Parallel road intelligent identification method and system based on generative label
The parallel road identification method combining generative labeling and topological adaptive graph neural network (TAGCN) solves the problems of insufficient identification accuracy and efficiency in traditional methods, achieves efficient and accurate parallel road identification, and can be extended to other geographic information processing tasks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-13
- Publication Date
- 2026-03-24
AI Technical Summary
Existing technologies struggle to efficiently identify parallel roads in urban road networks, especially in complex and diverse urban environments. Traditional methods rely on manually labeled data and have limited generalization capabilities, failing to fully utilize semantic information, resulting in insufficient recognition accuracy and efficiency.
A parallel road intelligent recognition method based on generative labeling is adopted. Combining geometric and semantic features, a recognition model is constructed using Topological Adaptive Graph Neural Network (TAGCN). By automatically generating training data and using semi-supervised learning, manual annotation is reduced, thereby improving the model's adaptability and accuracy.
It improves the accuracy and efficiency of parallel road identification, reduces labor costs, enhances the model's adaptability to different urban environments, and extends to other geographic information processing tasks such as urban functional zoning and traffic flow prediction.
Smart Images
Figure CN119785302B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of road network data processing technology, and in particular to a method and system for intelligent identification of parallel roads based on generative labels. Background Technology
[0002] Parallel roads, as the backbone of urban road networks, are crucial for understanding urban spatial structure, optimizing traffic flow distribution, and improving map data accuracy. However, with the acceleration of urbanization, the complexity of urban road networks is constantly increasing. Existing methods, mostly based on manual identification or simple geometric feature matching, can no longer meet the requirements of high accuracy and efficiency. Furthermore, while automated methods exist, they rely on large amounts of manually labeled training data, which is not only costly but also has limited generalization ability and is difficult to adapt to diverse urban environments. In summary, the main problems with existing parallel road identification technologies can be summarized as follows: 1. Difficulty in geometric feature identification: Urbanization leads to subtle road geometric features, making it difficult for traditional geometric feature-based identification methods to distinguish between parallel and non-parallel roads. 2. Complex influence from multiple factors: Parallel road identification is affected by various factors, such as road width, number of lanes, and traffic flow. The combined effect of these factors complicates the identification process. 3. Reliance on manually labeled data: Supervised learning methods rely on manually labeled training data, which not only consumes a lot of manpower and time but also limits the model's generalization ability in different urban environments. 4. Lack of effective semantic feature utilization: Existing methods fail to fully utilize the semantic information of roads, such as road names, functions, and surrounding points of interest (POIs), which are crucial for improving recognition accuracy. 5. Limited model generalization ability: Existing models perform well on specific datasets, but their performance degrades on unfamiliar data, lacking good generalization ability. Therefore, there is an urgent need for a solution capable of identifying parallel roads in spatial data at different scales to meet the needs of intelligent urban road network management. Summary of the Invention
[0003] To address this, the present invention provides a method and system for intelligent identification of parallel roads based on generative labels, which solves the problems of insufficient accuracy and efficiency of existing manual or simple geometric feature matching, and the difficulty of adapting automated methods to the diverse needs of urban environments.
[0004] According to the design scheme provided by this invention, on the one hand, a method for intelligent identification of parallel roads based on generative labels is provided, comprising:
[0005] Extract road network features from the target area;
[0006] The road network features of the target area are input into a pre-trained parallel road recognition model. The parallel road recognition model is used to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules.
[0007] As part of the parallel road intelligent recognition method based on generative labeling of this invention, further, the road network features of the target area are extracted, including:
[0008] The road network of the target area is transformed into a graph structure composed of road meshes, and multi-dimensional road network features are extracted. The road meshes represent closed areas enclosed by road segments. The multi-dimensional road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes.
[0009] As part of the parallel road intelligent recognition method based on generative labels of the present invention, the geometric features further include: the width, aspect ratio, compactness, area and perimeter of the road mesh.
[0010] As part of the parallel road intelligent recognition method based on generative labels of this invention, the model training based on training data and using a semi-supervised method further includes:
[0011] Obtain a road sample dataset, transform the road network in the road sample dataset into a graph structure composed of road meshes, and extract road network features from geometric and semantic dimensions. The road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes.
[0012] Based on preset rules, parallel road labels or non-parallel road labels are automatically generated for the road network in the road sample dataset, and training data is generated. The preset rules are road construction standards and knowledge in the road construction field.
[0013] The parallel road recognition model was trained based on the training data and using a semi-supervised learning method.
[0014] As part of the parallel road intelligent identification method based on generative labeling of the present invention, the method further includes automatically generating parallel road labels or non-parallel road labels according to preset rules, comprising:
[0015] Parallel road identification parameters are set according to road construction standards and knowledge in the field of road construction. The parallel road identification parameters include: the range of parallel road mesh width, the distribution of points of interest, and the aspect ratio threshold.
[0016] Based on the road network features in the road sample dataset, and according to the parallel road identification parameters, parallel road labels or non-parallel road labels are generated for the road network in the road sample dataset.
[0017] As a parallel road intelligent recognition method based on generative labels in this invention, the parallel road recognition model is further constructed based on a topological adaptive graph neural network and trained using a cross-entropy loss function.
[0018] As a parallel road intelligent recognition method based on generative labeling of the present invention, the parallel road recognition model further includes an input layer, a hidden layer and an output layer, wherein the hidden layer contains three topology adaptive graph neural network convolutional layers and one fully connected layer to encode the input road network graph structural features into feature vectors and generate feature classification probabilities.
[0019] Furthermore, this invention also provides a parallel road intelligent recognition system based on generative labeling, comprising: a road feature extraction module and a road feature recognition module, wherein,
[0020] The road feature extraction module is used to extract road network features of the target area;
[0021] The road feature recognition module is used to input the road network features of the target area into a pre-trained parallel road recognition model, and use the parallel road recognition model to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules.
[0022] The beneficial effects of this invention are:
[0023] This invention utilizes generative labeling to automatically generate training data, reducing labor costs, improving data processing efficiency, and decreasing reliance on manually labeled data. It combines road network geometric and semantic features to enhance the model's accuracy in identifying parallel roads. The invention employs a Topological Adaptive Graph Neural Network (TAGCN) to construct the identification model, enhancing its adaptability to road networks in different urban environments. By introducing Points of Interest (POI) data and fully utilizing road semantic information to assist in parallel road identification, this invention not only improves the efficiency and accuracy of parallel road identification but also extends to other related fields, such as urban functional zoning and traffic flow prediction, providing technical support for intelligent urban management. It has promising applications in urban planning and design, intelligent traffic management, geographic information system updates, navigation systems for autonomous vehicles, and automated analysis of urban road networks. Attached image description:
[0024] Figure 1 This is a schematic diagram of the intelligent recognition process for parallel roads based on generative labels in the embodiment.
[0025] Figure 2 This is a schematic diagram of the road network structure in the embodiment;
[0026] Figure 3 This is a schematic diagram illustrating the features of the road network in the embodiment;
[0027] Figure 4 This is a schematic diagram illustrating the overlay of vector graphics and POIs in some areas of the example, used to explain the rationality of generating labels;
[0028] Figure 5 This is a schematic diagram of the parallel road recognition framework based on TAGCN in the embodiment;
[0029] Figure 6 This is a schematic diagram of the experimental data in the example;
[0030] Figure 7 This is a schematic diagram of the parallel road recognition results from the experimental data in the example. Detailed implementation method:
[0031] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and technical solutions.
[0032] Traditional supervised learning algorithms rely on large amounts of manually labeled data, which is not only time-consuming and labor-intensive but also limits the model's generalization ability. Most existing technologies focus on analyzing the geometric features of roads while neglecting the importance of semantic information. Therefore, this invention provides embodiments, see [link to embodiments]. Figure 1 As shown, a method for intelligent recognition of parallel roads based on generative labels is provided, comprising:
[0033] S101. Extract road network features of the target area.
[0034] Specifically, the road network of the target area can be transformed into a graph structure composed of road meshes, and multi-dimensional road network features can be extracted. The road meshes represent closed areas enclosed by road segments, and the multi-dimensional road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes.
[0035] like Figure 2 and 3 As shown, a "polygonalization" operation is performed on the road network, transforming it into a graph structure composed of road meshes. Each mesh represents a closed region enclosed by road segments. If two meshes share a common edge or intersection, they are considered connected, thus constructing the graph structure of the road meshes.
[0036] Features of the road mesh are extracted from both geometric and semantic dimensions. Geometric features include mesh width, aspect ratio, compactness, area, and perimeter. These features describe the geometric morphology of the mesh, providing fundamental information for identification. Semantic features are obtained by analyzing the distribution of Points of Interest (POIs) within the mesh to acquire semantic information. For example, feature vectors can be constructed to describe the functional semantics of the mesh based on the presence of POIs representing different building types within it. By incorporating POI data related to road meshes, both geometric and semantic features are extracted, providing a richer and more accurate feature description for parallel road identification.
[0037] S102. Input the road network features of the target area into the pre-trained parallel road recognition model, and use the parallel road recognition model to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules.
[0038] Specifically, model training based on training data and using a semi-supervised method can be designed to include:
[0039] Obtain a road sample dataset, transform the road network in the road sample dataset into a graph structure composed of road meshes, and extract road network features from geometric and semantic dimensions. The road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes.
[0040] Based on preset rules, parallel road labels or non-parallel road labels are automatically generated for the road network in the road sample dataset, and training data is generated. The preset rules are road construction standards and knowledge in the road construction field.
[0041] The parallel road recognition model was trained based on the training data and using a semi-supervised learning method.
[0042] Among them, automatically generating parallel road labels or non-parallel road labels according to preset rules includes:
[0043] Parallel road identification parameters are set according to road construction standards and knowledge in the field of road construction. The parallel road identification parameters include: the range of parallel road mesh width, the distribution of points of interest, and the aspect ratio threshold.
[0044] Based on the road network features in the road sample dataset, and according to the parallel road identification parameters, parallel road labels or non-parallel road labels are generated for the road network in the road sample dataset.
[0045] like Figure 4As shown, labels for training data are automatically generated based on road construction standards and domain knowledge. Width range: The width range of PLR meshes for parallel lines roads is set according to road design specifications, such as 7.5 meters to 15 meters. POI distribution: The distribution of POIs within the mesh is used to help determine whether a mesh is PLR. Aspect ratio: A threshold is set, and the aspect ratio of the mesh is used to further determine whether a mesh is PLR. Using these criteria, PLR and non-PLR meshes are automatically labeled, generating training data with generative labels. By automatically generating training data with generative labels based on rules based on road construction standards and domain knowledge, the workload of manual labeling is significantly reduced, while improving the diversity and coverage of training samples.
[0046] The parallel road recognition model can be built based on the Topology Adaptive Graph Neural Network (TAGCN) and trained using the cross-entropy loss function. TAGCN can adapt to complex and varied node connections, making it suitable for processing the graph structure data of the road network in this embodiment. Using TAGCN to build the recognition model allows the network to adapt to graph data with complex and varied node connections, and compared to other graph neural networks, it exhibits superior performance when processing this type of data. The use of TAGCN improves the model's ability to recognize complex road networks. The model structure consists of an input layer, hidden layers, and an output layer. The hidden layer contains three TAGCN convolutional layers and one fully connected layer, responsible for encoding the input graph data into feature vectors and outputting node classification probabilities. Activation function: Leaky ReLU is selected as the activation function to alleviate the gradient vanishing problem. Loss function: Cross-entropy loss is selected to improve the model's focus on minority classes. Optimizer: The Adam optimizer is used for model training.
[0047] During training, model performance was optimized by adjusting hyperparameters (such as learning rate, number of filter layers, number of hidden layer channels, etc.). After training, the model was applied to road network sensors in other regions to identify parallel roads, and the recognition performance was evaluated based on metrics such as accuracy, precision, recall, and F1 score.
[0048] Furthermore, based on the above method, this embodiment of the invention also provides a parallel road intelligent recognition system based on generative labeling, comprising: a road feature extraction module and a road feature recognition module, wherein,
[0049] The road feature extraction module is used to extract road network features of the target area;
[0050] The road feature recognition module is used to input the road network features of the target area into a pre-trained parallel road recognition model, and use the parallel road recognition model to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules.
[0051] To verify the effectiveness of this solution, the following explanation is based on experimental data:
[0052] like Figure 5 and 6 As shown, experiments were conducted on real road datasets from cities A and B to verify the effectiveness of the proposed solution. Figure 7 As shown in the experimental results, the proposed scheme demonstrates excellent performance in parallel road recognition tasks across different regions, achieving an accuracy rate exceeding 93%. Compared to unsupervised methods, the proposed scheme shows a significant improvement in recognition accuracy. Even compared to supervised methods (such as SVM and GCN), the model in this proposed scheme exhibits performance advantages and does not rely on labeled data.
[0053] The data above demonstrates that the proposed solution reduces the need for manual annotation, lowering both manpower and time costs. By combining geometric features and semantic information, the accuracy of parallel road identification is improved. The TAGCN model and semi-supervised learning framework enhance the model's adaptability to road networks in different urban environments. Automatically generated training data and an optimized model structure improve the efficiency of model training and prediction. The use of automatically generated training data labels reduces human intervention and minimizes the impact of subjective bias on model performance, making the proposed solution applicable not only to parallel road identification but also extendable to other geographic information processing tasks, such as urban functional zoning and traffic flow prediction.
[0054] Unless otherwise specifically stated, the relative steps, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention.
[0055] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0056] The units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations are not considered to be beyond the scope of this invention.
[0057] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware or as a software functional module. This invention is not limited to any particular combination of hardware and software.
[0058] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for intelligent identification of parallel roads based on generative labels, characterized in that, Include: Extracting road network features of the target area includes: transforming the road network of the target area into a graph structure composed of road meshes, and extracting multi-dimensional road network features, wherein the road meshes represent closed areas enclosed by road segments, and the multi-dimensional road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes. The road network features of the target area are input into a pre-trained parallel road recognition model. The parallel road recognition model is used to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules. The model training based on training data and using a semi-supervised method includes: Obtain a road sample dataset, transform the road network in the road sample dataset into a graph structure composed of road meshes, and extract road network features from geometric and semantic dimensions. The road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes. Based on preset rules, parallel road labels or non-parallel road labels are automatically generated for the road network in the road sample dataset, and training data is generated. The preset rules are road construction standards and knowledge in the road construction field. The parallel road recognition model was trained based on the training data and using a semi-supervised learning method. The parallel road recognition model is built on a topological adaptive graph neural network and trained using the cross-entropy loss function. The parallel road recognition model includes an input layer, a hidden layer, and an output layer. The hidden layer contains three topology adaptive graph neural network convolutional layers and one fully connected layer to encode the structural features of the input road network graph into feature vectors and generate feature classification probabilities.
2. The intelligent identification method for parallel roads based on generative labels according to claim 1, characterized in that, The geometric features include: the width, aspect ratio, compactness, area, and perimeter of the road mesh.
3. The intelligent identification method for parallel roads based on generative labels according to claim 1, characterized in that, Automatically generate parallel road labels or non-parallel road labels based on preset rules, including: Parallel road identification parameters are set according to road construction standards and knowledge in the field of road construction. The parallel road identification parameters include: the range of parallel road mesh width, the distribution of points of interest, and the aspect ratio threshold. Based on the road network features in the road sample dataset, and according to the parallel road identification parameters, parallel road labels or non-parallel road labels are generated for the road network in the road sample dataset.
4. A parallel road intelligent recognition system based on generative tags, characterized in that, Includes: a road feature extraction module and a road feature recognition module. in, The road feature extraction module is used to extract road network features of the target area, including: converting the road network of the target area into a graph structure composed of road meshes, and extracting multi-dimensional road network features, wherein the road meshes represent closed areas enclosed by road segments, and the multi-dimensional road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes. The road feature recognition module is used to input the road network features of the target area into a pre-trained parallel road recognition model, and use the parallel road recognition model to obtain the parallel roads of the road network in the target area. The parallel road recognition model is trained based on training data and using a semi-supervised method. The training data automatically generates labels for the road sample dataset according to preset rules. The model training based on training data and using a semi-supervised method includes: Obtain a road sample dataset, transform the road network in the road sample dataset into a graph structure composed of road meshes, and extract road network features from geometric and semantic dimensions. The road network features include geometric features for describing the geometric shape of the road meshes and semantic features for describing the distribution of interest points inside the road meshes. Based on preset rules, parallel road labels or non-parallel road labels are automatically generated for the road network in the road sample dataset, and training data is generated. The preset rules are road construction standards and knowledge in the road construction field. The parallel road recognition model was trained based on the training data and using a semi-supervised learning method. The parallel road recognition model is built on a topological adaptive graph neural network and trained using the cross-entropy loss function. The parallel road recognition model includes an input layer, a hidden layer, and an output layer. The hidden layer contains three topology adaptive graph neural network convolutional layers and one fully connected layer to encode the structural features of the input road network graph into feature vectors and generate feature classification probabilities.
5. An electronic device, characterized in that, include: At least one processor, and a memory coupled to said at least one processor; The memory stores a computer program that can be executed by the at least one processor to implement the method as described in any one of claims 1 to 3.
6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, enables the implementation of the method as described in any one of claims 1 to 3.