CN115424435B

CN115424435B - Training method of cross link road identification network and method for identifying cross link road

Info

Publication number: CN115424435B
Application number: CN202210956434.3A
Authority: CN
Inventors: 郭翊麟; 杜舒; 唐俊杰
Original assignee: Alibaba China Co Ltd
Current assignee: Alibaba China Co Ltd
Priority date: 2022-08-10
Filing date: 2022-08-10
Publication date: 2024-01-23
Anticipated expiration: 2042-08-10
Also published as: CN115424435A

Abstract

The specification provides a training method of a cross link road recognition network and a method for recognizing a cross link road, and a road data set is obtained, wherein the road data set comprises a training data subset and a test data subset; training a cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; predicting the test data subset according to the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network is inconsistent with the label in the test data subset; and aiming at the error division data, acquiring the update information of the road relation graph, and retraining the cross link road identification network based on the update information.

Description

Training method of cross link road identification network and method for identifying cross link road

Technical Field

One or more embodiments of the present disclosure relate to the field of computer application technologies, and in particular, to a training method for identifying a cross link road and a method for identifying a cross link road.

Background

In navigating a vehicle for a navigation application, the following problems may exist: as shown in fig. 1, the navigation route is a_link (link represents a path) to b_link and then to c_link, but because the length of b_link is shorter, the user cannot immediately respond when the user walks to b_link to hear the guiding voice of "right turn to c_link", so that the user's vehicle may go straight from b_link to c' -link, and thus the user misses the correct navigation route, which causes inconvenience to the user. The triplet (i.e., a group of three consecutive links, such as a_link, b_link, and c_link) with the above-described problem is called a cross-link road. The related art lacks a method capable of accurately identifying the existence of a cross link road.

Disclosure of Invention

In view of this, one or more embodiments of the present disclosure provide a training method for a cross link road identification network and a method for identifying a cross link road.

According to a first aspect of one or more embodiments of the present disclosure, a training method for a cross link road identification network is provided, where the method includes:

acquiring a road data set, wherein the road data set comprises road information of a plurality of triples and labels corresponding to the triples respectively, and the labels are used for indicating whether the triples are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected;

Training a cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the position relation of link in the two triples;

predicting the test data subset according to the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network is inconsistent with the label in the test data subset;

and acquiring updating information of the road relation graph aiming at the error division data, wherein the updating information comprises the following components: node data corresponding to the error data and edge data corresponding to the error data;

retraining the cross link road identification network based on the updated information.

According to a second aspect of one or more embodiments of the present disclosure, there is provided a method of identifying a cross link road, comprising:

obtaining data to be detected, and constructing a road relation diagram to be detected according to the data to be detected;

Identifying the road relation diagram to be detected through a link-crossing road identification network to obtain an identification result of whether nodes in the road relation diagram to be detected are link-crossing road data or not; the cross link road recognition network is trained by the training method of the cross link road recognition network.

According to a third aspect of embodiments of the present disclosure, there is provided a training device for a cross link road recognition network, including:

the road data collection acquisition unit is used for acquiring a road data collection, wherein the road data collection comprises road information of a plurality of triplets and labels corresponding to the triplets respectively, and the labels are used for indicating whether the triplets are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected;

the cross link road recognition network training unit is used for training the cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the position relation of link in the two triples;

The test data subset prediction unit is used for predicting the test data subset according to the cross link road identification network and determining error division data of which the identification result of the cross link road identification network in the test data subset is inconsistent with the label;

an update information obtaining unit, configured to obtain, for the error score data, update information of the road relation graph, where the update information includes: node data corresponding to the error data and edge data corresponding to the error data;

and the cross link road recognition network retraining unit is used for retraining the cross link road recognition network based on the updated information.

According to a fourth aspect of embodiments of the present disclosure, there is provided an apparatus for identifying a cross link road, including:

the road relation diagram construction unit is used for acquiring the data to be detected and constructing a road relation diagram to be detected according to the data to be detected;

the identification result acquisition unit is used for carrying out identification processing on the road relation diagram to be detected through a link-crossing road identification network to obtain an identification result of whether nodes in the road relation diagram to be detected are link-crossing road data or not; the cross link road recognition network is trained by the training method of the cross link road recognition network.

According to a fifth aspect of embodiments of the present specification, there is provided a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the foregoing training method of a cross link road identification network or the method of identifying a cross link road.

According to a sixth aspect of embodiments of the present specification, there is provided a computer device comprising:

a processor;

a memory for storing processor-executable instructions;

the processor executes the executable instructions to realize the training method of the cross link road identification network or the cross link road identification method.

The specification provides a training method of a cross link road recognition network and a method for recognizing a cross link road, which are used for acquiring a road data set, wherein the road data set comprises road information of a plurality of triples and labels respectively corresponding to the triples, and the labels are used for indicating whether the triples are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected; training a cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the position relation of link in the two triples; predicting the test data subset according to the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network is inconsistent with the label in the test data subset; and acquiring updating information of the road relation graph aiming at the error division data, wherein the updating information comprises the following components: node data corresponding to the error data and edge data corresponding to the error data; retraining the cross link road identification network based on the updated information.

By processing the error data (the data which is classified by the cross link road identification network in error), the correct road information of the data which is classified by error in the error data due to inaccurate road information can be determined, and the cross link road identification network is updated through the correct road information, so that the network iteration learns to more accurate knowledge through the self-adaptive feedback mechanism, the influence of the inaccuracy of the existing road data set can be eliminated by the network training, the obtained cross link road identification network is more accurate, the identification result of the cross link is more accurate, the navigation of the cross link road data can be corrected, the yaw loss of a user is reduced, and the use experience of the user is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and together with the description, serve to explain the principles of the specification.

Fig. 1 is a schematic diagram of a cross link problem shown in the present specification.

Fig. 2 is a flow chart illustrating a training method of a cross link road identification network according to an exemplary embodiment of the present disclosure.

Fig. 3 is a road relationship diagram illustrating an exemplary embodiment of the present specification.

Fig. 4 is a flowchart illustrating a method of identifying cross link roads according to an exemplary embodiment of the present disclosure.

Fig. 5 is a block diagram of a training method of a cross link road recognition network according to an embodiment of the present disclosure.

Fig. 6 is a block diagram of a training apparatus of a cross link road identification network according to an exemplary embodiment of the present disclosure.

Fig. 7 is a block diagram of an apparatus for identifying cross link roads according to an exemplary embodiment of the present disclosure.

Fig. 8 is a hardware configuration diagram of a training apparatus of a cross link road recognition network or a computer device where the apparatus for recognizing a cross link road is located according to an exemplary embodiment of the present disclosure.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with one or more embodiments of the present specification. Rather, they are merely examples of apparatus and methods consistent with aspects of one or more embodiments of the present description as detailed in the accompanying claims.

It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than described in this specification. Furthermore, individual steps described in this specification, in other embodiments, may be described as being split into multiple steps; while various steps described in this specification may be combined into a single step in other embodiments.

Navigation applications typically have a vehicle navigation function for assisting a user in driving a vehicle to a shorter/faster/more convenient route to a destination by voice guidance information or other forms of guidance action (such as an on-screen display, etc.) during the user's driving of the vehicle.

In map application, firstly, a road network is divided into a plurality of sections according to a certain rule, each section of road can be regarded as a link, and when in navigation planning, navigation routes are provided for users according to corresponding links according to departure places and destinations of the users, and the users are reminded to walk according to the navigation routes in the running process, so that corresponding losses caused by the fact that the users walk wrong are avoided.

In some cases, cross link road data exists in the navigation route, and the cross link road data may yaw the user. Specifically, as shown in fig. 1, there is a triplet including three consecutive links in the navigation route, a_link, b_link, and c_link, and the triplet is cross-link road data, which will cause the following problems: when the vehicle passes through the a_link, the user is reminded to drive to the right front direction, the user is reminded to turn right when the vehicle passes through the b_link to the c_link, and the user can hear the guiding action (which can be voice or other forms) of turning right to the c_link when the vehicle passes through the b_link due to the short length of the b_link, so that the user can not respond immediately when the user is in chatting or the like, and the vehicle driven by the user can go straight from the b_link to the c' -link, so that the user deviates from the navigation route (namely yaw), and inconvenience is caused to the user.

Therefore, the cross link road data in the map needs to be identified to give the user a better use experience.

The related art has the following methods for identifying link-crossing road data: in the related art, cross link road data is screened out through a certain strategy, for example, the length of a link in the middle of a triplet is smaller than a certain length threshold (for example, 50 meters), and if the minimum included angle between the link in the middle of the triplet and any link in the rest is larger than a certain value, the triplet is determined to be the cross link road data, but the cross link road data which can be screened out in this way is inaccurate, for example, some screened data are not the cross link road data, and some cross link road data can not be identified.

Therefore, a method for accurately identifying the cross link road data is lacking in the related art.

In order to solve the above problem, it is considered that the road data corresponding to the triplets can be identified by the neural network, and further, the unstructured data is the road data corresponding to the triplets, so that the cross link road data can be identified by the graph neural network. Further, the result of identifying the cross link road data by the graph neural network is inaccurate, and the reason thereof is found that in some cases, the actual condition of the road is updated, but the stored data for training is not updated (for example, the road width of a certain road after repair is changed, or the road composition of the road is changed, but the stored data for training is not updated), or the information of the road which is originally collected is inaccurate, which results in that the trained graph neural network is trained based on inaccurate data, and the accuracy of the trained graph neural network is affected.

And if each training data is manually corrected, more time will be wasted. In order to solve the above problem, it is considered that the training data can be updated by an adaptive feedback method, so that the data for training can be closer to the actual situation, and the neural network of the graph obtained by training is more accurate. After the training of the graph neural network is completed, data which are misclassified by the graph neural network (hereinafter referred to as misclassified data, namely data with wrong recognition results of a cross link road in the cross link road recognition process) can be found, and the misclassified data are analyzed to obtain updated information, namely information which is closer to the misclassified data in reality, so that the obtained updated misclassified data are used as training data to update (retrain) the graph neural network, and the graph neural network can learn a correct graph structure.

In other words, the present disclosure provides a training method and apparatus for a cross link road recognition network, to obtain a road data set, where the road data set includes road information of multiple triplets, and a label corresponding to each triplet, where the label is used to indicate whether the triplets are cross link road data; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected; training a cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the position relation of link in the two triples; predicting the test data subset according to the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network is inconsistent with the label in the test data subset; and acquiring updating information of the road relation graph aiming at the error division data, wherein the updating information comprises the following components: node data corresponding to the error data and edge data corresponding to the error data; retraining the cross link road identification network based on the updated information.

Next, a training method of the cross link road recognition network shown in the present specification will be described in detail.

As shown in fig. 2, fig. 2 is a flowchart of a training method of a cross link road recognition network shown in the present specification, including the following steps:

step 201, a road data set is acquired.

The road data set comprises road information of a plurality of triples and labels corresponding to the triples respectively, wherein the labels are used for indicating whether the triples are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet includes three links connected.

And step 203, training a cross link road recognition network through the training data subset.

The input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the link position relation in the two triples.

Next, step 201 and step 203 will be described.

The nouns in step 201 and step 203 will be explained first.

Link, i.e. a section of road, in map navigation application, the real road network is often divided according to the attribute information (such as length, name, number of lanes, road class, road composition, speed limit, direction, city, etc.) of the road, so as to obtain different Link sections. The specific partitioning method can be referred to as a method in the related art, and will not be described herein.

The triplet, namely, the road data combination comprising three links connected in sequence, is used as the identification unit of the cross link problem according to the reason of the cross link problem in the specification. It should be noted that, in this specification, triplets are selected instead of the middle link of the triplets or the last two links in the triplets as a unit for identifying the cross-link problem, because it cannot be determined whether the cross-link problem exists or not by only relying on the two links without other information, and the existence of the cross-link problem is related to all three links of the triplets, and the lack of any information cannot accurately determine whether the cross-link problem exists or not.

The cross link road recognition network in the specification at least comprises a graph neural network. For a general graph neural network, the input is a graph including a plurality of nodes and edges (for example, as shown in fig. 3), in the process of processing the input by the graph neural network, the nodes and edges in the graph are continuously updated (the update is generally obtained by multiplying other nodes and edges around the nodes and weight convolution) in the multi-layer processing, and finally, the node vector in the output graph is obtained. In addition, sometimes, in order to achieve the corresponding purpose, a classification sub-network is further connected behind the graph neural network, for example, in the context of the present specification, the sub-network takes a node vector as input, and takes a result of whether the triplet corresponding to the node is a cross link road as output.

For purposes of this specification, a cross link road identification network may include a neural network (the neural network includes a sub-network for updating node vectors and a classification sub-network, the two sub-networks being trained together as a whole during training). The cross link road identification network may also be a graph neural network including a node vector for updating the road relationship graph, and a classification network (both networks are trained separately). Of course, the cross link road recognition network may have other structures, and the specific structure of the cross link road recognition network is not limited in this specification.

For training of the graph neural network, the graph neural network is trained, that is, a process of constructing a graph through training data and training to obtain weights of the graph neural network through the constructed graph. It should be noted that, in the present description, a corresponding road relationship graph may be constructed according to road data, and supervised learning may be performed according to labels of the training data subset, so that the network may learn the relationship between the graph structures and the road interaction information (attention). The specific training method can be referred to the training method in the related art, and the specific training method for the graph neural network is not limited in this specification.

It should be noted that the above-mentioned Graph neural network may be any common Graph neural network, for example, may be a Graph convolution network (Graph Convolutional Network, GCN), may be a Graph self-encodings (GAT) network, may be a Graph self-encodings (GAE), and is merely an example of a few Graph neural networks, and the specific form of the Graph neural network is not limited in this specification.

The road relation graph, that is, the graph constructed by the road data set or the training data subset, may refer to a method in the related art, and the description is omitted herein. In order to solve the problems in the navigation scene in the specification, in the constructed road relation graph, the nodes represent triples, the feature vectors of the nodes are vectors used for representing road information of the triples, the edges represent road relations between the triples and the triples, and the feature vectors of the edges are used for representing the road relations between the two triples. The specific meaning of the road information and the road relationship will be described in detail below, and will not be described in detail herein.

Further, the reason for the above-described problem in the present specification is that the road data in the road data set does not coincide with the actual situation, in other words, the road data set in the present specification has a problem of inaccuracy, or the road data set in the present specification cannot fully reflect the actual road situation.

It should be noted that, the training data subset is only used as training data in the present iteration, the test data is also used as test data in the present iteration, after the execution of step 209 is completed, the data in the test data subset may be used as training data again to retrain the graph neural network, and the data in the training data subset may be used as test data again. The test data subset and the training data subset in this specification do not represent data for both test data or for both training data forever. Furthermore, the partitioning of training data and test data may be random.

After the description of the nouns in step 201 and step 203, a specific implementation of the above steps will be described next.

For the specific acquisition of the road data, the step 201 may be to randomly select a part of the data from the database storing the road data as the road data, or may be to use the total data of the database as the road data set.

The specific acquisition of the road data may be, in addition to the above method: because for the network, if the data is data which has great difference with the cross link data, such as the directions of three links in the triples are identical, the network can easily distinguish the triples from the triples corresponding to the cross link road. In other words, it is more desirable in this specification for the network to have the ability to screen cross link road triplets and triplets that are very similar to cross link roads. Therefore, in the training process, in order to enable more training data to be data which enables the model to learn more graph structures, further reduce training cost and avoid noise influence, and achieve the optimal training effect with the least data, the acquired road data can be data comprising a cross link road triplet and a triplet which is very similar to the cross link road.

Based on the above analysis, in step 201, the road data may be screened according to the strategy of screening the link-crossing road in the related art, and the screened road data may be used as training data.

In other words, step 201 includes: obtaining three links connected with each other; and under the condition that the three links meet preset road conditions, taking the three links as the triples in the road data set, wherein the preset road conditions comprise: the road length of the middle link in the three links is smaller than a length threshold (for example, may be 50 m), and the included angle between the middle link and any other link in the three links is larger than a preset included angle threshold (for example, may be 30 degrees).

It should be noted that, the three links connected are three links connected in sequence, and the included angle refers to the minimum included angle between two links. In addition, in addition to the screening according to the above conditions, road data meeting the requirements may be screened further according to the following manner, for example, road data may be screened according to information on road attributes such as road type, road class, road composition, and the like.

The road types may include: unidirectional lanes, bidirectional lanes, etc.; the road class may include: national roads, provinces, urban express, major/minor roads, etc.; the road construction may include: upper and lower line separation, rotary island, service area, main road, auxiliary road, etc. The need to filter road data by information about road attributes is due to the lack of cross link problems under certain attributes, such as the lack of cross link problems for the small roads and service areas.

In addition, after the road data set is acquired, if the road data set is acquired through a screening mode, in order to enrich the road data, a cross link road recognition network can learn richer knowledge in the training process, and after the road data set is acquired, triples of front links and rear links (namely triples formed by links connected with any link in the triples) of the road data can be acquired. In this way, in the construction of the road relation graph, the triples and the triples connected with the triples have a connection relation, and then the information of the connected triples is considered in updating the screened triples, so that the cross link road recognition network can learn a richer graph structure in the process of updating the road relation graph corresponding to the training data.

In other words, after step 201, the method further comprises: acquiring adjacent triples with positions adjacent to the link according to the link positions in the triples in the road data set; the adjacent triples are added to the road data set.

Finally, it should be further noted that, in order to make the graph neural network learn a richer graph structure, the road relationship graph can be constructed by using more than the training data subset, and can be constructed by using the full amount of road data, so that the relationship in the constructed graph is more comprehensive and rich, and the network learning result can be more accurate.

In other words, step 203 includes: and constructing a road relation diagram according to the training data subset and the test data subset, training the labels of the triples in the data subset according to the road relation diagram, and training to obtain the cross link road identification network. Of course, it is also possible to construct the graph neural network from only a subset of training data.

The training obtains a cross link road identification network by acquiring road data comprising a training data subset and a test data subset, and performing supervised learning through the training data subset until the network converges. But the recognition effect of the network is not good, and the recognition effect of the network needs to be further improved through steps 205-209.

And 205, predicting the test data subset according to the cross link road identification network, and determining wrong-distribution data of which the identification result of the cross link road identification network and the label are inconsistent in the test data subset.

Step 207, obtaining update information of the road relation graph according to the misclassification data.

Wherein the update information includes: node data corresponding to the error data, and edge data corresponding to the error data.

Step 205 and step 207 are feedback processes of closed loop feedback (adaptive feedback). Next, step 205 and step 207 will be collectively described.

The nouns of step 205 and step 207 will be first described.

The wrong data is the data of the wrong identification of the cross link road identification network in the test data subset, and the test data is also provided with the label, so that the wrong data can be found out by comparing the label of the test data in the test data subset with the identification result of the cross link road identification network.

Because of the misclassification error of the wrong score data, the cross link road recognition network may not learn the correct node-node relationship, or may learn the relationship between the edge and the node by mistake, so in order to make the cross link road recognition network have a better recognition effect, the update in step 205 includes node data and edge data. That is, the update information is used to update the feature vectors of the nodes and edges in the road relation graph, and in some cases, the structure of the graph may be changed due to the feature vectors of the edges.

And predicting the test data subset according to the cross link road recognition network, namely updating node feature vectors of the road relation graph corresponding to the test data through the cross link road recognition network, and obtaining a recognition result of whether each node is a cross link road or not, wherein the misclassification data is data with inconsistent recognition results and test data labels.

Having described the nouns involved in step 205 and step 207, a specific implementation of step 205 and step 207 will be described below.

For step 205, the data used for prediction may be all the data in the test data subset, and in the case that there are more data in the test data subset, a better test effect may be achieved by using only part of the test data, so that in order to improve the overall efficiency, the data used for testing may also be part of the data in the test data subset.

In the latter case, step 205 comprises: selecting prediction data from the subset of test data; and predicting the prediction data through the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network and the label are inconsistent in the prediction data.

For the specific method of selecting the predicted data described above, a part of the data may be randomly selected from the test data subset as the predicted data.

In addition, it is considered that if the predicted data is randomly screened, deviation may occur in the predicted result, for example, the network only has a better recognition effect on part of road data, and the recognition effect on other road data is not good. For example, when the road relationship includes multiple relationships (the case that the road relationship includes multiple relationships will be described in detail below and will not be repeated here), the network may only have a good prediction effect on the data having a part of the road relationship, so in order to accurately determine the prediction effect of the network, it should be ensured that the data is unbiased in terms of the road relationship, that is, the predicted data should be screened out from each relationship, so as to determine whether the network has a good prediction effect on all the road relationships.

In other words, where the edges of the road relationship graph are used to characterize multiple road relationships between two triples, the selecting of predictive data from the subset of test data includes: and extracting part of road data from the road data corresponding to each road relation aiming at the test data subset, and forming the extracted road data into prediction data.

In addition, in order to determine that the network has a good prediction effect on all triplets of traffic, it should also be ensured that the prediction data is unbiased in traffic layering, i.e. each traffic has corresponding prediction data.

In other words, the selecting of the predicted data from the subset of test data includes: counting the traffic flow of links included in each triplet aiming at the test data subset, and determining the traffic flow range of the average value of the traffic flows of the three links included in each triplet; and extracting part of road data from the road data corresponding to each traffic flow range, and forming the extracted road data into prediction data.

The traffic flow can be counted based on daily flow, or can be counted based on monthly flow, quaternary flow or annual flow. Of course, statistics may be performed based on daily traffic, monthly traffic, quaternary traffic, and annual traffic, and then partial data may be extracted from all traffic ranges corresponding to each statistical method. For example, partial data may be extracted from a plurality of traffic flow ranges corresponding to annual traffic flow, partial data may be extracted from a plurality of traffic flow ranges corresponding to quaternary traffic flow, and so on.

For the execution timing of step 207, the update of the error score data may be performed as long as there is error score data. In addition, in order to improve the network training efficiency, the update can be performed only when the proportion of the error data to the prediction data (the data used for prediction can be all data in the test data subset or part of data in the test data subset) is greater than a certain value, so that the task network training can reach usable precision under the condition of less error data, and the network training efficiency can be improved by reducing the update.

In other words, step 207 comprises: determining the error division ratio of the error division data to the predicted data; the prediction data is part or all of a subset of test data; and under the condition that the error rate is larger than a preset rate threshold value, acquiring the update information of the road relation graph.

In the above case, considering the different traffic flows, if the number of wrong-split data is the same, for data with a larger traffic flow range, the wrong-split data will lead to more vehicle yaw, and the present application aims to reduce the number of vehicles that yaw as much as possible. Therefore, under the condition that whether the misclassification proportion is larger than the proportion threshold value is judged, the misclassification proportion can be divided into a plurality of groups of data according to the vehicle flow range, the larger the vehicle flow is, the lower the misclassification proportion is, so that the network has better cross link road recognition effect on the group with the larger vehicle flow, and therefore, the yaw of most vehicles is prevented.

In other words, the method further comprises: and counting the traffic flow of links included in each triplet aiming at the test data subset, and determining the traffic flow range of the average value of the traffic flows of the three links included in each triplet. Determining the error division ratio of the error division data to the predicted data; under the condition that the error rate example is larger than a preset rate threshold value, acquiring the update information of the road relation graph comprises the following steps: for each traffic flow range, performing: determining the error division ratio of error division data corresponding to the traffic flow range to predicted data; acquiring updated information of the road relation graph under the condition that the error rate is larger than a ratio threshold corresponding to the traffic flow range; the larger the average value of the traffic flow of the road data corresponding to the traffic flow range is, the lower the proportion threshold corresponding to the traffic flow range is.

After detailing the execution timing of step 207, a specific method of updating the misclassified data will be described next.

Updating the wrong score data may be by manually analyzing the wrong score data to determine correct edge data and node data (the node data is used for representing road information of the triples corresponding to the nodes, and the edge data is used for representing road relations between the triples).

In addition to identification by a method of manual analysis, it is considered that if there are a plurality of road information or a plurality of road relationships with respect to the road information or the road relationship, any one of the road information or the road relationship changes is related to the remaining road information or road relationship. For example, if a road changes from a unidirectional road to a bidirectional road, the road may be a bidirectional road due to the widening of the road, so that in the case that part of the road information or the road relationship is accurate and part of the information is inaccurate, the interaction between different road information or different road relationships can be learned through the relationship network, and whether each road information or road relationship is correct can be determined through the relationship network. In other words, step 207 may also be implemented through a relational network.

In other words, step 207 comprises: acquiring updated information of the road relation graph through a relation network; the input of the relation network is the characteristic vector of the node or the edge in the road relation diagram, and the output is updated node data or edge data.

The relationship network will be further described. The output of the relationship network may be: whether each bit feature value in the feature vector of the node or edge is accurate. In the case that a plurality of road relations exist in the road relation diagram or a plurality of road information exists in the nodes, the feature value of the feature vector represents one road relation or one road information, and then the output of the relation network can be understood as follows: whether each road relation or each road information is accurate.

For the road relation, because the edge often represents whether the relation exists between the two, the value of each characteristic value in the characteristic vector of the edge is generally non-0, namely 1, so that whether the characteristic value is accurate or not can be output through the relation network, and what is the accurate characteristic vector of the edge can be determined through the recognition result of the relation network.

In addition, for a node, the road information of the node is attribute information of the node, and each attribute information generally includes several classifications, for example, road information such as a road type, which generally includes a unidirectional road and a bidirectional road, so that the feature vector of the node can also be updated through a relational network. Specifically, still using the above example, the output of the relationship network may be whether the road type of the node is a unidirectional road, whether the road type of the node is a bidirectional road, and so on. The visible relational network can update feature vectors of nodes and edges well.

It should be noted that, of course, whether the reason of misclassification data is caused by inaccurate feature vectors of nodes or edges may be determined through the relational network, if so, the feature vectors of the nodes or edges may not be updated through the relational network, and may be updated through a manual analysis method, so that a better updating effect may be achieved.

In addition, in order to enable the relation network to better update the error data, the relation network can be updated (retrained) through the identification result of the error data before the feature vector of the node or the edge is updated through the relation network each time, so that the relation network can be more suitable for the current scene.

In other words, the obtaining, through the relational network, update information of the road relational graph includes: selecting part of data from the error data, and acquiring a relationship tag of the selected part of data, wherein the relationship tag is used for representing accurate edge or node data corresponding to the error data; the selected partial data are used as relation training data, and a pre-trained relation network is updated through the relation training data; and acquiring updated information corresponding to the residual data through the updated relation network aiming at the residual data except the relation training data in the error division data.

The accurate edge or node data may be represented by the result of whether each feature value is accurate or not, or may be represented by whether the feature value is a or B (i.e. whether the road type corresponding to the node is a unidirectional road and the road type of the node is a bidirectional road or not).

Step 209, retraining the cross link road identification network based on the updated information.

Specifically, the updated error data is added into the training data subset (meanwhile, the part of data is removed from the test data subset), the constructed road relation diagram is updated based on the updated error data, and the cross link road recognition network is retrained based on the new road relation diagram, so that the network achieves a good cross link recognition effect.

In addition, it should be noted that, after steps 201 to 209 are performed, in order to make the network have a better recognition effect, steps 201 to 209 may be repeatedly performed, so that the network may update the model through a continuous feedback process, learn richer information, and further improve the recognition effect of the model.

After the above method is described in its entirety, a description will be given of a method that can improve accuracy of network identification.

For the identification of the cross link road through the graph neural network, the road relationship may be one road relationship or multiple road relationships, and similarly, the road information may include one road information or multiple road information.

If multiple road information and multiple road relations are adopted, a graph with more complex relations can be constructed, and as the neural network learns the combination relation between different data in practice, the more the types of data in training data are, the more the contents of edges and nodes are, the better the identification effect of the cross link road identification network after training is finished, the more abundant road relation graphs can be generated by the road relation graphs with multiple road relations, and the graph neural network can learn the more abundant structural information of the graph through the road relation graphs, so that the trained graph neural network has better identification effect on the cross link road.

In other words, the road relationship in the present specification may include various relationships, and of course, the road information may include various information at the same time. The road relation graph comprises a plurality of road relation subgraphs, each road relation subgraph comprises a plurality of nodes and edges for connecting the nodes, and the edges for connecting the nodes in different road relation subgraphs represent different road relations. The road relation diagram with the above relation is shown in fig. 3.

It should be noted that, the boxes in fig. 3 represent a sub-graph corresponding to a relationship, the connection between the nodes of different boxes represents that two nodes are the same node, and the connection between the nodes in fig. three is a full connection relationship, i.e. if the node a is connected to the node B, and if the node B is connected to the node C, the connection between the node C and the node a is not shown in the diagram, but the two nodes are actually connected. The above-described road relation subgraph, i.e., the region characterized by the box in fig. 3.

Other relationships in fig. 3, that is, a combination of two or more of the above relationships, such as the road type between two triples being identical and the road class being identical, may be considered to be one other relationship between two triples. The other relationship may be preset, or may be obtained by the foregoing update information acquisition. For example, by manual analysis, it is found that the combination of two or more road relationships is more relevant to the cross link road, and other relationships can be added to the road relationship graph.

It should be noted that fig. 3 is not a diagram actually used by the computer, but is a diagram drawn for the convenience of the user to read and view the training results. Meanwhile, although the same node in fig. 3 is represented by a plurality of nodes in the graph, in the training process of the actual graph neural network, the plurality of nodes are identical in feature vector, and the change is synchronous.

Next, first, description will be made of road relation and road information, which may be information for characterizing basic attributes of a road, such as upstream-downstream relation, road length, road width, number of lanes, road departure, road entrance, road direction, road angle, and the like, which may include at least one of the following road class, road type, road composition, city to which the road belongs, and other triples.

The road class, the road type, and the road composition are already described above and will not be described in detail here. The city to which the road belongs, i.e. the city in which the position of the road is located. The road length represents the length of each link in the triplet, the road width represents the width of each road in the triplet, and the road exit and the road entrance represent the number of roads which can pass through at the downstream of one road and the number of roads which can enter the current road at the upstream of one road. The road direction and the road angle are similar, i.e. the azimuth information characterizing the road. If the information is a numerical value, the numerical value may be directly used as a feature value in the feature vector of the node, and if the information is not a numerical value (such as a category feature), the information needs to be encoded and then processed.

Similar to the road information, the road relation characterizes whether two triplets have the same specific road information, for example, if all links of the two triplets are bidirectional roads, the road relation with the road category between the two triplets is considered. The road relationship may also characterize the triplet-to-triplet positional relationship. For example, if two triples have the same link between them, then the two triples may be considered to have an upstream-downstream relationship.

The road relationship may include at least one of: whether it is the same road type, whether it is the same road composition, whether it is the same road type, whether it belongs to the same city, or whether it has an upstream-downstream relationship.

It should be noted that, the upstream-downstream relationship not only represents that two triples include the same link, but also that two triples are within the same region, and that one triplet may walk to another triplet, which is also referred to as having an upstream-downstream relationship between the two triples.

It should be noted that, the graph neural network learns the relationships between the combinations and the outputs of the various information, and a certain relationship exists between the combinations of the various information and the cross link road, so that the graph neural network for identifying the cross link road can be obtained through the data training including the various relationships and/or the various road information.

After explaining a training method of the cross link road recognition network, a method of recognizing a cross link road shown in the present specification will be explained next.

As shown in fig. 4, fig. 4 is a flowchart illustrating a method of identifying a cross link road according to an exemplary embodiment of the present disclosure, including:

Step 401, obtaining data to be detected, and constructing a road relation diagram to be detected according to the data to be detected.

The data to be detected may be manually input, and the method for constructing the road relation graph to be detected according to the data to be detected may be similar to the method for constructing the graph in the related art, and will not be described herein. The data to be detected can be all data of the map, and then the cross link road can be marked in advance. The data to be detected can also be that when the navigation route is generated, the combination of three links connected in sequence in the navigation route is regarded as a triplet, and all triplets are regarded as the data to be detected, so that the link crossing road in the navigation route can be identified in a targeted manner, and the guiding action corresponding to the link crossing road can be prompted in a targeted manner.

And step 403, performing recognition processing on the road relation diagram to be detected through a link-crossing road recognition network to obtain a recognition result of whether nodes in the road relation diagram to be detected are link-crossing road data.

The cross link road recognition network is obtained by training the cross link road recognition network training method.

And inputting the constructed road relation diagram to be detected into the network to obtain the identification result of whether each node in the road relation diagram to be detected is the cross link road data or not, and thus the cross link road can be identified.

When the cross link road data exists in the navigation route, the guiding actions of the triples corresponding to the cross link road need to be processed, for example, the two guiding actions of the triples are simultaneously broadcasted or displayed, so that yawing of a user can be avoided, and the use experience of the user is improved.

A training method of the cross link road recognition network shown in the present specification will be described below by way of a specific embodiment.

The overall architecture of the training method of the graph neural network is shown in fig. 5, and the training method comprises a data acquisition module, a relationship construction module, a deep learning model module and a relationship updating module. The data acquisition module is used for acquiring training data and test data; the relationship construction module is used for constructing a road relationship graph comprising a plurality of road relationships. In addition, when the relation update judgment result in the closed loop feedback mechanism is yes, a new relation diagram is constructed and training data is updated; the deep learning model module is used for improving the model identification effect by learning information from different relation diagrams; the relation updating module consists of an expert experience module and a relation generating module, wherein the expert experience module is used for extracting part of test data and performing quality judgment on the part of test data (judging whether the error division ratio is smaller than a ratio threshold value), the relation generating module is used for redetermining a new road relation from the error division data, and finally, the relation updating module is fed back to the relation constructing module based on whether the new relation exists, so that the effect of improving the model identification effect by using the graph relation information with richer model learning is realized.

The specific functions of the above modules will be described in detail.

And a data acquisition module: road data is obtained from a database, cross link data is screened out according to a certain strategy, and the cross link data is expressed as a triplet, wherein the screening strategy is generally generated empirically according to methods such as road types, road grades, road construction and the like, and front link data and rear link data of the batch of data are obtained. And providing the data with manual labels, wherein the labels mean whether link road data are crossed, and the rest data are used as test data for prediction, so that a road data set, a test data subset and a training data subset are obtained. When a batch of data is obtained by using the strategy screening, real cross link road data and non-cross link road data exist in the data, and the real cross link road needs to be identified through the following modules.

And the relation construction module is used for: constructing different road relation subgraphs according to different road relation types through a road data set, wherein points represent a triplet, and the types of edges mainly consist of the following relations: upstream-downstream relationship, same road type, same road composition, same city, etc. Each node also contains basic information of each road link in the triplet, such as information of road length, road width, lane number, road departure, road entrance, road direction, road angle and the like, which are all used for processing in a subsequent deep learning model module, wherein numerical characteristics are directly used, and category characteristics are processed by corresponding codes. In this way, a road relation sub-graph comprising a plurality of road relations can be constructed by the relation construction module, and the plurality of sub-graphs jointly form the road relation graph.

Deep learning model module: after the relationship building module has built the entire relationship graph, some graph models may be utilized, such as: GCN, GAT, GAE, etc., to learn the whole graph structure relationship and road interaction information, and then to conduct supervised learning with respect to the training data until the model converges.

The relation updating module mainly comprises screening test data, an expert experience sub-module and a relation generating sub-module, wherein the expert experience module mainly carries out quality judgment on the test data, and the relation generating module mainly analyzes and generates a new relation according to a quality judgment result and feeds the new relation back to the relation constructing module.

And after model learning in the deep learning model module converges, screening whether the model learning is test data or not to obtain a test data subset in the road data set, and obtaining a classification result of the test data.

The expert experience submodule is used for extracting part of data from the test data subset as prediction data and performing quality judgment on the prediction data, and the extraction method is as follows:

firstly, randomly extracting a part of data from different relations according to a relation construction module, wherein the data is assumed to be data_A;

and secondly, counting traffic flow of each triplet in the data_A, such as daily flow, monthly flow, quaternary flow, annual flow and the like, and randomly extracting a part of data from the traffic flow, wherein the data is assumed to be data_B.

And evaluating the data_B data, and comparing the classification result of the dataB with the label to find out error-separated data. And judging whether the error rate example corresponding to the error rate data is smaller than a preset proportion threshold value, if so, not updating the error rate data, and if so, updating the error rate data through the relation generation sub-module.

The relation generation submodule is used for analyzing the error data to obtain accurate feature vectors of edges and nodes of the error data. Analysis may be performed manually or through a relational network. After analysis is completed, the road relation diagram is required to be updated through the updated information, the part of data is added to training data, and the cross link road recognition network is retrained through the training data, so that the recognition effect of the network can be improved.

Corresponding to the embodiments of the aforementioned method, the present specification also provides embodiments of the apparatus and the terminal to which it is applied.

As shown in fig. 6, fig. 6 is a block diagram of a training apparatus of a cross link road recognition network according to an exemplary embodiment of the present disclosure, the apparatus comprising:

A road data set obtaining unit 610, configured to obtain a road data set, where the road data set includes road information of a plurality of triplets, and a label corresponding to each triplet, where the label is used to indicate whether the triplets are cross link road data; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected;

a network training unit 620, configured to train the cross link road recognition network through the training data subset; the input of the cross link road identification network is a road relation graph, and the output is an identification result of the cross link road; the road relation graph comprises a plurality of nodes and edges for connecting the nodes, each node is used for representing a triplet, and each edge is used for representing the road relation between two triples; the road relation is used for representing whether the same road information exists in the two triples or representing the position relation of link in the two triples;

a test data subset prediction unit 630, configured to predict the test data subset according to the cross link road identification network, and determine error-division data in the test data subset, where the identification result of the cross link road identification network does not match the label;

An update information obtaining unit 640, configured to obtain, for the misclassification data, update information of the road relation graph, where the update information includes: node data corresponding to the error data and edge data corresponding to the error data;

and a network retraining unit 650, configured to retrain the cross link road identification network based on the updated information.

In an alternative embodiment, the road data set acquiring unit 610 is configured to acquire three links connected to each other; and under the condition that the three links meet preset road conditions, taking the three links as the triples in the road data set, wherein the preset road conditions comprise: the road length of the middle link in the three links is smaller than a length threshold, and the included angle between the middle link and any other link in the three links is larger than a preset included angle threshold.

In an optional embodiment, the system further includes a road data set adding unit 660 (not shown in the figure), configured to obtain, according to a link position in the triples in the road data set, an adjacent triplet having a position adjacent to the link position; the adjacent triples are added to the road data set.

In an alternative embodiment, the road relation graph comprises a plurality of road relation subgraphs, each road relation subgraph comprises a plurality of nodes and edges connecting the nodes, and the edges connecting the nodes in different road relation subgraphs represent different road relations.

In an alternative embodiment, the road relation includes at least one of: whether the road type is the same road type, whether the road is formed by the same road, whether the road type is the same road type, whether the road type belongs to the same city or whether the road type has an upstream-downstream relationship; and/or; the road information includes at least one of: road class, road type, road composition, city of the road, and upstream-downstream relationship of other triplets, road length, road width, number of lanes, road exit, road entrance, road direction, road angle.

In an alternative embodiment, test data subset prediction unit 630 includes: a prediction data screening subunit 631 (not shown) for selecting prediction data from the subset of test data; a prediction data prediction subunit 632 (not shown in the figure) for: and predicting the prediction data through the cross link road identification network, and determining wrong-partition data of which the identification result of the cross link road identification network and the label are inconsistent in the prediction data.

In an alternative embodiment, the edges of the road relationship graph are used to characterize various road relationships between two triples; and a prediction data screening subunit 631 configured to extract, for the subset of test data, part of road data from the road data corresponding to each road relationship, and combine the extracted road data into prediction data.

In an optional embodiment, the predicted data filtering subunit 631 is configured to, for the test data subset, count traffic of links included in each triplet, and determine a traffic range to which a traffic average value of three links included in each triplet belongs; and extracting part of road data from the road data corresponding to each traffic flow range, and forming the extracted road data into prediction data.

In an alternative embodiment, the update information obtaining unit 640 is configured to determine a misclassification ratio of the misclassification data to the prediction data; the prediction data is part or all of a subset of test data; and under the condition that the error rate is larger than a preset rate threshold value, acquiring the update information of the road relation graph.

In an optional embodiment, the system further includes a traffic flow statistics module 670 (not shown in the figure) configured to, for the subset of test data, count traffic flows of links included in each triplet, and determine a traffic flow range to which a traffic flow average value of three links included in each triplet belongs; an update information acquisition unit 640 for performing, for each traffic flow range: determining the error division ratio of error division data corresponding to the traffic flow range to predicted data; acquiring updated information of the road relation graph under the condition that the error rate is larger than a ratio threshold corresponding to the traffic flow range; the larger the average value of the traffic flow of the road data corresponding to the traffic flow range is, the lower the proportion threshold corresponding to the traffic flow range is.

In an alternative embodiment, the update information obtaining unit 640 is configured to obtain, through a relationship network, update information of the road relationship graph; the input of the relation network is the characteristic vector of the node or the edge in the road relation diagram, and the output is updated node data or edge data.

In an alternative embodiment, the update information obtaining unit 640 is configured to update the pre-trained relationship network with the selected partial data as relationship training data; and acquiring updated information corresponding to the residual data through the updated relation network aiming at the residual data except the relation training data in the error division data.

As shown in fig. 7, fig. 7 is a block diagram of an apparatus for identifying a cross link road according to an exemplary embodiment, the apparatus comprising:

the road relation diagram to be detected construction unit 710 is configured to obtain data to be detected, and construct a road relation diagram to be detected according to the data to be detected;

the identification result obtaining unit 720 is configured to perform identification processing on the to-be-detected road relationship graph through a cross link road identification network, so as to obtain an identification result of whether a node in the to-be-detected road relationship graph is cross link road data; the cross link road recognition network is obtained by training the cross link road recognition network training method.

The implementation process of the functions and roles of each module in the above device is specifically shown in the implementation process of the corresponding steps in the above method, and will not be described herein again.

For the device embodiments, reference is made to the description of the method embodiments for the relevant points, since they essentially correspond to the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the modules illustrated as separate components may or may not be physically separate, and the components shown as modules may or may not be physical, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purposes of the present description. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

As shown in fig. 8, fig. 8 is a hardware structure diagram of a computer device where a training apparatus of a cross link road recognition network or an apparatus for recognizing a cross link road according to an embodiment is located, where the device may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 implement communication connections therebetween within the device via a bus 1050.

The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit ), microprocessor, application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc. for executing relevant programs to implement the technical solutions provided in the embodiments of the present disclosure.

The Memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory ), static storage device, dynamic storage device, or the like. Memory 1020 may store an operating system and other application programs, and when the embodiments of the present specification are implemented in software or firmware, the associated program code is stored in memory 1020 and executed by processor 1010.

The input/output interface 1030 is used to connect with an input/output module for inputting and outputting information. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.

Communication interface 1040 is used to connect communication modules (not shown) to enable communication interactions of the present device with other devices. The communication module may implement communication through a wired manner (such as USB, network cable, etc.), or may implement communication through a wireless manner (such as mobile network, WIFI, bluetooth, etc.).

Bus 1050 includes a path for transferring information between components of the device (e.g., processor 1010, memory 1020, input/output interface 1030, and communication interface 1040).

It should be noted that although the above-described device only shows processor 1010, memory 1020, input/output interface 1030, communication interface 1040, and bus 1050, in an implementation, the device may include other components necessary to achieve proper operation. Furthermore, it will be understood by those skilled in the art that the above-described apparatus may include only the components necessary to implement the embodiments of the present description, and not all the components shown in the drawings.

The embodiments of the present disclosure also provide a computer readable storage medium having a computer program stored thereon, which when executed by a processor implements the foregoing training method or identifying a cross link road identification network.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

The specification also provides a computer program for implementing the training method of the cross link road recognition network or the method for recognizing the cross link road.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.

The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A training method for a cross link road identification network, the method comprising:

acquiring a road data set, wherein the road data set comprises road information of a plurality of triples and labels corresponding to the triples respectively, and the labels are used for indicating whether the triples are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected, the road length of the middle link in the three links is smaller than a length threshold, and the included angle between the middle link and any other link in the three links is larger than a preset included angle threshold;

2. The method of claim 1, the road relationship graph comprising a plurality of road relationship subgraphs, each road relationship subgraph comprising a plurality of nodes and edges connecting nodes, the edges connecting nodes in different road relationship subgraphs representing different road relationships.

3. The method of claim 2, wherein predicting the subset of test data according to the cross link road identification network, determining the wrong-split data of the subset of test data for which the identification result of the cross link road identification network does not match the label, comprises:

extracting part of road data from the road data corresponding to each road relation aiming at the test data subset, and forming the extracted road data into prediction data;

and predicting the prediction data according to the cross link road recognition network, and determining wrong-partition data of which the recognition result of the cross link road recognition network is inconsistent with the label in the prediction data.

4. The method according to claim 1,

the road relation comprises at least one of the following: whether the road types are the same, whether the road types belong to the same city or whether the road types have an upstream-downstream relationship; and/or;

the road information includes at least one of: road class, road type, road composition, city of the road, and upstream-downstream relationship of other triplets, road length, road width, number of lanes, road exit, road entrance, road direction, road angle.

5. The method of claim 1, wherein the obtaining, for the misclassification data, update information of the road relation graph includes:

determining the error division ratio of the error division data to the predicted data; the prediction data is part or all of a subset of test data;

and under the condition that the error rate is larger than a preset rate threshold value, acquiring the update information of the road relation graph.

6. The method according to claim 1,

the method further comprises the steps of: counting the traffic flow of links included in each triplet aiming at the test data subset, and determining the traffic flow range of the average value of the traffic flows of the three links included in each triplet;

The obtaining the update information of the road relation graph for the misclassification data includes:

for each traffic flow range, performing:

determining the error division ratio of error division data corresponding to the traffic flow range to predicted data;

acquiring updated information of the road relation graph under the condition that the error rate is larger than a ratio threshold corresponding to the traffic flow range;

the larger the average value of the traffic flow of the road data corresponding to the traffic flow range is, the lower the proportion threshold corresponding to the traffic flow range is.

7. The method of claim 1, the obtaining updated information of the road relationship graph, comprising:

acquiring updated information of the road relation graph through a relation network; the input of the relation network is the characteristic vector of the node or the edge in the road relation diagram, and the output is updated node data or edge data.

8. The method of claim 7, wherein the obtaining, through a relationship network, updated information of the road relationship graph comprises:

selecting partial data from the error data, and acquiring a relationship tag of the selected partial data, wherein the relationship tag is used for representing accurate edge or node data;

The selected partial data are used as relation training data, and a pre-trained relation network is updated through the relation training data;

and acquiring updated information corresponding to the residual data through the updated relation network aiming at the residual data except the relation training data in the error division data.

9. A method of identifying a cross link, comprising:

identifying the road relation diagram to be detected through a link-crossing road identification network to obtain an identification result of whether nodes in the road relation diagram to be detected are link-crossing road data or not; wherein the cross link road identification network is trained by the method of any one of claims 1 to 8.

10. A training device for a cross link road identification network, comprising:

the road data collection acquisition unit is used for acquiring a road data collection, wherein the road data collection comprises road information of a plurality of triplets and labels corresponding to the triplets respectively, and the labels are used for indicating whether the triplets are cross link road data or not; the road data set comprises a training data subset and a test data subset; the triplet comprises three links which are connected; the road length of the middle link in the three links is smaller than a length threshold, and the included angle between the middle link and any other link in the three links is larger than a preset included angle threshold;

11. An apparatus for identifying a cross link, comprising:

the identification result acquisition unit is used for carrying out identification processing on the road relation diagram to be detected through a link-crossing road identification network to obtain an identification result of whether nodes in the road relation diagram to be detected are link-crossing road data or not; wherein the cross link road identification network is trained by the method of any one of claims 1 to 8.

12. A computer device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to implement the method of any of claims 1-9 by executing the executable instructions.

13. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the method of any of claims 1-9.