feature value, these factors are comprehensively considered, and the differences and similarities between vectors can be found more precisely, so as to more accurately determine the first training error.

[0076] The ability to determine the first training error involved in the embodiments of the present invention directly affects the optimization of the visual mapping network of the first model. If the first training error can accurately reflect the problems in the process of mining visual element vectors, relevant parameters in the visual mapping network of the first model can be adjusted accordingly. For example, if it is found that the first training error is mainly caused by a low common feature value in calculating the relative position of buildings, the part related to position calculation in the visual mapping network of the first model can be adjusted to improve the accuracy of mining visual element vectors.

[0077] At the same time, this accurate determination of the first training error also indirectly affects the model visual rendering network. Because an accurate visual mapping network of the first model can provide more accurate visual element vectors for the model visual rendering network, thereby improving the input quality of the model visual rendering network, and ultimately contributing to improving the accuracy and quality of the entire visualization system from the original 3D mapping data to the final target visual rendering output result.

[0078] It can be seen that by calculating the common feature values between different vectors to determine the first training error, this way comprehensively considers the multi-faceted feature relationships of each vector in the 3D mapping data samples and the visual rendering output samples. Compared with traditional methods, it can more accurately reflect the accuracy problems in the process of mining visual element vectors, help optimize the visual mapping network of the first model targeted, indirectly improve the input quality of the model visual rendering network, and thus improve the accuracy and quality of the entire visualization system, providing more reliable technical support for the model visualization of digital cities.

[0079] In an exemplary technical solution, determining the first training error according to the first feature commonality value corresponding to each of the Y three-dimensional mapping data samples and the second feature commonality value corresponding to each of the X visual rendering output samples includes: determining the third feature commonality value between the second visual element vector of the p-th three-dimensional mapping data sample and each of the U second visual sample vectors; determining the fourth feature commonality value between the second visual sample vector of the q-th visual rendering output sample and each of the V second visual element vectors; determining the first local training error between the visual element vector and the visual sample vector according to the first feature commonality value and the third feature commonality value corresponding to each of the Y three-dimensional mapping data samples; determining the second local training error between the visual sample vector and the visual element vector according to the second feature commonality value and the fourth feature commonality value corresponding to each of the X visual rendering output samples; and obtaining the first training error according to the first local training error and the second local training error.

[0080] In this exemplary technical solution, the process of determining the first training error is a multi-step and complex operation, which involves in-depth analysis and comprehensive processing of multiple feature commonality values. First, for the p-th three-dimensional mapping data sample among the Y three-dimensional mapping data samples, it is necessary to determine the third feature commonality value between its second visual element vector and each of the U second visual sample vectors. Taking a three-dimensional mapping data sample of a digital city as an example, for example, in the building part of the p-th three-dimensional mapping data sample, its second visual element vector contains some characteristic values. For example, the structural complexity value of the building is 0.6 (0.6 in the embodiments of the present invention is a value for quantifying the structural complexity), and the internal space layout ratio value is 0.3 (indicating the proportional relationship of different functional spaces).

[0081] For a certain second visual sample vector among the U second visual sample vectors, for example, it contains characteristic values regarding visual presentation. For example, the structural visual effect value of the building in visualization is 0.5 (indicating the perceived visual complexity of the structure), and the space occupancy ratio value of the internal space in visualization is 0.25.

[0082] When determining the third feature commonality value, various numerical features in these vectors should be comprehensively considered. For the structural complexity and the structural visual effect, the common part value may be calculated based on the degree of difference between the two and their relative importance in the overall visualization. For example, if the difference between the structural complexity value of 0.6 and the structural visual effect value of 0.5 is large, and the structural complexity is considered relatively important in the overall visualization, then the influence weight of this part on the third feature commonality value is relatively large. A similar method is also adopted for the internal space layout ratio and the spatial occupancy ratio of the internal space in the visualization. The common part values calculated in each aspect are integrated (such as by weighted summation, etc.), and finally the third feature commonality value between the second visual element vector and this second visual sample vector is obtained. Such an operation is performed on each of the U second visual sample vectors to obtain the third feature commonality value between the p-th 3D mapping data sample and each second visual sample vector.

[0083] Next, for the q-th visual rendering output sample among the X visual rendering output samples, determine the fourth feature commonality value between its second visual sample vector and each of the V second visual element vectors. For example, for the building part in the q-th visual rendering output sample, its second visual sample vector contains some visualization-related feature values. For example, the numerical value of the richness of the appearance details of the building in the visualization is 0.7 (indicating the visual manifestation degree of the appearance details), and the numerical value of the degree of integration with the surrounding environment is 0.4 (indicating the visual integration degree with the surrounding environment).

[0084] For a certain second visual element vector among the V second visual element vectors, for example, it contains a vector after the conversion of some relevant original numerical values of the building. For example, the relative numerical value after the conversion of the actual appearance complexity value of the building is 0.65, and the proportional value after the conversion of the actual correlation degree with the surrounding ground features is 0.35.

[0085] When determining the fourth feature commonality value, various numerical features should also be comprehensively considered. For the richness of the appearance details and the appearance complexity, the common part value is calculated based on the difference between the two numerical values and their importance weights in the visualization. For the degree of integration with the surrounding environment and the actual correlation degree with the surrounding ground features, in a similar way, the common part values calculated in each aspect are integrated (such as by weighted summation, etc.), so as to obtain the fourth feature commonality value between the second visual sample vector and this second visual element vector. Such an operation is performed on each of the V second visual element vectors to obtain the fourth feature commonality value between the q-th visual rendering output sample and each second visual element vector.

[0086] Then, based on the first feature commonality value and the third feature commonality value corresponding to each of the Y three-dimensional mapping data samples, determine the first local training error between the visual element vector and the visual sample vector. For example, for a certain three-dimensional mapping data sample, its first feature commonality value represents the degree of commonality with the visual sample vector in some feature aspects (such as the height and location of buildings), and the third feature commonality value represents the degree of commonality with the visual sample vector in some other feature aspects (such as the structural complexity of buildings and the internal space layout).

[0087] When determining the first local training error, the first feature commonality value and the third feature commonality value can be regarded as an overall data set, and the first local training error can be determined by analyzing the statistical features of this data set. For example, the numerical value of the degree of dispersion of this data set can be calculated. If the degree of dispersion is large, it indicates that the commonality between the visual element vector and the visual sample vector in multiple feature aspects is poor, and then the first local training error is large. Or a weighted method can be adopted. According to the importance of the features involved in the first feature commonality value and the third feature commonality value in the overall visualization, different weights are given respectively, and then the weighted comprehensive value is calculated to represent the first local training error.

[0088] Similarly, based on the second feature commonality value and the fourth feature commonality value corresponding to each of the X visual rendering output samples, determine the second local training error between the visual sample vector and the visual element vector. For example, for a certain visual rendering output sample, its second feature commonality value represents the degree of commonality with the visual element vector in some visualization feature aspects (such as the height ratio of buildings in the visualization and the distance ratio from the surrounding roads), and the fourth feature commonality value represents the degree of commonality with the visual element vector in some other visualization feature aspects (such as the richness of the appearance details of buildings and the degree of integration with the surrounding environment).

[0089] When determining the second local training error, a method similar to that for determining the first local training error is also adopted. A certain statistical feature of the data set containing the second feature commonality value and the fourth feature commonality value can be calculated, or the comprehensive value can be calculated by weighting according to the importance of the features to represent the second local training error.

[0090] Finally, based on the first local training error and the second local training error, obtain the first training error. This process may involve comprehensive processing in various ways. For example, the first local training error and the second local training error can be directly added to obtain the first training error. Or according to their relative importance in the overall error evaluation, different weights are given, and then the weighted sum is obtained to get the first training error.

[0091] The above method for determining the first training error has significant advantages. When traditional data visualization techniques determine the training error, they often do not have such a comprehensive and detailed analysis method. Usually, they simply compare some vector values or use a single method to evaluate the error. In this embodiment, by separately determining the feature commonality values at different levels, and then determining the local training error, and finally obtaining the first training error, it can more accurately reflect the relationship between the visual element vector and the visual example vector in the process from 3D mapping data to visual rendering output.

[0092] The error determination method involved in the embodiments of the present invention directly affects the optimization of the visual mapping network of the first model. If the first training error can accurately reflect the differences and common problems between the visual element vector and the visual example vector, relevant parameters in the visual mapping network of the first model can be adjusted accordingly. For example, if the first local training error is mainly caused by problems in the features related to the building structure complexity, the part related to the calculation of the structure complexity in the visual mapping network of the first model can be adjusted, so as to improve the accuracy of mining the visual element vector.

[0093] At the same time, the accurate determination of the first training error also indirectly affects the model visual rendering network. Because an accurate visual mapping network of the first model can provide more accurate visual element vectors for the model visual rendering network, thereby improving the input quality of the model visual rendering network, and ultimately contributing to improving the accuracy and quality of the entire visualization system from the original 3D mapping data to the final target visual rendering output result.

[0094] Therefore, by determining multiple feature commonality values to calculate the first training error, this method comprehensively and deeply considers the relationship between the visual element vector and the visual example vector in multiple aspects. Compared with the traditional method, it can more accurately reflect the relationship problem between the two, which helps to optimize the visual mapping network of the first model targeted, indirectly improve the input quality of the model visual rendering network, and thus improve the accuracy and quality of the entire visualization system, providing more reliable technical support for the model visualization of digital cities.

[0095] Based on the above exemplary technical solution, determining the first local training error between the visible element vector and the visible example vector according to the first feature commonality value and the third feature commonality value corresponding to each three-dimensional mapping data example among the Y three-dimensional mapping data examples includes: obtaining the prior feature commonality value between the visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors; determining the first local training error according to the prior feature commonality value between the visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors, the first feature commonality value between the first visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors, and the third feature commonality value between the second visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors.

[0096] Further, determining the first local training error according to the prior feature commonality value between the visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors, the first feature commonality value between the first visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors, and the third feature commonality value between the second visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors includes: performing feature enhancement on the prior feature commonality value between the visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors and the third feature commonality value between the second visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors to obtain the first common visible feature information; performing bit-by-bit weighting on the first feature commonality value between the first visible element vector of the p-th three-dimensional mapping data example and the U second visible example vectors to obtain the second common visible feature information; determining the first local training error according to the first common visible feature information and the second common visible feature information corresponding to each three-dimensional mapping data example among the Y three-dimensional mapping data examples.

[0097] In the next step, determining the first local training error according to the first common visible feature information and the second common visible feature information corresponding to each three-dimensional mapping data example among the Y three-dimensional mapping data examples includes: performing iterative summation on the first common visible feature information and the second common visible feature information corresponding to each three-dimensional mapping data example among the Y three-dimensional mapping data examples to obtain the first local training error.

[0098] In an embodiment based on the exemplary technical solution, the process of determining the first local training error first involves obtaining the prior feature commonality value between the visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors. Taking the building in a specific 3D mapping data sample as an example, the visual element vector contains feature information such as the height value of the building (e.g., 100 meters) and the floor area value (e.g., 2000 square meters). For one of the U second visual sample vectors, it may contain similar feature values mined from other relevant visualization samples, such as the relative height value of the building in a certain visualization (e.g., 0.8, which is a relative ratio value) and the relative floor area value (e.g., 0.2, also a relative ratio value). The obtaining of the prior feature commonality value is to comprehensively consider the relationship between each numerical feature in these vectors. This process will calculate the commonality degree between them based on certain rules and algorithms. For example, a comprehensive prior feature commonality value is determined according to the numerical relationships between height and relative height, floor area and relative floor area, as well as their importance in the overall visualization.

[0099] Next, the first local training error is determined based on the prior feature commonality value between the visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors, the first feature commonality value between the first visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors, and the third feature commonality value between the second visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors.

[0100] In this process, the prior feature commonality value between the visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors is feature-enhanced with the third feature commonality value between the second visual element vector of the p-th 3D mapping data sample and the U second visual sample vectors to obtain the first common visual feature information. Feature enhancement is an operation to deeply mine and integrate vector relationships. For example, the prior feature commonality value may reflect the commonality degree of the basic features of the building (such as height, area, etc.) at the initial level, while the third feature commonality value may focus on the commonality degree of the internal structure or other deeper-level features of the building. Through the feature enhancement operation, the common features in these two aspects are fused and optimized to obtain the first common visual feature information that more comprehensively reflects the relationship between the visual element vector and the visual sample vector.

[0101] Meanwhile, the first feature commonality value between the first visual feature vector of the p-th 3D mapping data sample and the U second visual sample vectors is weighted bit by bit to obtain the second common visual feature information. Bit-by-bit weighting is an operation of weighting according to the importance of each feature in the vector. For example, the building height in the first visual feature vector has a relatively high importance in the overall visualization. Then, when calculating the first feature commonality value, a higher weight will be assigned to the numerical part related to the height. The second common visual feature information obtained through this bit-by-bit weighting method can more accurately reflect the common relationship under different feature importances.

[0102] Then, based on the first common visual feature information and the second common visual feature information corresponding to each of the Y 3D mapping data samples, the first local training error is determined. Specifically, the first common visual feature information and the second common visual feature information corresponding to each of the Y 3D mapping data samples are iteratively summed to obtain the first local training error. This iterative summation process gradually accumulates the relevant information of each 3D mapping data sample, thereby obtaining an error value that comprehensively reflects the relationship between the visual feature vectors and the visual sample vectors of all 3D mapping data samples. This value is the first local training error.

[0103] The above method for determining the first local training error has obvious advantages compared with traditional methods. When traditional data visualization techniques determine similar errors, they often lack a method of comprehensively considering multiple feature commonality values and deeply integrating them. Usually, they simply compare some vector features or adopt a relatively single calculation method. In this embodiment, by first obtaining the prior feature commonality value, then performing operations such as feature enhancement and bit-by-bit weighting, and finally obtaining the first local training error through iterative summation, it can more accurately reflect the complex relationship between the visual feature vector and the visual sample vector.

[0104] The method for determining the first local training error involved in the embodiment of the present invention directly affects the optimization of the first model visual mapping network. If the first local training error can accurately reflect the differences and commonalities between the visual feature vector and the visual sample vector at multiple feature levels, the relevant parameters in the first model visual mapping network can be adjusted accordingly. For example, if it is found that there is a large deviation in the result of the height-related features of a certain building after feature enhancement and bit-by-bit weighting when calculating the first local training error, the part related to height calculation or processing in the first model visual mapping network can be adjusted to improve the accuracy of visual feature vector mining.

[0105] Meanwhile, the accurate determination of the first local training error also indirectly affects the model visual rendering network. Because an accurate first model visual mapping network can provide more accurate visual element vectors for the model visual rendering network, thereby improving the input quality of the model visual rendering network, and ultimately contributing to enhancing the accuracy and quality of the entire visualization system from the original 3D surveying data to the final target visual rendering output result.

[0106] With such a design, the first local training error is determined by obtaining the prior feature commonality value and performing operations such as feature enhancement, bit-by-bit weighting, and iterative summation. This approach comprehensively and deeply considers various relationships between the visual element vector and the visual example vector, and can more accurately reflect the relationship problem between the two compared with traditional methods. It helps to optimize the first model visual mapping network targeted, indirectly improve the input quality of the model visual rendering network, thereby enhancing the accuracy and quality of the entire visualization system, and providing more reliable technical support for the model visualization of digital cities.

[0107] Based on the above exemplary technical solution, determining the second local training error between the visual example vector and the visual element vector according to the second feature commonality value and the fourth feature commonality value corresponding to each visual rendering output example among the X visual rendering output examples includes: obtaining the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V second visual element vectors; determining the second local training error according to the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V visual example vectors, the second feature commonality value between the first visual element vector of the q-th visual rendering output example and the V second visual element vectors, and the fourth feature commonality value between the second visual element vector of the q-th visual rendering output example and the V second visual element vectors.

[0108] Further, determining the second local training error based on the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V visual example vectors, the second feature commonality value between the first visual element vector of the q-th visual rendering output example and the V second visual element vectors, and the fourth feature commonality value between the second visual element vector of the q-th visual rendering output example and the V second visual element vectors includes: enhancing the feature of the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V visual example vectors and the third feature commonality value between the second visual element vector of the q-th visual rendering output example and the V visual example vectors to obtain third common visual feature information; determining fourth common visual feature information by using the first feature commonality value between the first visual element vector of the q-th visual rendering output example and the V visual example vectors; and determining the second local training error based on the third common visual feature information and the fourth common visual feature information corresponding to each visual rendering output example among the X visual rendering output examples.

[0109] In the next step, determining the second local training error based on the third common visual feature information and the fourth common visual feature information corresponding to each visual rendering output example among the X visual rendering output examples includes: iteratively summing the third common visual feature information and the fourth common visual feature information corresponding to each visual rendering output example among the X visual rendering output examples to obtain the second local training error.

[0110] In this embodiment based on the above exemplary technical solution, the process of determining the second local training error first involves obtaining the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V second visual feature vectors. For example, in the building part of a visual rendering output example, the visual example vector contains specific visual feature values. For example, the color feature value of the building in the visual rendering output is 0.6 (0.6 in the embodiments of the present invention is a numerical value for quantifying color performance, such as it can represent the proportion of a certain color in the overall vision or the vividness of the color, etc.), and its position relationship value in the visual scene is 0.3 (a quantitative value representing the relative position relationship with other elements). For a certain one of the V second visual feature vectors, it may contain relevant values mined from the 3D survey data example, such as a certain quantitative value of the original color of the building being 0.55, and the converted ratio of the original position relationship value being 0.25. The obtaining of the prior feature commonality value is to comprehensively consider the relationship between the respective numerical features in these vectors and calculate the commonality degree between them according to specific algorithms and rules. This prior feature commonality value reflects the commonality relationship between the visual example vector and the second visual feature vector at the initial level.

[0111] Next, the second local training error is determined based on the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V visual example vectors, the second feature commonality value between the first visual feature vector of the q-th visual rendering output example and the V second visual feature vectors, and the fourth feature commonality value between the second visual feature vector of the q-th visual rendering output example and the V second visual feature vectors.

[0112] In this process, the prior feature commonality value between the visual example vector of the q-th visual rendering output example and the V visual example vectors is feature-enhanced with the third feature commonality value between the second visual feature vector of the q-th visual rendering output example and the V visual example vectors to obtain the third common visual feature information. For example, the prior feature commonality value reflects the initial commonality of the visual example vector and the second visual feature vector in basic visual features (such as color, position relationship, etc.), while the third feature commonality value may focus on other deeper or different aspects of feature commonality (such as the relationship between the detailed performance of the building in the visualization and the corresponding features in the original data, etc.). Through the feature enhancement operation, the common features in these two aspects are integrated and optimized, so as to obtain the third common visual feature information that more comprehensively reflects the relationship between the visual example vector and the visual feature vector.

[0113] Meanwhile, the fourth common visual feature information is determined by using the first common feature value between the first visual element vector of the q-th visual rendering output example and the V visual example vectors. This process constructs the fourth common visual feature information based on the information contained in the first common feature value. The first common feature value reflects the common degree of the first visual element vector of the visual rendering output example and the V visual example vectors in certain specific features. By processing and transforming these common information, the fourth common visual feature information that can accurately reflect the specific relationship is obtained.

[0114] Then, according to the third common visual feature information and the fourth common visual feature information corresponding to each visual rendering output example in the X visual rendering output examples, the second local training error is determined. Specifically, the third common visual feature information and the fourth common visual feature information corresponding to each visual rendering output example in the X visual rendering output examples are iteratively summed to obtain the second local training error. This iterative summation process gradually accumulates the relevant information of each visual rendering output example, and finally obtains an error value that comprehensively reflects the relationship between the visual example vectors and the visual element vectors of all visual rendering output examples, that is, the second local training error.

[0115] This way of determining the second local training error has significant advantages compared with traditional methods. In traditional data visualization techniques, when determining similar errors, there is often a lack of such a method for comprehensively processing multiple feature common values. The traditional method may simply compare some vector features or adopt a relatively single calculation method, which is difficult to comprehensively and accurately reflect the complex relationship between the visual example vectors and the visual element vectors. In this embodiment, by obtaining the prior feature common value, performing feature enhancement, determining the fourth common visual feature information, and finally obtaining the second local training error through iterative summation, the internal relationship between the visual example vectors and the visual element vectors can be more accurately mined.

[0116] The way of determining the second local training error involved in the embodiment of the present invention directly affects the optimization of the first model visual mapping network and the model visual rendering network. For the first model visual mapping network, if the second local training error can accurately reflect the differences and common problems between the visual example vectors and the visual element vectors at multiple feature levels, it can provide a basis for adjusting the first model visual mapping network. For example, if it is found that there are deviations in the processing of certain features such as the color or position relationship of a building when calculating the second local training error, the relevant parts of the first model visual mapping network related to these features can be adjusted to improve the accuracy of visual element vector mining.

[0117] For the model visual rendering network, accurate determination of the second local training error helps improve the input quality of the model visual rendering network. Since the accurate determination of the second local training error reflects the accuracy of the relationship between the visual example vector and the visual element vector, this indirectly affects the quality of the visual element vector, and further affects the input of the model visual rendering network, and ultimately helps improve the accuracy and quality of the entire visualization system from the original 3D mapping data to the final target visual rendering output result.

[0118] The second local training error is determined by obtaining the prior feature common value, performing feature enhancement, determining the relevant common visual feature information, and iterative summation. This method comprehensively considers various relationships between the visual example vector and the visual element vector, and can more accurately reflect the relationship problem between the two compared with traditional methods. It helps optimize the first model visual mapping network and the model visual rendering network, improves the accuracy and quality of the entire visualization system, and provides more reliable technical support for the model visualization of digital cities.

[0119] Based on the above preferred technical idea, in an independent embodiment, the joint debugging of the first model visual mapping network and the model visual rendering network according to the first training error and the second training error includes: optimizing the network model variables in the first model visual mapping network and the model visual rendering network through the first training error and the second training error to obtain the network model optimization variables of the first model visual mapping network and the network model optimization variables of the model visual rendering network; optimizing the network model variables in the first model visual mapping network according to the network model optimization variables of the first model visual mapping network, and optimizing the network model variables in the model visual rendering network according to the network model optimization variables of the model visual rendering network.

[0120] In addition, the method further includes: determining the network model optimization variables of the second model visual mapping network according to the network model optimization variables of the first model visual mapping network; optimizing the network model variables in the second model visual mapping network according to the network model optimization variables of the second model visual mapping network.

[0121] Further, the determining the network model optimization variables of the second model visual mapping network according to the network model optimization variables of the first model visual mapping network includes: determining the optimized network model weights of the second model visual mapping network according to the current network model weights of the second model visual mapping network and the optimized network model weights of the first model visual mapping network.

[0122] In this independent embodiment, jointly debugging the first model visual mapping network and the model visual rendering network according to the first training error and the second training error is a key part of the entire technical solution. For the first model visual mapping network and the model visual rendering network, network model variables are important factors determining their functions and performances. These network model variables can be regarded as a set of elements with specific numerical characteristics, existing in the form of multi-dimensional numerical feature vectors. For example, the network model variables in the first model visual mapping network may include vector components related to the mining of different types of visual elements. For example, when processing building data in a digital city, the numerical feature vector of the network model variables in one dimension is related to the mining of the geometric shape of buildings, and its value can be expressed as [0.3, 0.5, 0.2], and another dimension is related to the mining of building textures, with a value of [0.4, 0.1, 0.5]. The network model variables in the model visual rendering network also have such a form of multi-dimensional numerical feature vectors. For example, in the network model variables related to the rendering of light and shadow effects, one dimension may be related to the calculation of the light direction, with a value of [0.2, 0.6, 0.2], and another dimension is related to the rendering of shadow colors, with a value of [0.5, 0.3, 0.2].

[0123] When optimizing the network model variables in these two networks according to the first training error and the second training error, it is based on the analysis and adjustment of these multi-dimensional numerical feature vectors. For the first model visual mapping network, the first training error reflects the deviation situation in the process of mining visual element vectors. If a large deviation from the actual situation (reflected by the first training error) is found when mining the visual element vector of the building geometry, then the numerical feature vector [0.3, 0.5, 0.2] of the network model variables related to the mining of the building geometry will be adjusted. For example, it is adjusted to [0.2, 0.6, 0.2]. This adjustment process is based on the information provided by the first training error, changing the values of each dimension to reduce the mining error. Similarly, the numerical feature vectors of other network model variables related to the first training error will be adjusted accordingly, so as to obtain the network model optimization variables of the first model visual mapping network. For the model visual rendering network, the second training error reflects the deviation in the visual rendering operation. If it is found that the rendering of shadow colors does not match the expectation in the light and shadow effect rendering (reflected by the second training error), then the numerical feature vector [0.5, 0.3, 0.2] of the network model variables related to the rendering of shadow colors will be adjusted, such as adjusted to [0.4, 0.4, 0.2]. After such adjustment, the network model optimization variables of the model visual rendering network are obtained.

[0124] Then, according to the network model optimization variables of the first model visual mapping network, the network model variables in the first model visual mapping network are optimized. This is a process of directly applying the previously obtained optimization variables to the first model visual mapping network. For example, the numerical feature vector [0.2, 0.6, 0.2] related to building geometry mining after adjustment is used to replace the original vector [0.3, 0.5, 0.2], so as to improve the accuracy of the first model visual mapping network in visual feature vector mining. For the model visual rendering network, optimizing its own network model variables according to its network model optimization variables is a similar process, where the adjusted vector related to shadow color rendering is used to replace the original vector, thereby enhancing the accuracy of the visual rendering operation.

[0125] In addition, according to the network model optimization variables of the first model visual mapping network, the network model optimization variables of the second model visual mapping network are determined. The key in the embodiments of the present invention is to determine the optimized network model weights of the second model visual mapping network based on the current network model weights (also in the form of multi-dimensional numerical feature vectors) of the second model visual mapping network and the optimized network model weights of the first model visual mapping network. For example, when the current network model weights of the second model visual mapping network are processing a certain feature of a building (such as internal structure mining), its numerical feature vector is [0.4, 0.3, 0.3]. After the first model visual mapping network is optimized, the numerical feature vector of the optimized network model weights related to similar building internal structure mining becomes [0.3, 0.4, 0.3]. According to the relationship between the two (this relationship may be based on a certain complex algorithm, such as may involve distance calculation between vectors, proportional relationships or other logical relationships, etc.), the numerical feature vector of the optimized network model weights of the second model visual mapping network in building internal structure mining is calculated, such as becoming [0.35, 0.35, 0.3]. Then, according to this optimized network model weight, the network model variables in the second model visual mapping network are optimized. If there is a numerical feature vector [0.5, 0.2, 0.3] of a network model variable related to building internal structure mining in the second model visual mapping network, according to the new optimized network model weight [0.35, 0.35, 0.3], this vector can be adjusted to [0.4, 0.3, 0.3].

[0126] The above-mentioned joint debugging method has significant advantages. Traditional data visualization technologies often lack such a comprehensive, systematic, and optimized method based on network model variables (in the form of multi-dimensional numerical feature vectors) in network debugging. Traditional methods may simply adjust some parameters in the network without considering the multi-dimensionality of network model variables and the complex relationships between different networks. In this embodiment, by using the first training error and the second training error to jointly debug the first model visual mapping network and the model visual rendering network, and further associating with the optimization of the second model visual mapping network, the accuracy of the entire visualization system can be comprehensively improved from multiple dimensions and the relationships between multiple networks.

[0127] From the perspective of the first model visual mapping network and the model visual rendering network, by optimizing the multi-dimensional numerical feature vectors of network model variables through accurate training errors, the accuracy of visual element vector mining and the quality of visual rendering operations can be improved. This makes the entire process from 3D mapping data to visual rendering output more accurate, reducing the accumulation and propagation of errors between different dimensions and different networks. For the second model visual mapping network, the associated optimization with the first model visual mapping network further ensures the coordination and consistency between different links in the entire visualization process, improving the overall performance of the entire system from data mining to visual presentation.

[0128] In this way, by using the first training error and the second training error to jointly debug the first model visual mapping network and the model visual rendering network, including the optimization of network model variables (in the form of multi-dimensional numerical feature vectors) and the associated optimization of the second model visual mapping network. This method considers the multi-dimensionality of network model variables and the relationships between networks more comprehensively and systematically than traditional methods, improving the accuracy of visual element mining, visual rendering operations, as well as the coordination and consistency of the entire visualization system, and enhancing the overall performance from 3D mapping data to visual rendering output.

[0129] In another independently implementable embodiment, determining the second training error based on the visual rendering training results respectively corresponding to the Y 3D mapping data samples includes: obtaining the visual rendering authentication outputs respectively corresponding to the Y 3D mapping data samples; determining the second training error based on the visual rendering authentication outputs and the visual rendering training results respectively corresponding to the Y 3D mapping data samples.

[0130] Specifically, for Y three-dimensional mapping data samples, the visual rendering training result is the result obtained after the model's visual rendering network performs a visual rendering operation on the first visual feature vectors of the Y three-dimensional mapping data samples. These results contain information on how to convert the visual feature vectors in the three-dimensional mapping data samples into visual presentations. For example, for the building part in a certain three-dimensional mapping data sample, the numerical feature of the building's height in the visual rendering training result may be 80 meters (representing the presented height in the visualization), and the numerical feature of its color in the visualization may be 0.6 (representing the numerical value after a certain color quantization, such as the vividness of the color or its relative position in the entire color space, etc.).

[0131] Then, obtain the visual rendering authentication outputs corresponding to the Y three-dimensional mapping data samples respectively. The visual rendering authentication output is a visual result used as a standard or reference. For example, for the same building, the numerical feature of the building's height in the visual rendering authentication output should be 100 meters (different from the 80 meters in the visual rendering training result, reflecting the possible error), and the numerical feature of the color should be 0.7.

[0132] Determine the second training error based on the visual rendering authentication outputs and the visual rendering training results corresponding to the Y three-dimensional mapping data samples respectively. This process quantifies the error by comparing the differences between the two. For the height of the building, there is a 20-meter gap between the 80 meters in the visual rendering training result and the 100 meters in the visual rendering authentication output, and this gap reflects the error part in height rendering. For the numerical feature of the color, there is a 0.1 gap between 0.6 in the visual rendering training result and 0.7 in the visual rendering authentication output. By combining the gaps in different aspects (such as height, color, etc.), the second training error can be determined. This combination may involve weighted processing of different features. For example, since height is more important in the overall visualization, its gap may be given a larger weight when calculating the second training error, while the color gap is given a relatively smaller weight. In this way, a second training error value that can comprehensively reflect the difference between the visual rendering training result and the visual rendering authentication output is finally determined.

[0133] In this way, the second training error is determined through the visual rendering authentication output and the visual rendering training result. This method quantifies the error by comparing the difference between the reference standard and the actual rendering result, and performs comprehensive calculations considering the importance of different features. Compared with traditional methods, it can more accurately reflect the gap between the rendering effect of the model's visual rendering network and the expectation, which helps to optimize the model's visual rendering network targeted, thereby improving the accuracy and quality of the entire visualization system and providing more reliable technical support for the model visualization of digital cities.

[0134] In summary, the embodiments of the present invention can effectively process 3D mapping data including various types of data such as terrain, features, and textures. The visual element vectors are mined through the first model visual mapping network, providing accurate input for subsequent rendering and improving the visualization efficiency. The model visual rendering network performs rendering operations based on the mined vectors to achieve the visual display of the target digital city area. By jointly debugging the two networks based on the first training error and the second training error, the network performance can be accurately optimized, enhancing the accuracy and fidelity of the visualization effect, better presenting the spatial characteristics of the digital city, and meeting various requirements such as urban planning and management.

[0135] Further, Figure 2 FIG. 5 is a schematic structural diagram of a data visualization display system 200 provided by an embodiment of the present invention. As Figure 2 shown, the data visualization display system 200 includes a processor 210. The processor 210 can call and run a computer program from a memory to implement the method in the embodiments of the present invention. Optionally, as Figure 2 shown, the data visualization display system 200 may further include a memory 230. Among them, the processor 210 can call and run a computer program from the memory 230 to implement the method in the embodiments of the present invention. Among them, the memory 230 can be an independent device separate from the processor 210 or integrated in the processor 210. Optionally, as Figure 2 shown, the data visualization display system 200 may further include a transceiver 220. The processor 210 can control the transceiver 220 to interact with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices. Optionally, the data visualization display system 200 can implement the corresponding processes of the storage engine or components (such as processing modules) in the storage engine or the device deployed with the storage engine in each method of the embodiments of the present invention. For the sake of brevity, it will not be elaborated here.

[0136] It should be understood that the processor in the embodiments of the present invention may be an integrated circuit chip with signal processing capabilities.

[0137] It can be understood that the memory in the embodiments of the present invention can be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, suitable types of memories.

[0138] Based on the above, a readable storage medium is provided. Programs or instructions are stored on the readable storage medium, and when the programs or instructions are executed by a processor, the steps of the above method are implemented.

[0139] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the embodiments of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the embodiments of the present invention and the scope protected by the embodiments of the present invention, and all of them fall within the protection of the embodiments of the present invention.

Claims

1. A data visualization display method based on a business model, characterized in that: The method is applied to a data visualization display system, and the method comprises: Acquire the three-dimensional surveying and mapping data to be processed, wherein the three-dimensional surveying and mapping data to be processed is used to describe the multi-mode spatial representation vector of the target digital city area, and the three-dimensional surveying and mapping data to be processed includes terrain data, ground object data and texture data; Performing visual element vector mining on the three-dimensional surveying and mapping data to be processed through a first model visual mapping network to obtain visual element vectors of the three-dimensional surveying and mapping data to be processed; Performing visual rendering operations on the visual element vectors of the three-dimensional surveying and mapping data to be processed through a model visual rendering network to obtain a target visual rendering output result; the target visual rendering output result is used to realize model visualization display of the target digital city area; The model visual rendering network and the first model visual mapping network are obtained by joint debugging based on the first training error and the second training error.

2. The method according to claim 1, characterized in that The debugging steps of the first model visual mapping network and the model visual rendering network include: Obtaining X visual rendering output samples and Y three-dimensional surveying and mapping data samples included in the current debugging sample set, and performing visual element vector mining on each three-dimensional surveying and mapping data sample through a first model visual mapping network to obtain a first visual element vector of each three-dimensional surveying and mapping data sample, and performing visual element vector mining on each visual rendering output sample to obtain a first visual sample vector of each visual rendering output sample, wherein the three-dimensional surveying and mapping data sample is used to describe a multi-modal spatial representation vector of a sample digital city area, and X and Y are both positive integers; Determining a first training error according to the first visible element vectors of the Y three-dimensional mapping data samples and the first visible sample vectors of the X visible rendering output samples; Performing a visual rendering operation on the first visual element vectors of the Y three-dimensional surveying and mapping data samples through a model visual rendering network to obtain visual rendering training results corresponding to the Y three-dimensional surveying and mapping data samples, and determining a second training error according to the visual rendering training results corresponding to the Y three-dimensional surveying and mapping data samples; Jointly debugging the first model visual mapping network and the model visual rendering network according to the first training error and the second training error; The determining a first training error according to the first visual element vectors of the Y three-dimensional mapping data samples and the first visual sample vectors of the X visual rendering output samples comprises: Determine a first visible vector chain and a second visible vector chain, wherein the first visible vector chain includes U second visible sample vectors, and the second visible vector chain includes V second visible element vectors, wherein U and V are both positive integers; The first training error is determined by performing joint training based on the first visual element vectors of the Y three-dimensional mapping data samples, the first visual sample vectors of the X visual rendering output samples, the U second visual sample vectors included in the first visual vector chain, and the V second visual element vectors included in the second visual vector chain.

3. The method according to claim 2, characterized in that The determining of the first visible vector chain and the second visible vector chain comprises: Performing visual element vector mining on the Y three-dimensional surveying and mapping data samples through the second model visual mapping network to obtain a second visual element vector of each three-dimensional surveying and mapping data sample, and performing visual element vector mining on the X visual rendering output samples to obtain a second visual sample vector of each visual rendering output sample; Migrating the second visual element vectors of the Y three-dimensional surveying and mapping data samples to the second visual vector chain; Migrating the second visual element vectors of the X visual rendering output samples into the first visual vector chain.

4. The method according to claim 3, characterized in that The determining the first training error by performing joint training based on the first visual element vectors of the Y three-dimensional mapping data samples, the first visual sample vectors of the X visual rendering output samples, the U second visual sample vectors included in the first visual vector chain, and the V second visual element vectors included in the second visual vector chain includes: For a p-th three-dimensional surveying and mapping data sample among the Y three-dimensional surveying and mapping data samples, determining a first characteristic commonality value between a first visible element vector of the p-th three-dimensional surveying and mapping data sample and each second visible sample vector among the U second visible sample vectors, wherein p is a positive integer not greater than Y; For a qth visual rendering output sample among the X visual rendering output samples, determining a second characteristic commonality value between a first visual sample vector of the qth visual rendering output sample and each second visual element vector among the V second visual element vectors, where q is a positive integer not greater than X; The first training error is determined according to a first feature commonality value corresponding to each 3D surveying and mapping data sample in the Y 3D surveying and mapping data samples and a second feature commonality value corresponding to each 3D surveying and mapping output sample in the X 3D surveying and mapping output samples.

5. The method according to claim 4, characterized in that The determining the first training error according to the first feature commonality value corresponding to each 3D surveying and mapping data sample in the Y 3D surveying and mapping data samples and the second feature commonality value corresponding to each 3D surveying and mapping output sample in the X 3D surveying and mapping output samples comprises: Determining a third characteristic commonality value between the second visible element vector of the p-th three-dimensional surveying and mapping data sample and each second visible sample vector of the U second visible sample vectors; Determining a fourth characteristic commonality value between the second visual sample vector of the qth visual rendering output sample and each second visual element vector of the V second visual element vectors; Determining a first local training error between a visual element vector and a visual sample vector according to a first feature commonality value and a third feature commonality value corresponding to each of the Y three-dimensional surveying and mapping data samples; Determining a second local training error between a visual sample vector and a visual element vector according to a second feature commonality value and a fourth feature commonality value corresponding to each of the X visual rendering output samples; The first training error is obtained according to the first local training error and the second local training error.

6. The method according to claim 5, characterized in that The determining, based on the first feature commonality value and the third feature commonality value corresponding to each of the Y three-dimensional surveying and mapping data samples, a first local training error between the visual element vector and the visual sample vector comprises: Obtaining a priori feature commonality values ​​between the visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors; The first local training error is determined based on a priori feature commonality values ​​between the visual element vector of the p-th three-dimensional mapping data sample and the U second visual sample vectors, a first feature commonality value between the first visual element vector of the p-th three-dimensional mapping data sample and the U second visual sample vectors, and a third feature commonality value between the second visual element vector of the p-th three-dimensional mapping data sample and the U second visual sample vectors.

7. The method according to claim 6, characterized in that The determining the first local training error according to the prior feature commonality value between the visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors, the first feature commonality value between the first visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors, and the third feature commonality value between the second visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors, comprises: Performing feature enhancement on the prior feature commonality value between the visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors and the third feature commonality value between the second visual element vector of the p-th three-dimensional surveying and mapping data sample and the U second visual sample vectors to obtain first common visual feature information; Performing bit-by-bit weighting on the first feature commonality value between the first visible element vector of the p-th three-dimensional surveying and mapping data sample and the U second visible sample vectors to obtain second commonality visual feature information; Determining the first local training error according to the first common visual feature information and the second common visual feature information corresponding to each of the Y three-dimensional surveying and mapping data samples; The determining the first local training error based on the first common visual feature information and the second common visual feature information corresponding to each of the Y three-dimensional surveying and mapping data samples includes: iteratively summing the first common visual feature information and the second common visual feature information corresponding to each of the Y three-dimensional surveying and mapping data samples to obtain the first local training error.

8. The method according to claim 5, characterized in that The determining, according to the second feature commonality value and the fourth feature commonality value corresponding to each of the X visual rendering output samples, a second local training error between the visual sample vector and the visual element vector comprises: Obtaining a priori feature commonality values ​​between the visual sample vector of the qth visual rendering output sample and the V second visual element vectors; The second local training error is determined based on a prior feature commonality value between the visual sample vector of the qth visual rendering output sample and the V visual sample vectors, a second feature commonality value between the first visual element vector of the qth visual rendering output sample and the V second visual element vectors, and a fourth feature commonality value between the second visual element vector of the qth visual rendering output sample and the V second visual element vectors.

9. The method according to claim 8, characterized in that The determining the second local training error according to a priori feature commonality value between the visual sample vector of the qth visual rendering output sample and the V visual sample vectors, a second feature commonality value between the first visual element vector of the qth visual rendering output sample and the V second visual element vectors, and a fourth feature commonality value between the second visual element vector of the qth visual rendering output sample and the V second visual element vectors, comprises: Performing feature enhancement on a priori feature commonality value between the visual sample vector of the qth visual rendering output sample and the V visual sample vectors and a third feature commonality value between the second visual element vector of the qth visual rendering output sample and the V visual sample vectors to obtain third common visual feature information; Determine fourth common visual feature information by using a first feature commonality value between the first visual element vector of the qth visual rendering output sample and the V visual sample vectors; determining the second local training error according to the third common visual feature information and the fourth common visual feature information corresponding to each of the X visual rendering output samples; The determining the second local training error according to the third common visual feature information and the fourth common visual feature information corresponding to each of the X visual rendering output samples comprises: iteratively summing the third common visual feature information and the fourth common visual feature information corresponding to each of the X visual rendering output samples to obtain the second local training error.

10. A data visualization display system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 9.

Images