Geographic surveying and mapping data accurate processing method based on machine learning

Through the combination of multiple deep learning technologies, the spatial, statistical and texture characteristics of geographic surveying and mapping data are extracted and generated, and the problems of insufficient data processing accuracy and efficiency in the existing technology are solved, achieving high precision, high efficiency and high authenticity Geographic surveying and mapping data processing.

CN119027793AInactive Publication Date: 2024-11-26YUYAO PLANNING SURVEYING & DESIGN CO LTD
View PDF 0 Cites 6 Cited by

Patent Information

Application Number
CN202411184096.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively improve processing accuracy and efficiency in geographic surveying and mapping data processing. Especially when facing large-scale, high-dimensional and complex and diverse data, traditional methods are easily affected by noise and distortion, and a single deep learning technology is difficult to comprehensively Improve the processing effect.

Method used

Combining various deep learning technologies such as generative adversarial networks (GAN), graph convolutional networks (GCN), variational autoencoders (VAE), self-attention mechanisms, convolutional neural networks (CNN) and visual Transformer (ViT), Through multi-technology fusion methods, spatial features, statistical features and texture features are extracted and generated, and the ability to design multi-objective loss function optimization generators and discriminators is achieved to achieve high accuracy, high authenticity and high efficiency processing of data.

Benefits of technology

It significantly improves the processing accuracy and efficiency of geographic surveying and mapping data, ensures the authenticity and consistency of the generated data in various dimensions, enhances the robustness of the model and the quality of the generated data, and meets the practical application needs of geographic surveying and mapping data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027793A_ABST
    Figure CN119027793A_ABST
Patent Text Reader

Abstract

The invention discloses a machine learning-based geographic surveying and mapping data precise processing method, which comprises the following steps of S1, collecting geographic surveying and mapping data, and constructing a data set; s2, designing a generative adversarial network; s3, extracting and generating spatial features from the data set by using the graph convolutional network of the generator and a Transform model; s4, extracting and generating statistical features from the data set by using a variational auto-encoder and a self-attention mechanism of the generator; s5, using the convolutional neural network of the generator and a visual Transform model to extract and generate texture features from the data set; s6, fusing the spatial features, the statistical features and the texture features through a full connection layer, and generating final geographic surveying and mapping data; s7, discriminating the authenticity of the final geographic surveying and mapping data by using a discriminator; and S8, designing a multi-objective loss function, and optimizing the generation capability of the generator and the discrimination capability of the discriminator. The geographic surveying and mapping data is precisely processed by utilizing the generative adversarial network and the deep learning technology, and the method has the advantages of high precision and high efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of geographic surveying and mapping data processing, and in particular to a method for accurately processing geographic surveying and mapping data based on machine learning. Background Art

[0002] Geographic surveying and mapping data are widely used in the fields of land and resources management, urban planning, environmental monitoring, disaster warning, etc. With the development of remote sensing technology and geographic information system (GIS), the acquisition of geographic surveying and mapping data has become more convenient and efficient. However, the existing technology still has many shortcomings in the processing and accuracy improvement of geographic surveying and mapping data, and a new method is urgently needed to improve the processing accuracy and efficiency of geographic surveying and mapping data.

[0003] Existing geographic surveying and mapping data processing methods mainly rely on traditional image processing and statistical analysis techniques. Although these methods can meet the processing needs of geographic surveying and mapping data to a certain extent, their processing efficiency and accuracy are often unsatisfactory when faced with large-scale, high-dimensional and complex and diverse data. For example, when processing texture features in geographic surveying and mapping data, traditional image processing technology is easily affected by noise and distortion, resulting in reduced data accuracy. In addition, when processing statistical features in geographic surveying and mapping data, traditional statistical analysis methods often rely on prior assumptions and empirical models, making it difficult to accurately capture the potential distribution and feature relationships of the data.

[0004] In recent years, with the rapid development of deep learning technology, especially the emergence of technologies such as generative adversarial networks (GAN), graph convolutional networks (GCN), variational autoencoders (VAE), self-attention mechanisms, convolutional neural networks (CNN) and visual transformers (ViT), new solutions have been provided for geographic surveying and mapping data processing. Generative adversarial networks (GAN) can generate high-quality geographic surveying and mapping data through adversarial training of generators and discriminators. However, when processing geographic surveying and mapping data, GAN often ignores the complex relationship between the spatial features, statistical features and texture features of the data, resulting in unstable quality of generated data.

[0005] Graph convolutional network (GCN) is a neural network that specializes in processing graph structure data. It can effectively extract information about nodes and their neighboring nodes through convolution operations. Applying GCN to geographic surveying and mapping data processing can better capture the relationship between spatial locations and improve the ability to extract spatial features of data. However, it is difficult to fully mine other feature information in the data using GCN alone.

[0006] Variational autoencoders (VAEs) generate new data by learning the probability distribution of data, and can retain the statistical properties of features while generating data. Applying VAE to the statistical feature processing of geographic surveying and mapping data can effectively capture and generate high-dimensional statistical features. However, when VAE processes high-dimensional data, the features generated may be fuzzy and imprecise.

[0007] The self-attention mechanism captures the internal relationship of the data by calculating the correlation between the various parts of the input data, and is widely used in various deep learning models. The self-attention mechanism has significant advantages in feature enhancement, and can enhance the generation and discrimination capabilities of statistical features in the generator and discriminator. However, it is difficult to fully improve the processing effect of geographic surveying and mapping data using the self-attention mechanism alone.

[0008] Convolutional neural network (CNN) is a neural network that specializes in processing image data. It extracts local features of images through convolutional layers. Applying CNN to texture feature processing in geographic surveying and mapping data can effectively extract and generate low-level texture features. However, CNN has certain limitations in capturing high-level texture features.

[0009] Therefore, how to provide a method for accurately processing geographic surveying and mapping data based on machine learning is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0010] One purpose of the present invention is to propose a method for accurately processing geographic surveying and mapping data based on machine learning. The present invention makes full use of various deep learning technologies such as generative adversarial networks (GANs), graph convolutional networks (GCNs), variational autoencoders (VAEs), self-attention mechanisms, convolutional neural networks (CNNs), and visual transformers (ViTs), and accurately processes geographic surveying and mapping data through a multi-technology fusion method. The method can effectively extract and generate spatial features, statistical features, and texture features, and has the advantages of high precision, high authenticity, and high efficiency, significantly improving the processing effect and application value of geographic surveying and mapping data.

[0011] A method for accurately processing geographic surveying and mapping data based on machine learning according to an embodiment of the present invention comprises the following steps: S1. Collect geographic surveying and mapping data using remote sensing technology and geographic information systems, perform preprocessing, and construct data sets; S2. Design a generative adversarial network, including the neural network models of the generator and the discriminator; S3, using the generator’s graph convolutional network and Transformer model to extract and generate spatial features from the dataset; S4. Extract and generate statistical features from the dataset using the variational autoencoder and self-attention mechanism of the generator; S5, using the generator's convolutional neural network and visual Transformer model to extract and generate texture features from the dataset; S6, fusing spatial features, statistical features and texture features through a fully connected layer to generate the final geographic mapping data; S7. Use the discriminator to determine the authenticity of the final geographic surveying and mapping data, including using the graph convolutional network and the Transformer model to determine the authenticity of spatial features, using the variational autoencoder and the self-attention mechanism to determine the authenticity of statistical features, and using the convolutional neural network and the visual Transformer model to determine the authenticity of texture features; S8. Design a multi-objective loss function, including generator loss and discriminator loss, to optimize the generation ability of the generator and the discrimination ability of the discriminator.

[0012] Furthermore, the S3 specifically includes: S31. Define the input of the graph convolutional network as graph structure data ,in Represents a collection of nodes. Represents a set of edges, each node With characteristic vector ; S32. Perform convolution operation on the node features and calculate the new feature representation of each node: ; in, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node The neighbor nodes of Representation Node The degree, Representation Node The degree, and represents the trainable weight matrix, represents the activation function; S33. Represent the spatial features extracted by the graph convolutional network Input into the Transformer model, defining the input of the Transformer model as a feature sequence ,in Indicates spatial eigenvectors; S34. Calculate the self-attention representation of each feature vector: ; in, , , , represents the query matrix, represents the key matrix, represents the value matrix, , and represents the trainable weight matrix, represents the dimension of the key vector; S35. Calculate multiple self-attention representations and concatenate them through Transformer's multi-head self-attention mechanism: ; in, , , , and represents a trainable weight matrix; S36. Output of the graph convolutional network Fusion with Transformer's multi-head self-attention output: ; in, , and Represents an adjustable weight parameter used to balance the weighted sum of graph convolutional network output, Transformer self-attention output, and node features. Representation Node The trainable weights of Representation Node The spatial eigenvector of Representation Node The degree, Represents the total number of spatial features; S37, use the feedforward neural network to perform nonlinear transformation on each fused feature representation, and generate the final spatial feature through the stacking of multiple fusion layers and feedforward neural networks. .

[0013] Furthermore, the S4 specifically includes: S41. Input the data related to the statistical features in the data set into the encoder of the variational autoencoder. The encoder maps the input data to the latent space to obtain the latent variable representation. ,in is from a Gaussian distribution The latent variable representation obtained by sampling from ; S42. Calculate latent variable representation The mean and variance : ; ; in, Represents input data, and The weight matrices representing the mean and variance, and The bias vectors representing the mean and variance respectively; S43. Sampling in latent space to obtain latent variable representation : ; in, represents bit-wise multiplication of elements, represents noise sampled from a standard normal distribution; S44, input the latent variable representation into the decoder of the variational autoencoder, and the decoder Remap to the original data space to generate statistical features ; S45. Calculate and generate statistical features and input data The reconstruction error between: ; in, represents the Euclidean distance; S46. Use the self-attention mechanism to enhance the generation ability of statistical features. The self-attention mechanism calculates the self-attention weight matrix of the input data : ; in, represents the query matrix, represents the key matrix, represents the dimension of the key vector; S47, the self-attention weight matrix Apply to input data , and the enhanced statistical feature representation is obtained : ; in, represents the value matrix; S48, statistical features will be generated And the enhanced statistical feature representation Combine to generate the final statistical features .

[0014] Furthermore, the S5 specifically includes: S51. Construct a multi-layer convolutional network. In each convolutional layer, extract local texture features through convolution operations: ; in, represents the convolution output, represents the weight of the convolution kernel, Represents the pixel value of the input image; S52, apply activation function after each convolution layer , use the pooling layer to downsample and reduce the size of the feature map: ; in, represents the pooling window; S53, serializing the low-level texture features into a one-dimensional vector sequence as the input of the visual Transformer model, in which the self-attention mechanism is used to capture the long-distance dependencies between features and extract high-level texture features; S54, converting high-level texture features back to two-dimensional feature maps to generate final texture features .

[0015] Furthermore, the S6 specifically includes: S61, the spatial features generated by the generator Input to the fully connected layer; S62, the statistical features generated by the generator Input to the fully connected layer; S63, the texture features generated by the generator Input to the fully connected layer; S64, perform linear transformation and weighted summation to obtain the fused features : ; in, , and Respectively represent the weight matrices of spatial features, statistical features, and texture features in the fully connected layer, , and Represent the bias items of spatial features, statistical features and texture features respectively, , and represents the fusion weight parameter, represents the rectified linear unit activation function; S65. The fused features Through the activation function Perform nonlinear transformation to increase the nonlinear expression ability of the model: ; S66, the features processed by the activation function As the final geographic surveying and mapping data output, the output data has higher accuracy and richer feature information, and can better meet the actual application needs of geographic surveying and mapping.

[0016] Furthermore, the S8 specifically includes: S81. Design a multi-objective loss function, including generator loss and discriminator loss, where the generator loss includes adversarial loss, feature matching loss and reconstruction loss, and the discriminator loss includes adversarial loss; S82. Calculate the generator loss: ; in, and Represents the importance weight parameter for balancing each loss item, Represents resistance to loss, represents the feature matching loss, represents the reconstruction loss; S83. Define adversarial loss : ; in, Represents real data, Represents the discriminator's judgment result on the real data, represents the input noise of the generator, Represents the discriminator's judgment result on the generated data; S84. Define feature matching loss : ; in, represents the feature extraction function, Indicates the generated data. represents the L2 norm; S85. Define reconstruction loss : ; in, represents the encoder, Represents reconstructed data; S86. Calculate the discriminator loss: ; S87. Use automated machine learning techniques to search and optimize model structures and hyperparameters.

[0017] The beneficial effects of the present invention are: First, by using graph convolutional networks (GCN) and Transformer models to extract and generate spatial features, the spatial relationships in geographic surveying and mapping data are effectively captured, and the extraction accuracy of spatial features is improved. At the same time, the application of variational autoencoders (VAE) and self-attention mechanisms in statistical feature extraction and generation enables the statistical features of the data to retain the probability distribution of the original data, and enhances the generation ability of statistical features and improves the generation accuracy of statistical features. In addition, the combination of convolutional neural networks (CNN) and visual transformers (ViT) fully extracts low-level and high-level texture features in geographic surveying and mapping data, significantly improving the detail retention and generation quality of texture features.

[0018] Secondly, by using the graph convolutional network, Transformer model, variational autoencoder and self-attention mechanism in the discriminator, the spatial features, statistical features and texture features of the generated data are discriminated to ensure the authenticity and consistency of the generated data in all dimensions. At the same time, a multi-objective loss function is designed, combining the adversarial loss, feature matching loss and reconstruction loss of the generator and the discriminator to optimize the generation ability of the generator and the discrimination ability of the discriminator, further improving the authenticity of the generated data.

[0019] In addition, the feature fusion operation of the fully connected layer organically integrates spatial features, statistical features, and texture features to generate high-quality geographic surveying and mapping data, achieving efficient integration of data processing.

[0020] Finally, through the organic combination of multiple deep learning technologies, the advantages of each are fully utilized to comprehensively improve the overall performance of the model. The combination of graph convolutional network and Transformer enhances the ability to capture spatial features, the combination of variational autoencoder and self-attention mechanism enhances the ability to generate statistical features, and the combination of convolutional neural network and visual Transformer improves the ability to retain details of texture features. The design of multi-objective loss function balances the losses of the generator and discriminator, improving the robustness of the model and the quality of generated data. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1A flowchart of a method for accurately processing geographic surveying and mapping data based on machine learning proposed by the present invention; Figure 2 This is a schematic diagram of the generative adversarial network structure of a method for accurately processing geographic surveying and mapping data based on machine learning proposed in the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0023] refer to Figure 1 and Figure 2 , a method for accurately processing geographic surveying and mapping data based on machine learning, comprising the following steps: S1. Collect geographic surveying and mapping data using remote sensing technology and geographic information systems, perform preprocessing, and construct data sets; S2. Design a generative adversarial network, including the neural network models of the generator and the discriminator; S3, using the generator’s graph convolutional network and Transformer model to extract and generate spatial features from the dataset; S4. Extract and generate statistical features from the dataset using the variational autoencoder and self-attention mechanism of the generator; S5, using the generator's convolutional neural network and visual Transformer model to extract and generate texture features from the dataset; S6, fusing spatial features, statistical features and texture features through a fully connected layer to generate the final geographic mapping data; S7. Use the discriminator to determine the authenticity of the final geographic surveying and mapping data, including using the graph convolutional network and the Transformer model to determine the authenticity of spatial features, using the variational autoencoder and the self-attention mechanism to determine the authenticity of statistical features, and using the convolutional neural network and the visual Transformer model to determine the authenticity of texture features; S8. Design a multi-objective loss function, including generator loss and discriminator loss, to optimize the generation ability of the generator and the discrimination ability of the discriminator.

[0024] In this implementation, S3 specifically includes: S31. Define the input of the graph convolutional network as graph structure data ,in Represents a collection of nodes. Represents a set of edges, each node With characteristic vector ; S32. Perform convolution operation on the node features and calculate the new feature representation of each node: ; in, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node The neighbor nodes of Representation Node The degree, Representation Node The degree, and represents the trainable weight matrix, represents the activation function; S33. Represent the spatial features extracted by the graph convolutional network Input into the Transformer model, defining the input of the Transformer model as a feature sequence ,in Indicates spatial eigenvectors; S34. Calculate the self-attention representation of each feature vector: ; in, , , , represents the query matrix, represents the key matrix, represents the value matrix, , and represents the trainable weight matrix, represents the dimension of the key vector; S35. Calculate multiple self-attention representations and concatenate them through Transformer's multi-head self-attention mechanism: ; in, , , , and represents a trainable weight matrix; S36. Output of the graph convolutional network Fusion with Transformer's multi-head self-attention output: ; in, , and Represents an adjustable weight parameter used to balance the weighted sum of graph convolutional network output, Transformer self-attention output, and node features. Representation Node The trainable weights of Representation Node The spatial eigenvector of Representation Node The degree, Represents the total number of spatial features; S37, use the feedforward neural network to perform nonlinear transformation on each fused feature representation, and generate the final spatial feature through the stacking of multiple fusion layers and feedforward neural networks. .

[0025] In this implementation, S4 specifically includes: S41. Input the data related to the statistical features in the data set into the encoder of the variational autoencoder. The encoder maps the input data to the latent space to obtain the latent variable representation. ,in is from a Gaussian distribution The latent variable representation obtained by sampling from ; S42. Calculate latent variable representation The mean and variance : ; ; in, Represents input data, and The weight matrices representing the mean and variance, and The bias vectors representing the mean and variance respectively; S43. Sampling in latent space to obtain latent variable representation : ; in, represents bit-wise multiplication of elements, represents noise sampled from a standard normal distribution; S44, input the latent variable representation into the decoder of the variational autoencoder, and the decoder Remap to the original data space to generate statistical features ; S45. Calculate and generate statistical features and input data The reconstruction error between: ; in, represents the Euclidean distance; S46. Use the self-attention mechanism to enhance the generation ability of statistical features. The self-attention mechanism calculates the self-attention weight matrix of the input data : ; in, represents the query matrix, represents the key matrix, represents the dimension of the key vector; S47, the self-attention weight matrix Apply to input data , and the enhanced statistical feature representation is obtained : ; in, represents the value matrix; S48, statistical features will be generated And the enhanced statistical feature representation Combine to generate the final statistical features .

[0026] In this implementation manner, S5 specifically includes: S51. Construct a multi-layer convolutional network. In each convolutional layer, extract local texture features through convolution operations: ; in, represents the convolution output, represents the weight of the convolution kernel, Represents the pixel value of the input image; S52, apply activation function after each convolution layer , use the pooling layer to downsample and reduce the size of the feature map: ; in, represents the pooling window; S53, serializing the low-level texture features into a one-dimensional vector sequence as the input of the visual Transformer model, in which the self-attention mechanism is used to capture the long-distance dependencies between features and extract high-level texture features; S54, converting high-level texture features back to two-dimensional feature maps to generate final texture features .

[0027] In this implementation manner, S6 specifically includes: S61, the spatial features generated by the generator Input to the fully connected layer; S62, the statistical features generated by the generator Input to the fully connected layer; S63, the texture features generated by the generator Input to the fully connected layer; S64, perform linear transformation and weighted summation to obtain the fused features : ; in, , and Respectively represent the weight matrices of spatial features, statistical features, and texture features in the fully connected layer, , and Represent the bias items of spatial features, statistical features and texture features respectively, , and represents the fusion weight parameter, represents the rectified linear unit activation function; S65. The fused features Through the activation function Perform nonlinear transformation to increase the nonlinear expression ability of the model: ; S66, the features processed by the activation function As the final geographic surveying and mapping data output, the output data has higher accuracy and richer feature information, and can better meet the actual application needs of geographic surveying and mapping.

[0028] In this implementation manner, S8 specifically includes: S81. Design a multi-objective loss function, including generator loss and discriminator loss, where the generator loss includes adversarial loss, feature matching loss and reconstruction loss, and the discriminator loss includes adversarial loss; S82. Calculate the generator loss: ; in, and Represents the importance weight parameter for balancing each loss item, Represents resistance to loss, represents the feature matching loss, represents the reconstruction loss; S83. Define adversarial loss : ; in, Represents real data, Represents the discriminator's judgment result on the real data, represents the input noise of the generator, Represents the discriminator's judgment result on the generated data; S84. Define feature matching loss : ; in, represents the feature extraction function, Indicates the generated data. represents the L2 norm; S85. Define reconstruction loss : ; in, represents the encoder, Represents reconstructed data; S86. Calculate the discriminator loss: ; S87. Use automated machine learning techniques to search and optimize model structures and hyperparameters.

[0029] Embodiment 1: In order to verify the feasibility of the present invention in implementation, the present invention is applied to a certain urban planning project. The municipal government hopes to support urban infrastructure construction and optimize traffic layout through high-precision geographic surveying and mapping data. Traditional surveying and mapping data processing methods cannot meet the high requirements of urban planning due to insufficient accuracy and low processing efficiency. In order to improve the accuracy and processing efficiency of geographic surveying and mapping data, the municipal government decided to adopt a precise processing method for geographic surveying and mapping data based on machine learning.

[0030] The municipal government first used remote sensing technology and geographic information systems (GIS) to collect the city's geographic surveying and mapping data, including topographic maps, building distribution, road networks, etc. After preprocessing, these data were used to construct a comprehensive data set containing spatial features, statistical features, and texture features. In order to improve the accuracy and efficiency of data processing, the municipal government chose a precise processing method for geographic surveying and mapping data based on machine learning.

[0031] The core of this method is the Generative Adversarial Network (GAN), which includes the neural network models of the generator and the discriminator. The generator is responsible for extracting and generating high-quality spatial features, statistical features, and texture features from the dataset, while the discriminator is used to determine the authenticity of the generated data. Specifically, the generator uses a graph convolutional network (GCN) and a Transformer model to extract and generate spatial features, a variational autoencoder (VAE) and a self-attention mechanism to extract and generate statistical features, and a convolutional neural network (CNN) and a visual Transformer (ViT) to extract and generate texture features.

[0032] In the data processing process, the graph convolutional network (GCN) is first used to extract spatial features such as latitude, longitude and elevation data from geographic surveying data, and the Transformer model is used to further enhance these features to ensure that the complex relationship between spatial locations is captured. Next, the generator extracts statistical features such as the mean and variance of terrain undulations from the dataset through the variational autoencoder (VAE), and enhances the generation ability of these features through the self-attention mechanism. Then, the convolutional neural network (CNN) is used to extract low-level texture features in the data, such as edge information of roads and buildings, and the visual transformer (ViT) further extracts high-level texture features to ensure the details and accuracy of the generated data.

[0033] After the generator generates data, the discriminator uses graph convolutional networks and Transformer models to determine the authenticity of spatial features, uses variational autoencoders and self-attention mechanisms to determine the authenticity of statistical features, and uses convolutional neural networks and visual Transformer models to determine the authenticity of texture features. In this process, the generator's generation ability and the discriminator's discrimination ability are optimized through multi-objective loss functions, including generator loss and discriminator loss, to ensure high accuracy and authenticity of the generated data.

[0034] Table 1 Comparison of geographic surveying and mapping data processing effects based on machine learning Table 1 above covers several key indicators such as data processing time, terrain height error, building location error, commuting time reduction, flood warning accuracy and economic loss avoidance.

[0035] First, the machine learning-based method performs well in terms of data processing time. It only takes 24 hours to process 50 square kilometers of geographic surveying and mapping data, while the traditional method takes 72 hours, which is 3 times more efficient. This significant efficiency improvement is mainly due to the faster data processing achieved by the machine learning method through automation and efficient algorithms, which is particularly important for real-time data processing and decision support in emergency situations.

[0036] Secondly, in terms of terrain height error, the machine learning-based method reduces the error to ±1 meter, while the traditional method has an error of ±5 meters, which is a 5-fold improvement in accuracy. This significant improvement in accuracy means that more accurate terrain information can be provided in applications such as urban planning, environmental monitoring, and disaster warning, thereby reducing planning errors and decision-making errors caused by data errors.

[0037] In terms of processing building location errors, the machine learning-based method reduces the error to ±2 meters, while the traditional method has an error of ±10 meters, and the accuracy is also improved by 5 times. This is of great significance for urban planning, building management and emergency response, and can provide more accurate building distribution information and support more sophisticated management and decision-making.

[0038] In terms of reducing commuting time, the machine learning-based method reduced residents’ commuting time by about 15 minutes by optimizing public transportation routes. Although there is no specific data in this regard for traditional methods, it is obvious that traffic optimization based on high-precision geographic mapping data can directly improve the quality of life of citizens and increase the efficiency of public transportation.

[0039] In terms of flood warning accuracy, the machine learning-based method has increased the accuracy of warnings from 80% of traditional methods to 100%, an increase of 20%. This improvement is particularly critical in disaster warning, as it can more accurately predict disasters and take measures in advance to reduce losses. Specific data shows that the flood warning system based on precise terrain data has effectively avoided economic losses of about 5 million yuan.

[0040] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for accurately processing geographic surveying and mapping data based on machine learning, characterized in that: The steps include: S1. Collect geographic surveying and mapping data using remote sensing technology and geographic information systems, perform preprocessing, and construct data sets; S2. Design a generative adversarial network, including the neural network models of the generator and the discriminator; S3, using the generator’s graph convolutional network and Transformer model to extract and generate spatial features from the dataset; S4. Extract and generate statistical features from the dataset using the variational autoencoder and self-attention mechanism of the generator; S5, using the generator's convolutional neural network and visual Transformer model to extract and generate texture features from the dataset; S6, fusing spatial features, statistical features and texture features through a fully connected layer to generate the final geographic mapping data; S7. Use the discriminator to determine the authenticity of the final geographic surveying and mapping data, including using the graph convolutional network and the Transformer model to determine the authenticity of spatial features, using the variational autoencoder and the self-attention mechanism to determine the authenticity of statistical features, and using the convolutional neural network and the visual Transformer model to determine the authenticity of texture features; S8. Design a multi-objective loss function, including generator loss and discriminator loss, to optimize the generation ability of the generator and the discrimination ability of the discriminator.

2. The method for accurately processing geographic surveying and mapping data based on machine learning according to claim 1, characterized in that: The S3 specifically includes: S31. Define the input of the graph convolutional network as graph structure data ,in Represents a collection of nodes. Represents a set of edges, each node With characteristic vector ; S32. Perform convolution operation on the node features and calculate the new feature representation of each node: ; in, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node In the The feature representation of the layer, Representation Node The neighbor nodes are combined, Representation Node The degree, Representation Node The degree, and represents the trainable weight matrix, represents the activation function; S33. Represent the spatial features extracted by the graph convolutional network Input into the Transformer model, defining the input of the Transformer model as a feature sequence ,in Indicates spatial eigenvectors; S34. Calculate the self-attention representation of each feature vector: ; in, , , , represents the query matrix, represents the key matrix, represents the value matrix, , and represents the trainable weight matrix, represents the dimension of the key vector; S35. Calculate multiple self-attention representations and concatenate them through Transformer's multi-head self-attention mechanism: ; in, , , , and represents a trainable weight matrix; S36. Output of the graph convolutional network Fusion with Transformer's multi-head self-attention output: ; in, , and Represents an adjustable weight parameter used to balance the weighted sum of graph convolutional network output, Transformer self-attention output, and node features. Representation Node The trainable weights of Representation Node The spatial eigenvector of Representation Node The degree, Represents the total number of spatial features; S37, use the feedforward neural network to perform nonlinear transformation on each fused feature representation, and generate the final spatial feature through the stacking of multiple fusion layers and feedforward neural networks. .

3. The method for accurately processing geographic surveying and mapping data based on machine learning according to claim 1, characterized in that: The S4 specifically includes: S41. Input the data related to the statistical features in the data set into the encoder of the variational autoencoder. The encoder maps the input data to the latent space to obtain the latent variable representation. ,in is from a Gaussian distribution The latent variable representation obtained by sampling from S42. Calculate latent variable representation The mean and variance : ; ; in, Represents input data, and The weight matrices representing the mean and variance, and The bias vectors representing the mean and variance respectively; S43. Sampling in latent space to obtain latent variable representation : ; in, represents bit-wise multiplication of elements, represents noise sampled from a standard normal distribution; S44, input the latent variable representation into the decoder of the variational autoencoder, and the decoder Remap to the original data space and generate statistical features ; S45. Calculate and generate statistical features and input data The reconstruction error between: ; in, represents the Euclidean distance; S46. Use the self-attention mechanism to enhance the generation ability of statistical features. The self-attention mechanism calculates the self-attention weight matrix of the input data : ; in, represents the query matrix, represents the key matrix, represents the dimension of the key vector; S47, the self-attention weight matrix Apply to input data , and the enhanced statistical feature representation is obtained : ; in, represents the value matrix; S48, statistical features will be generated And the enhanced statistical feature representation Combine to generate the final statistical features .

4. The method for accurately processing geographic surveying and mapping data based on machine learning according to claim 1, characterized in that: The S5 specifically includes: S51. Construct a multi-layer convolutional network. In each convolutional layer, extract local texture features through convolution operations: ; in, represents the convolution output, represents the weight of the convolution kernel, Represents the pixel value of the input image; S52, apply activation function after each convolution layer , use the pooling layer to downsample and reduce the size of the feature map: ; in, represents the pooling window; S53, serializing the low-level texture features into a one-dimensional vector sequence as the input of the visual Transformer model, in which the self-attention mechanism is used to capture the long-distance dependencies between features and extract high-level texture features; S54, converting high-level texture features back to two-dimensional feature maps to generate final texture features .

5. The method for accurately processing geographic surveying and mapping data based on machine learning according to claim 1, characterized in that: The S6 specifically includes: S61, the spatial features generated by the generator Input to the fully connected layer; S62, the statistical features generated by the generator Input to the fully connected layer; S63, the texture features generated by the generator Input to the fully connected layer; S64, perform linear transformation and weighted summation to obtain the fused features : ; in, , and Respectively represent the weight matrices of spatial features, statistical features, and texture features in the fully connected layer, , and Represent the bias items of spatial features, statistical features and texture features respectively, , and represents the fusion weight parameter, represents the rectified linear unit activation function; S65. The fused features Through the activation function Perform a nonlinear transformation: ; S66, the features processed by the activation function Output as final geographic mapping data.

6. The method for accurately processing geographic surveying and mapping data based on machine learning according to claim 1, characterized in that: The S8 specifically includes: S81. Design a multi-objective loss function, including generator loss and discriminator loss, where the generator loss includes adversarial loss, feature matching loss and reconstruction loss, and the discriminator loss includes adversarial loss; S82. Calculate the generator loss: ; in, and Represents the importance weight parameter for balancing each loss item, Represents resistance to loss, represents the feature matching loss, represents the reconstruction loss; S83. Define adversarial loss : ; in, Represents real data, Represents the discriminator's judgment result on the real data, represents the input noise of the generator, Represents the discriminator's judgment result on the generated data; S84. Define feature matching loss : ; in, represents the feature extraction function, Indicates the generated data. represents the L2 norm; S85. Define reconstruction loss : ; in, represents the encoder, Represents reconstructed data; S86. Calculate the discriminator loss: ; S87. Use automated machine learning techniques to search and optimize model structures and hyperparameters.

Citation Information

Cited By

  • Land property detection method based on space-time adaptive fusion and geographic features

    CN120635657A

  • Land property detection method based on spatio-temporal adaptive fusion and geographical features

    CN120635657B

  • Deep learning-based road and bridge construction measurement data processing method and system

    CN120726519A

  • Peanut quality character selective breeding method based on whole genome SNP (Single Nucleotide Polymorphism) and application

    CN121171335A

  • Peanut quality trait selection breeding method based on whole genome snp and application

    CN121171335B