Multi-mode mixed road network decomposition and classification device fusing Transform model

Abstract

The invention discloses an urban multi-mode hybrid road network decomposition and classification prediction device fused with a Transform model. According to the method, the classic geography partitioning and layering thought is comprehensively applied, and the accuracy and stability of road network mode decomposition are remarkably improved through the two dimensions of geographic space and ground feature attributes. Meanwhile, through an independently researched and developed feature vector extraction algorithm, space structure and attribute information in the road network are efficiently mined, and eight road section features of the road network from macroscopic, mesoscopic and microscopic are extracted, including the distance between a road section midpoint and a road network central point, the direction of a connecting line between the road section midpoint and the road network central point, the number of connecting edges of two end points of the road section, the length of the road section and the like. Based on the feature vectors, an improved Transform model is utilized to realize classification prediction of a road grid mode, a parallel mode, a radiation mode, a complex interchange mode, an irregular complex interchange mode, a fishbone mode and other irregular modes. The invention provides an accurate and efficient multi-mode mixed road network decomposition and classification identification method for the fields of geographic information system analysis, urban and rural planning, resource optimization configuration and the like.

Description

Technical Field

[0001] The present invention relates to the fields of geographic information systems and deep learning, and specifically to a device for decomposing and classifying urban multi-mode hybrid road network vector data. Background Art

[0002] Road networks have distinct spatial distribution patterns, and their pattern recognition plays a crucial role in many fields, including cartography, map matching, and spatial query. Due to the differences in scale and configuration of urban elements, road networks form a variety of patterns. Road network pattern recognition is an important component of geographic information system research and map generalization.

[0003] Currently, existing road network pattern decomposition methods exist in a variety of theoretical and practical approaches. Many rely on road mesh features and human experience. However, road mesh features themselves have certain limitations, and the subjectivity and uncertainty of human experience lead to significant drawbacks in practical applications of these methods. Furthermore, most current road network pattern classification methods are confined to a single pattern classification algorithm framework, which is unable to fully adapt to and accurately analyze the complex characteristics of road networks under different patterns, thus limiting the efficiency, accuracy, and reliability of classification results.

[0004] Therefore, how to accurately decompose the patterns of complex urban road networks, how to effectively extract the characteristics of road network patterns, and how to efficiently identify multi-modal mixed road network patterns have become key issues that need to be solved urgently. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a multi-modal hybrid road network decomposition and classification method that integrates the Transformer model, which not only meets people's demand for urban multi-modal road networks, but also can accurately reflect the road network mode type.

[0006] To achieve the above objectives, the present invention provides a multi-mode hybrid road network decomposition and classification device integrating a Transformer model, comprising:

[0007] Based on the geographical zoning concept, main highways or autonomous driving lanes, main roads or vehicle lanes, primary roads and secondary roads are selected as the basis for main road zoning;

[0008] Based on the principles of spatial similarity and attribute similarity, the third-level roads, residential roads, undefined roads and complex interchange roads within the block are finely decomposed;

[0009] Based on the different characteristics of the road network, eight types of road segment features are constructed from macroscopic, mesoscopic to microscopic levels, including: the distance between the midpoint of the road segment and the center point of the road network, the direction of the line connecting the midpoint of the road segment and the center point of the road network, the number of edges connecting the two end points of the road segment, the length of the road segment, the aspect ratio of the minimum circumscribed rectangle of the road segment, the direction of the longest side of the minimum circumscribed rectangle of the road segment, and the proportion of the road segment length to the total length of the road network. The characteristic quantities of each road segment in the road network are obtained through calculation;

[0010] The nodes with an angle less than 110 degrees in the road section are segmented, and finally, the calculated feature quantities are converted into vector matrices.

[0011] The road network patterns are divided into seven categories: grid pattern, parallel pattern, radial pattern, complex interchange pattern, irregular complex interchange pattern, fishbone pattern and other irregular patterns.

[0012] The road network feature vector and classification dataset are input into the Transformer model and trained through the encoding layer input;

[0013] The trained model is evaluated for accuracy, and the targets to be classified are detected to ultimately obtain the accurate target type.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects:

[0015] (1) By introducing the concept of geographical zoning and stratification, the present invention can intelligently decompose different road network patterns according to the spatial distribution and attribute characteristics of the road network. Even when faced with complex and unevenly distributed road networks, it can be processed efficiently and flexibly to ensure accurate division. This decomposition method not only improves efficiency and accuracy, but is also particularly suitable for processing large-scale urban road networks;

[0016] (2) This invention improves the Transformer model and optimizes multiple features of the road network to achieve a model that can efficiently and accurately predict multiple road network patterns. This innovative classification method not only improves the speed and efficiency of the classification task, but also improves the refinement of the classification results through the feature optimization algorithm.

[0017] (3) The present invention obtains the final multi-mode hybrid road network classification result through a complete process from road network decomposition, feature extraction to classification and detection; this methodology has broad versatility, is compatible with various types of road network models, and has good portability and scalability, and can be widely used in multiple fields. This comprehensiveness and versatility give the present invention broad social application value and practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0019] Figure 1 This is a flow chart of a multi-mode hybrid road network classification device integrating a Transformer model according to an embodiment of the present invention;

[0020] Figure 2 This is a diagram showing the main road dataset of the present invention;

[0021] Figure 3 A block division display diagram according to an embodiment of the present invention;

[0022] Figure 4 A diagram showing a vector dataset of a block interior according to an embodiment of the present invention;

[0023] Figure 5 This is a split display diagram of a street street mode according to an embodiment of the present invention;

[0024] Figure 6 A road pattern type diagram according to an embodiment of the present invention;

[0025] Figure 7 A road network mode classification display diagram according to an embodiment of the present invention;

[0026] Figure 8 This is a confusion matrix diagram for evaluating the classification accuracy of the transformer model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] Solve the problem of how to decompose and classify the acquired multi-mode hybrid road network vector data, not only solve the problem that the multi-mode hybrid road network data is too numerous and the road network cannot be quickly and efficiently decomposed, but also meet the problem of classifying and identifying different road network modes. Therefore, the present invention proposes a multi-mode hybrid road network decomposition and classification device that integrates the Transformer model (see Appendix). Figure 1 ),include:

[0030] The pre-processed multi-mode mixed road network vector data is decomposed using the geographical partitioning and layering idea;

[0031] Segment the decomposed vector dataset into sections and extract the macro, meso and micro features of each section;

[0032] Based on the spatial structural characteristics of the road network, the pattern classification is carried out;

[0033] Train and use the Transformer model for learning, achieve classification goals and generate a confusion matrix for accuracy evaluation.

[0034] This example uses the Chengdu Geographic Information System Open Street Map (OSM). The projected coordinate system type is then checked and statistical methods and data visualization tools are used to identify and correct outliers in the dataset. These outliers may be caused by measurement errors, data entry errors, and other factors. For missing road sections in the data, automatic detection and repair of hanging points is used to ensure data integrity, resulting in cleaned vector data A'.

[0035] Furthermore, for the pre-processed standard vector dataset A', the main highways, main roads, primary roads and secondary roads in the OSM data classification attributes were separately exported and merged, and then the center lines of the double-line segments were extracted and replaced with the original double-line segments to obtain the main road dataset B (see Appendix). Figure 2 ).

[0036] The line features of the main road dataset A were converted into surface features. The surface features with an area of ​​less than 1 km² were eliminated and merged with the surrounding partitions. The surface features with an area of ​​more than 10 km² were divided into two sub-areas with reference to the third-level roads. Finally, the Chengdu partition dataset C (see Appendix) was obtained. Figure 3 ).

[0037] Select the residential roads, third-level roads, undefined roads and complex interchange roads in the OSM data classification attributes, export and merge them, extract the center line of the double-line segment and replace the original double-line segment. Then use the partition dataset C to split and cut the data to obtain the road network dataset D within each partition (see Appendix). Figure 4 ).

[0038] Based on the principle of spatial similarity and proximity, the road network dataset D is traversed and the road network patterns within the blocks are split one by one (see Appendix Figure 5 ), and finally the Chengdu multi-modal hybrid road network decomposition dataset E is obtained.

[0039] Furthermore, we break the road nodes to split the road network into separate segments. Then, we break the segments where the angle is less than 110 degrees, so that the segments can more clearly reflect the road network characteristics, and obtain the segmented segment dataset F.

[0040] The corresponding third-party toolkit in the programming language Python is used to calculate various characteristic quantities of the road section: the distance between the midpoint of the road section and the center point of the road network, the direction of the line connecting the midpoint of the road section and the center point of the road network, the number of connecting edges between the two end points of the road section, the length of the road section, the aspect ratio of the minimum circumscribed rectangle of the road section, the direction of the longest side of the minimum circumscribed rectangle of the road section, and the proportion of the road section length to the total length of the road network, and these are recorded in the road network attributes.

[0041] Export the calculated road network attribute table and generate the eigenvector matrix G.

[0042] Traverse the dataset F and summarize its spatial characteristics. Based on knowledge and experience, the road network patterns are divided into seven categories: grid pattern, parallel pattern, radial pattern, complex interchange pattern, irregular complex interchange pattern, fishbone pattern and other irregular patterns (see Appendix). Figure 6 ).

[0043] Generate a classification label set H for the dataset F, which is used to identify the classification results of each road network (see Appendix Figure 7 ).

[0044] The feature vector matrix G and the final classification results H are divided into a training set and a test set. The training set is used to train the Transformer model, and the test set is used to evaluate the model's performance. The training set is input into the Transformer model for training, and the model parameters are adjusted to minimize the classification error. During training, cross-validation is used to evaluate the model's stability and generalization ability.

[0045] Calculate the model's classification accuracy and recall rate to evaluate the model's performance, and use the confusion matrix to visually display the model's classification accuracy and misclassification (see Appendix). Figure 8 ).

[0046] New data G_new is fed into the trained Transformer model for prediction. The model generates a prediction H_new based on the feature vector of the input data. The prediction H_new includes the classification label and related information of the new data.

[0047] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A multi-mode hybrid road network intelligent decomposition and classification prediction device integrating the Transformer model, characterized by: include: The pre-processed urban road network vector data is decomposed using the geographical zoning and stratification idea; Segment the decomposed vector dataset into sections and extract the macro, meso and micro features of each section; Based on the spatial structural characteristics of the road network, the pattern classification is carried out; Train and use the Transformer model for learning, achieve classification goals and generate a confusion matrix for accuracy evaluation.

2. The urban multimodal mixed road network decomposition method according to claim 1, characterized in that: include: Preprocess the input road network vector data A to identify and clean up outliers and missing values ​​in the data; Standardization processing, including topology checking and coordinate conversion, ensures that data from different sources and formats can be compared and calculated at the same scale to obtain a standardized dataset A'; According to the geographical zoning idea, the main highways, main roads, primary roads and secondary roads in A' are selected as the zoning basis to complete the spatial division of the inner blocks of the city; Select the third-level roads, residential roads, undefined roads, and complex interchange roads in A' as the road network pattern data within the block. Use the main road zoning results to clip the road network within the block. Then convert the two-lane roads to single-lane roads, check the topology structure, and delete single non-patterned roads. Based on the principles of spatial similarity and attribute similarity, the road network within the block is finely decomposed to obtain each road network dataset B.

3. The method for extracting macroscopic, mesoscopic and microscopic features of each road section as claimed in claim 1, characterized in that: include: Based on the different characteristics of the road network, eight types of road segment features were constructed from macroscopic, mesoscopic to microscopic levels. These features include: the distance between the midpoint of the road segment and the center point of the road network, the direction of the line connecting the midpoint of the road segment and the center point of the road network, the number of edges connecting the two end points of the road segment, the length of the road segment, the aspect ratio of the minimum circumscribed rectangle of the road segment, the direction of the longest side of the minimum circumscribed rectangle of the road segment, and the proportion of the road segment length to the total length of the road network. Based on the road network dataset B, the road nodes are interrupted. In order to better reflect the characteristics of the road network, the angled sections (the range set in the present invention is less than 110°) are segmented again to obtain the road segment dataset B'; Based on the road segment dataset B', the feature quantity F of each road segment in the road network is calculated, and the calculated features are converted into a vector matrix F_opt.

4. The method for pattern classification based on the spatial structural characteristics of a road network as claimed in claim 1, characterized in that: include: Based on the pattern characteristics of the road network, it is divided into seven categories: grid pattern, parallel pattern, radial pattern, complex interchange pattern, irregular complex interchange pattern, fishbone pattern and other irregular patterns; The classification results R labels are marked one by one in the road segment dataset B'.

5. The method of training and using a Transformer model to perform learning and training, achieve classification goals and generate a confusion matrix for accuracy evaluation as described in claim 1, characterized in that: include: The feature vector matrix F_opt and the classification result R are used as input data, divided into training set and test set, and input into the Transformer model for training; After model training is completed, the performance of the model is evaluated using the test set. Evaluation indicators include classification accuracy, recall rate, etc. The unlabeled new road network data feature vector F_new is input into the trained Transformer model for classification prediction. The model will output the corresponding classification result R_new based on the feature vector of F_new.

Images