Urban road intelligent line selection method based on deep learning model

Abstract

The invention discloses an urban road intelligent line selection method based on a deep learning model, and belongs to the technical field of intelligent road line selection. In order to solve the problem of automatic road line selection reasoning, the method comprises the following steps: dividing an urban space according to a fixed interval to form three-dimensional pixel points, obtaining an urban space three-dimensional pixel data set, and inputting urban area elements according to the three-dimensional pixel points; setting and updating the traffic travel attributes of the three-dimensional pixel points in the urban space three-dimensional element data set by combining the spatial attributes and features of the urban area elements to obtain an updated urban space three-dimensional element data set; collecting urban data of a built road as training data, marking road line data, and then integrating the marked road line data with the updated urban space three-dimensional element data set to obtain an urban road intelligent line selection data set; and building an urban road intelligent line selection deep learning model based on Transform for training to obtain a trained urban road intelligent line selection deep learning model. Automatic road line selection is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent highway route selection, and particularly relates to an intelligent urban highway route selection method based on a deep learning model. Background Art

[0002] Existing highway intelligent route selection technologies mainly carry out route selection based on terrain, and the application scenarios are limited to mountainous areas, unable to consider more complex other background environments such as economy, population, infrastructure, etc. This also makes them inapplicable to urban roads. Existing highway intelligent route selection technologies are still in a semi-automatic mode, requiring manual setting of initial boundary parameters, and then the system automatically builds route selection models based on the parameters. Currently, there is no technical method that allows a large model to automatically read spatial information and automatically perform highway route selection reasoning. Summary of the Invention

[0003] The problem to be solved by the present invention is to realize automatic highway route selection reasoning, and a method for intelligent urban highway route selection based on a deep learning model is proposed.

[0004] To achieve the above object, the present invention is realized through the following technical solutions: An intelligent urban highway route selection method based on a deep learning model, comprising the following steps: S1. Divide the urban space at a fixed interval to form three-dimensional pixel points, and obtain a three-dimensional pixel dataset of the urban space; S2. Input urban elements into the three-dimensional pixel dataset of the urban space obtained in step S1 according to the three-dimensional pixel points, including natural elements R , material elements M , economic elements E , to obtain a three-dimensional element dataset of the urban space; S3. Combine the spatial attributes and characteristics of the urban elements to set and update the traffic travel attributes of the three-dimensional pixel points in the three-dimensional element dataset of the urban space, and obtain an updated three-dimensional element dataset of the urban space; S4. Collect urban data of existing highways as training data, mark the highway line data in the training data, and then integrate it with the updated three-dimensional element dataset of the urban space obtained in step S3 to obtain an intelligent urban highway route selection dataset; S5. Build a deep learning model for intelligent urban highway route selection based on Transformer; S6. Use the intelligent urban highway route selection dataset obtained in step S4 to train the deep learning model for intelligent urban highway route selection obtained in step S5 to obtain a trained deep learning model for intelligent urban highway route selection; S7. Collect the three-dimensional urban spatial element dataset of the to-be-built road and input it into the trained deep learning model for intelligent route selection of urban roads obtained in step S6 for prediction, so as to obtain the intelligent route selection scheme of the urban road of the to-be-built road.

[0005] Further, in step S1, the urban space is segmented into three-dimensional grids, each grid represents a three-dimensional pixel point, and the granularity of each three-dimensional pixel point is set to 1x1x1m.

[0006] Further, in step S2, the natural elements R include terrain and geology, greening river channels, vegetation systems, and landscape resources, and the physical elements M include built environment, transportation facilities, municipal public facilities, public spaces and greenlands, industries and production facilities, disaster prevention and safety facilities, and cultural and leisure facilities, and the social elements S include population, culture, system, and economy.

[0007] Further, in step S3, the range of traffic travel attributes is set to 0-10, where 0 represents the lowest traffic generation and attraction, and 10 represents the highest traffic generation and attraction.

[0008] Further, in step S4, the road line data in the training data is labeled, including one or more of the road boundary line, cost, and construction period, and the road route selection attribute of the pixel points within the road boundary line is set to 1.

[0009] Further, the specific implementation method of step S5 includes the following steps: S5.1. Set the deep learning model for intelligent route selection of urban roads based on Transformer, including a disassembly layer, an input layer, a first convolutional layer, a first extended convolutional layer, a second convolutional layer, a second extended convolutional layer, a flattening layer, a first connection layer, a second connection layer, and an output layer connected in sequence; S5.2. Establish a disassembly layer for disassembling the three-dimensional spatial pixel set in the dataset of intelligent route selection of urban roads. The expression formula of the three-dimensional space pixel is as follows: ; Among them, is the dataset of three-dimensional space pixel points, C is the channel data, H is the horizontal coordinate of the pixel point, W is the vertical coordinate of the pixel point, L is the height coordinate of the pixel point, and P is a function that integrates relevant parameters to form a dataset of three-dimensional space pixel points; The disassembly layer disassembles the dataset of three-dimensional space pixel points into multiple layers of two-dimensional data, and the expression is: ; Among them, is a function for storing height coordinates, is a dataset of pixel points in a two-dimensional space; is stored separately in matrix form; Then, the multi-layer two-dimensional data is projected onto the two-dimensional space to form input data, and the expression is: ; Among them, is a single-layer dataset obtained after projecting the multi-layer two-dimensional data onto a single-layer two-dimensional data, is a multi-layer data union function; S5.3. Establish a connection layer , and the expression is: ; Among them, is a linear function, is a non-linear activation function, is an offset and weight combination parameter, h is an input variable; ; Among them, W is the weight, b is the offset; , is the output dimension, is the input dimension, is the real number space; S5.4. Establish a convolutional layer , and the expression is: ; Among them, is a training parameter; The convolutional layer decomposes the two-dimensional disassembled image according to the sliding step, and the two-dimensional convolutional points of the input layer are expressed as: ; Among them, is the input variable, is the convolutional kernel filtering parameter, is the coordinate of the sliding center pixel point, is x the sliding step in the direction, y is p the sliding step in the x direction, q is y the cumulative sliding distance in the direction, is the convolution kernel filtering parameter of the current central pixel point; Read associated z direction pixel points, extract the associated pixel point information from and assign weights to obtain the expression: ; where is the z coordinate associated with the current central pixel point, is the weight, is the bias term; Then, based on optimize the two-dimensional convolution points to obtain the expression of the optimized two-dimensional convolution points: ; S5.5. Establish a dilated convolutional layer , and the expression is: ; where d is the dilated convolution order; The connection method between the input of the dilated convolutional layer and the convolutional kernel is as follows: ; where d is the dilated convolution order, w is the dilated convolutional kernel size, is the weight parameter of the convolutional kernel at the p-th position, is the value of the input x at the position i - d p d p is the sampling offset of the dilated convolution; S5.6. Set the deep learning model for intelligent route selection of urban roads to adopt a feedforward network structure. The expression of the th layer is: ; where is the input layer of the th layer, is the output layer of the th layer, is the training parameter, is the network expression; Truncate the deep learning model for intelligent route selection of urban roads from the th layer to obtain two expressions: ; where is the encoding function; ; where is the decoding function, is the final layer passed by the function, is the entire composite function 's parameter set, is the entire composite function 's parameter set; Then two parts of the entire neural network are obtained: ; ; Among them, is the latent variable of the input ; is 's decoded high-dimensional variable, is the encoding function of the pixel points in the three-dimensional space, is the decoding function that decompiles the low-dimensional variable to generate pixel points in the three-dimensional space.

[0010] Furthermore, in step S6, using the urban road intelligent route selection dataset obtained in step S4, taking the updated urban spatial three-dimensional element data as the input variable and the road route selection boundary as the output variable, the urban road intelligent route selection deep learning model obtained in step S5 is trained.

[0011] Advantages of the present invention: The urban road intelligent route selection method based on a deep learning model described in the present invention takes into account other urban elements besides the terrain, converts the urban road background environment into computer model-readable data, and uses the deep learning model to accurately and intelligently select road routes. Description of the drawings

[0012] Figure 1 is the flowchart of the urban road intelligent route selection method based on a deep learning model described in the present invention; Figure 2 is an example diagram of the urban spatial three-dimensional pixel dataset of the present invention; Figure 3 is an example diagram of the urban spatial input data of the present invention; Figure 4 is an example diagram of the urban road route selection output data of the present invention. Detailed implementation manners

[0013] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only a part of the embodiments of the present invention, rather than all of the specific embodiments. Usually, the components of the specific embodiments of the present invention described and shown in the accompanying drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0014] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0015] In order to further understand the content, features and effects of the present invention, the following specific embodiments are exemplified and accompanied by Figure 1 Appendix Figure 4 The detailed description is as follows:

[0016] Example 1: An intelligent route selection method for urban roads based on a deep learning model, comprising the following steps: S1. Divide the urban space at a fixed interval to form three-dimensional pixel points, and obtain a three-dimensional pixel dataset of the urban space; Further, in step S1, the urban space is divided into three-dimensional grids, each grid represents a three-dimensional pixel point, and the granularity of each three-dimensional pixel point is set to 1x1x1m, as Figure 2 shown.

[0017] S2. Input urban elements into the three-dimensional pixel dataset of the urban space obtained in step S1 according to the three-dimensional pixel points, including natural elements R , material elements M , economic elements E , to obtain a three-dimensional element dataset of the urban space; Further, in step S2, the natural elements R include terrain and geology, greening river channels, vegetation systems, landscape resources, and the material elements M include built environment, transportation facilities, municipal public facilities, public spaces and green spaces, industrial and production facilities, disaster prevention and safety facilities, cultural and recreational facilities, and the social elements S include population, culture, system, economy; S3. Combine the spatial attributes and characteristics of urban elements, set and update the traffic travel attributes of the three-dimensional pixel points in the urban spatial three-dimensional element dataset, and obtain an updated urban spatial three-dimensional element dataset; Further, in step S3, the range of the traffic travel attribute is set to 0-10, where 0 represents the lowest traffic generation and attraction, and 10 represents the highest traffic generation and attraction; Further, the obstacle area within the historical protection area boundary can be regarded as a high-impedance area, that is, the obstacle attribute of the pixel points within the boundary is set to 1; for the population aggregation area, it can be regarded as a priority area with high attraction, that is, the traffic travel attribute of the pixel points within the boundary is set to 10, as Figure 3 shown.

[0018] S4. Collect the urban data of the built roads as training data, mark the road line data in the training data, and then integrate it with the updated urban spatial three-dimensional element dataset obtained in step S3 to obtain an urban road intelligent route selection dataset; Further, in step S4, the road line data is marked in the training data, including marking one or several of the road boundary line, cost, and construction period, and the road route selection attribute of the pixel points within the road boundary line range is set to 1; S5. Build a deep learning model for urban road intelligent route selection based on Transformer; Further, the specific implementation method of step S5 includes the following steps: S5.1. Set the deep learning model for urban road intelligent route selection based on Transformer, including a disassembling layer, an input layer, a first convolutional layer, a first extended convolutional layer, a second convolutional layer, a second extended convolutional layer, a flattening layer, a first connection layer, a second connection layer, and an output layer connected in sequence; S5.2. Establish a disassembling layer for disassembling the three-dimensional spatial pixel set in the urban road intelligent route selection dataset. The expression formula of the three-dimensional space pixel is as follows: ; Among them, is the dataset of three-dimensional space pixel points, C is the channel data, H is the horizontal coordinate of the pixel point, W is the vertical coordinate of the pixel point, L is the height coordinate of the pixel point, and P is a function that integrates relevant parameters to form a three-dimensional space pixel point dataset; The disassembling layer disassembles the dataset of three-dimensional space pixel points into multiple layers of two-dimensional data, and the expression is: ; Among them, is a function for storing height coordinates, is a dataset of pixel points in a two-dimensional space; is stored separately in matrix form; Then, project the multi-layer two-dimensional data onto the two-dimensional space to form the input data, and the expression is: ; where, is the single-layer dataset obtained after projecting the multi-layer two-dimensional data onto the single-layer two-dimensional data, is the multi-layer data union function; S5.3. Establish the connection layer , and the expression is: ; where, is a linear function, is a non-linear activation function, is the offset and weight combination parameter, h is the input variable; ; where, W is the weight, b is the offset; , is the output dimension, is the input dimension, is the real number space; S5.4. Establish the convolutional layer , and the expression is: ; where, is the training parameter; The convolutional layer decomposes the two-dimensional disassembled image according to the sliding step, and the two-dimensional convolutional points of the input layer are expressed as: ; where, is the input variable, is the convolutional kernel filtering parameter, is the coordinate of the sliding center pixel point, is x the sliding step in the direction, y is the sliding step in the p direction, x is the cumulative sliding distance in the q direction, y is the cumulative sliding distance in the direction, is the convolutional kernel filtering parameter of the current center pixel point; Read associated z direction pixel points, extract associated pixel point information from and assign weights to obtain the expression: ; where, is the z - coordinate associated with the current central pixel point, is the weight, is the bias term; Then, based on optimize the two - dimensional convolution points to obtain the expression of the optimized two - dimensional convolution points: ; S5.5. Establish a dilated convolution layer , the expression is: ; where, d is the dilated convolution order; The connection method between the input of the dilated convolution layer and the convolution kernel is as follows: ; where, d is the dilated convolution order, w is the dilated convolution kernel size, is the weight parameter of the convolution kernel at the p - th position, is the value of the input x at the position i - d p , d p is the sampling offset of the atrous convolution; S5.6. Set the deep - learning model for intelligent route selection of urban roads to adopt a feed - forward network structure. The expression of the th layer is: ; where, is the input layer of the th layer, is the output layer of the th layer, is the training parameter, is the network expression; Truncate the deep - learning model for intelligent route selection of urban roads from the th layer to obtain two expressions: ; where, is the encoding function; ; where, is the decoding function, is the final layer of function passing, is the entire composite function 's parameter set, is the entire composite function 's parameter set; Then two parts of the entire neural network are obtained: ; ; Among them, is the input 's latent variable, is 's decoded high-dimensional variable, is the encoding function of three-dimensional space pixel points, is the decoding function that generates three-dimensional space pixel points by decompiling low-dimensional variables.

[0019] S6. Use the urban road intelligent route selection dataset obtained in step S4 to train the urban road intelligent route selection deep learning model obtained in step S5 to obtain a trained urban road intelligent route selection deep learning model; Furthermore, in step S6, using the urban road intelligent route selection dataset obtained in step S4, taking the updated urban spatial three-dimensional element data as the input variable and the road route selection boundary as the output variable, train the urban road intelligent route selection deep learning model obtained in step S5.

[0020] S7. Collect the urban spatial three-dimensional element dataset of the to-be-built road, input it into the trained urban road intelligent route selection deep learning model obtained in step S6 for prediction, and obtain the urban road intelligent route selection plan of the to-be-built road.

[0021] It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0022] Although the present application has been described above with reference to specific embodiments, various modifications can be made thereto and components thereof can be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in the present application can be combined with each other in any way, and the reason for not exhaustively describing the situations of these combinations in this specification is only to save space and resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for intelligent route selection of urban highways based on a deep learning model, characterized in that: The steps include: S1. Segment the urban space into three-dimensional pixels at fixed intervals to obtain a three-dimensional pixel data set of the urban space; S2. Input urban elements, including natural elements, according to three-dimensional pixel points in the urban space three-dimensional pixel data set obtained in step S1 R , Material elements M , Economic factors E , and obtain the urban space three-dimensional element dataset; S3. Based on the spatial attributes and characteristics of the urban elements, the traffic attributes of the three-dimensional pixels in the urban space three-dimensional element dataset are set and updated to obtain an updated urban space three-dimensional element dataset; S4. Collecting urban data of built roads as training data, marking the road line data in the training data, and then integrating it with the updated urban space three-dimensional element data set obtained in step S3 to obtain an urban road intelligent line selection data set; S5. Build a Transformer-based deep learning model for intelligent route selection of urban highways; S6. Using the urban highway intelligent route selection data set obtained in step S4 to train the urban highway intelligent route selection deep learning model obtained in step S5, a trained urban highway intelligent route selection deep learning model is obtained; S7. Collect the urban spatial 3D feature dataset of the highway to be built, input it into the trained urban highway intelligent route selection deep learning model obtained in step S6 for prediction, and obtain the urban highway intelligent route selection plan of the highway to be built.

2. According to claim 1, a method for intelligent route selection of urban highways based on a deep learning model is characterized in that: In step S1, the urban space is divided into three-dimensional grids, each grid represents a three-dimensional pixel point, and the granularity of each three-dimensional pixel point is set to 1x1x1m.

3. The method for intelligent route selection of urban highways based on a deep learning model according to claim 1 or 2, characterized in that: Natural elements in step S2 R Including topography and geology, green river channels, vegetation systems, landscape resources, material elements M Including architectural environment, transportation facilities, municipal utilities, public space and green space, industry and production facilities, disaster prevention and safety facilities, cultural and leisure facilities, social elements S Including population, culture, system and economy.

4. The method for intelligent route selection of urban highways based on a deep learning model according to claim 3 is characterized in that: In step S3, the range of the traffic attribute is set to 0-10, where 0 represents the lowest traffic attraction and 10 represents the highest traffic attraction.

5. The method for intelligent route selection of urban highways based on a deep learning model according to claim 4 is characterized in that: In step S4, the highway line data in the training data are annotated, including annotating one or more of the highway boundary line, construction cost, and construction period, and the highway line selection attribute of the pixel points within the highway boundary line range is set to 1.

6. The method for intelligent route selection of urban highways based on a deep learning model according to claim 5, characterized in that: The specific implementation method of step S5 includes the following steps: S5.

1. Setting up a Transformer-based urban highway intelligent route selection deep learning model includes a disassembly layer, an input layer, a first convolution layer, a first extended convolution layer, a second convolution layer, a second extended convolution layer, a flattening layer, a first connection layer, a second connection layer, and an output layer connected in sequence; S5.

2. Establish a disassembly layer to disassemble the three-dimensional spatial pixel set in the urban highway intelligent line selection dataset. The expression formula of the three-dimensional spatial pixel is as follows: ； in, is a dataset of pixels in three-dimensional space. C is the channel data, H is the horizontal coordinate of the pixel point, W is the vertical coordinate of the pixel point, L is the height coordinate of the pixel point, and P is a function that integrates relevant parameters to form a three-dimensional space pixel point data set; The decomposition layer decomposes the three-dimensional space pixel data set into multiple layers of two-dimensional data. The expression is: ； in, is a function that stores the height coordinates, is a dataset of two-dimensional space pixels; Stored separately in matrix form; Then the multi-layer two-dimensional data is projected into the two-dimensional space to form the input data, which is expressed as: ； in, It is a single-layer data set obtained by projecting multi-layer two-dimensional data onto single-layer two-dimensional data. Multi-layer data union function; S5.

3. Establishing the connection layer , the expression is: ； in, is a linear function, is a nonlinear activation function, is the offset and weight combination parameter, h is the input variable; ； in, W is the weight, b is the offset; , is the output dimension, is the input dimension, is the space of real numbers; S5.

4. Building convolutional layers , the expression is: ； in, is the training parameter; The convolution layer decomposes the two-dimensional decomposition image according to the sliding step size, and the two-dimensional convolution point of the input layer is expressed as: ； in, is the input variable, is the convolution kernel filtering parameter, is the coordinate of the sliding center pixel point, yes x The sliding step length in the direction, yes y The sliding step length in the direction, p yes x Cumulative sliding distance in the direction, q yes y Cumulative sliding distance in the direction, is the center pixel after sliding, is the convolution kernel filtering parameter of the current center pixel; Read Related z Direction pixel, from Extract the associated pixel information and assign weights to it, and the expression is: ； in, is the z coordinate associated with the current center pixel, is the weight, is the bias term; Then based on The two-dimensional convolution point is optimized, and the expression of the optimized two-dimensional convolution point is: ； S5.

5. Building dilated convolutional layers , the expression is: ； in, d is the dilated convolution order; The connection method between the input of the dilated convolution layer and the convolution kernel is as follows: ； in, d is the dilated convolution order, w is the dilated convolution kernel size, is the weight parameter of the convolution kernel at the pth position, is the input x at position id p The value of d p is the sampling offset of the dilated convolution; S5.

6. Set the urban highway intelligent route selection deep learning model to adopt a feedforward network structure. The expression of the layer is: ； in, It is The input layer of the layer, It is The output layer of the layer, is the training parameter, is a network expression; The deep learning model of intelligent route selection for urban highways is applied from Layer truncation, we get two expressions: ； in, is the encoding function; ； in, is the decoding function, is the final layer of function passing, is the entire composite function The parameter set, is the entire composite function The parameter set of Then we get two parts of the entire neural network: ； ； in, Yes Input The hidden variable, yes The decoded high-dimensional variable, is the encoding function of the pixels in three-dimensional space, It is a decoding function that decompiles low-dimensional variables to generate three-dimensional space pixels.

7. The method for intelligent route selection of urban highways based on a deep learning model according to claim 6, characterized in that: In step S6, the urban highway intelligent route selection dataset obtained in step S4 is used, the updated urban space three-dimensional element data is used as input variables, and the highway route selection boundary is used as the output variable to train the urban highway intelligent route selection deep learning model obtained in step S5.

Images