A method of road recognition and related apparatus

Abstract

Embodiments of the present application provide a road recognition method and related device, which can be applied to various scenarios such as artificial intelligence, intelligent transportation, assisted driving, Internet of Vehicles, etc. The method can consider traffic flow characteristics and traffic speed characteristics on the basis of remote sensing images, which can reduce the difficulty of road type recognition, and improve the recognition accuracy and recall rate. The method at least involves machine learning technology in artificial intelligence. The above method comprises: obtaining traffic flow characteristics and traffic speed characteristics of a to-be-recognized road area; obtaining a remote sensing image of the to-be-recognized road area; performing feature extraction processing on the remote sensing image to obtain road network characteristics of the to-be-recognized road area; performing road recognition processing on the traffic flow characteristics, the traffic speed characteristics and the road network characteristics based on a target road recognition model to obtain a target classification result, the target classification result being used to indicate a road type of a road in the to-be-recognized road area.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of artificial intelligence, in particular to a road recognition method and related device. BACKGROUND

[0002] Road recognition is an indispensable key step in road network generation and road missing mining. With the development of deep learning, the perception ability of road recognition has made great progress.

[0003] In related solutions, a deep learning model such as a segformer model is mainly used to recognize roads in remote sensing images, and then the type of the road is determined. However, the road structure is complex and is easily blocked by objects such as factories, houses or trees. Using the existing way to recognize and process remote sensing images can only recognize the roads that are not blocked by objects. For the road types in the areas blocked by objects, there is a great difficulty in recognition, resulting in low recognition accuracy and low recall rate. SUMMARY

[0004] Embodiments of the present application provide a road recognition method and related device, which can reduce the difficulty of road type recognition, improve recognition accuracy and recall rate.

[0005] In a first aspect, embodiments of the present application provide a road recognition method. The method comprises: obtaining a traffic flow feature and a traffic speed feature of a to-be-recognized road area, the traffic flow feature being used to indicate the vehicle flow passing through the to-be-recognized road area, and the traffic speed feature being used to indicate the driving speed of the vehicle passing through the to-be-recognized road area; obtaining a remote sensing image of the to-be-recognized road area; performing feature extraction processing on the remote sensing image to obtain a road network feature of the to-be-recognized road area; performing road recognition processing on the traffic flow feature, the traffic speed feature and the road network feature based on a target road recognition model to obtain a target classification result, the target classification result being used to indicate the road type of the road in the to-be-recognized road area.

[0006] In a second aspect, an embodiment of the present application provides a road recognition device. The road recognition device comprises an acquisition unit and a processing unit. The acquisition unit is configured to acquire traffic flow characteristics and traffic speed characteristics of a to-be-recognized road region, the traffic flow characteristics being used to indicate vehicle flow passing through the to-be-recognized road region, and the traffic speed characteristics being used to indicate driving speed of vehicles passing through the to-be-recognized road region. The acquisition unit is configured to acquire a remote sensing image of the to-be-recognized road region. The processing unit is configured to perform feature extraction processing on the remote sensing image to obtain road network characteristics of the to-be-recognized road region. The processing unit is configured to perform road recognition processing on the traffic flow characteristics, the traffic speed characteristics, and the road network characteristics based on a target road recognition model to obtain a target classification result, the target classification result being used to indicate a road type of a road in the to-be-recognized road region.

[0007] In some optional embodiments, the processing unit is configured to: traverse each trajectory point on the to-be-recognized road region, each trajectory point being used to indicate a positioning position of a vehicle passing through the to-be-recognized road region; and perform cumulative summation processing on each trajectory point to obtain the traffic flow characteristics of the to-be-recognized road region.

[0008] In some other optional embodiments, the processing unit is configured to: traverse each trajectory point on the to-be-recognized road region, each trajectory point being used to indicate a positioning position of a vehicle passing through the to-be-recognized road region; and calculate the traffic speed characteristics based on projection speeds of each trajectory point in at least two directions of the to-be-recognized road region.

[0009] In some other optional embodiments, the processing unit is configured to: calculate a projection speed of each trajectory point in a first direction of the to-be-recognized road region and a projection speed of each trajectory point in a second direction of the to-be-recognized road region, the first direction being perpendicular to the second direction, and the first direction and the second direction being any two of the at least two directions; perform average cumulative calculation on the projection speed of each trajectory point in the first direction of the to-be-recognized road region and the projection speed of each trajectory point in the second direction of the to-be-recognized road region, respectively, to obtain a first projection speed and a second projection speed, the first projection speed being used to indicate a passing speed component of the vehicle in the first direction, and the second projection speed being used to indicate a passing speed component of the vehicle in the second direction; and perform normalization processing on the first projection speed and the second projection speed to obtain the traffic speed characteristics.

[0010] In some optional embodiments, the obtaining unit is further configured to obtain a traffic flow feature representation, a traffic speed feature representation, and a road network feature representation of a training sample, the training sample being a sample having a preset road type labeled on a fusion feature obtained after the traffic flow feature representation, the traffic speed feature representation, and the road network feature representation are fused, and the road network feature representation being obtained from a remote sensing image of the training sample. The processing unit is further configured to take the traffic flow feature representation, the traffic speed feature representation, and the road network feature representation as inputs of an initial road recognition model to obtain a target road type, calculate a difference between the target road type and the preset road type to obtain a target loss value, and update model parameters of the initial road recognition model based on the target loss value to obtain the target road recognition model.

[0011] In some optional embodiments, the processing unit is configured to perform fusion processing on the traffic flow feature, the traffic speed feature, and the road network feature to obtain a target fusion feature, and take the target fusion feature as an input of the target road recognition model to obtain the target classification result.

[0012] In some optional embodiments, the target road recognition model comprises a D-linkNet network. The processing unit is configured to input the target fusion feature into the D-linkNet network, perform encoding processing on the target fusion feature by an encoder subnetwork in the D-linkNet network to obtain a first output feature, perform feature extraction processing on the first output feature in a feature extraction subnetwork in the D-linkNet network to obtain a second output feature, the feature extraction subnetwork being composed of a dilated convolution and a convolution block attention module, and perform decoding processing on the first output feature and the second output feature in a decoder subnetwork in the D-linkNet network to obtain the target classification result.

[0013] In some optional embodiments, the processing unit is configured to take the traffic flow feature as a first channel layer, the traffic speed feature as a second channel layer, and the road network feature as a third channel layer, the first channel layer, the second channel layer, and the third channel layer constituting an RGB channel, and perform modeling processing on the first channel layer, the second channel layer, and the third channel layer to obtain a target fusion feature.

[0014] In some optional embodiments, the processing unit is configured to: perform road recognition processing on the remote sensing image based on a preset recognition model to obtain a road segmentation map of the to-be-recognized road region; and use the road segmentation map of the to-be-recognized road region as the road network feature of the to-be-recognized road region.

[0015] In some optional embodiments, the remote sensing image includes a multispectral remote sensing image, a hyperspectral remote sensing image, or a low-spectrum remote sensing image.

[0016] The third aspect of the embodiments of the present application provides a road recognition device, which includes a memory, an input / output (I / O) interface, and a memory. The memory is configured to store program instructions. The processor is configured to execute the program instructions in the memory to perform the method of road recognition corresponding to the embodiments of the first aspect.

[0017] The fourth aspect of the embodiments of the present application provides a computer readable storage medium, which stores instructions. When the instructions are run on a computer, the computer is caused to perform the method corresponding to the embodiments of the first aspect.

[0018] The fifth aspect of the embodiments of the present application provides a computer program product containing instructions, which, when run on a computer or a processor, causes the computer or the processor to perform the method corresponding to the embodiments of the first aspect.

[0019] From the above technical solutions, it can be seen that the embodiments of the present application have the following advantages:

[0020] In the embodiments of the present application, a remote sensing image of a to-be-identified road region is acquired, and feature extraction processing is performed on the remote sensing image to obtain road network features of the to-be-identified road region. Moreover, since the traffic flow feature can indicate the vehicle flow passing through the to-be-identified road region, and the traffic speed feature can indicate the driving speed of the vehicle passing through the to-be-identified road region, the traffic flow feature and the traffic speed feature of the to-be-identified road region can also be acquired. Then, the target road identification model is used to perform road identification processing on the traffic flow feature, the traffic speed feature and the road network feature, and a target classification result is obtained, so as to indicate the road type of the to-be-identified road region through the target classification result. In the above manner, since the traffic flow feature and the traffic speed feature can depict the trajectory data of the vehicle, the influence of other objects on the road during road identification based on the remote sensing image can be eliminated. Therefore, on the basis of the remote sensing image, the traffic flow feature and the traffic speed feature are taken into account, so that the road type of the to-be-identified road region finally identified by the target road identification model is closer to the actual situation, can be applicable to the region scene blocked by objects, and can also be applicable to the region scene not blocked by objects, greatly makes up for the defects caused by depicting the road type only by the remote sensing image, reduces the difficulty of road type identification, and improves the recognition accuracy and recall rate. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 A flowchart of a road identification method provided by the embodiments of the present application is shown;

[0023] Figure 2 A trajectory static flow chart provided by the embodiments of the present application is shown;

[0024] Figure 3 A density diagram after modeling provided by the embodiments of the present application is shown;

[0025] Figure 4 A model training flowchart of a target road identification model provided by the embodiments of the present application is shown;

[0026] Figure 5 A flowchart of generating training samples provided by the embodiments of the present application is shown;

[0027] Figure 6A structural schematic diagram of a target road recognition model provided by an embodiment of the present application is shown.

[0028] Figure 7 A structural schematic diagram of a feature extraction subnetwork in a D-linkNet network is shown.

[0029] Figure 8 A first comparison schematic diagram of a recognition result of the present application and a recognition result of an existing scheme is shown.

[0030] Figure 9 A second comparison schematic diagram of a recognition result of the present application and a recognition result of an existing scheme is shown.

[0031] Figure 10 A structural schematic diagram of a road recognition device provided by an embodiment of the present application is shown.

[0032] Figure 11 A hardware structural schematic diagram of a road recognition device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0033] An embodiment of the present application provides a road recognition method and related device, which can reduce the difficulty of road type recognition, and improve recognition accuracy and recall rate.

[0034] It can be understood that in the specific embodiments of the present application, data related to user information, user's personal data, etc. is involved, and when the above embodiments of the present application are applied to specific products or technologies, the user's permission or consent needs to be obtained, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of countries and regions.

[0035] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0036] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, and above-mentioned drawings, if any, are used to distinguish similar objects and are not necessarily used to describe a particular sequential or chronological order. It is to be understood that the use of the terms so used herein is merely intended to distinguish the objects for reasons of convenience, and that the embodiments described herein can be carried out in other sequences than the one described or illustrated here. Moreover, the terms "comprises", "comprising", "includes", "including" and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a list of steps or units can not necessarily be limited to those listed steps or units but can include other not expressly listed steps or units, or such steps or units inherent to such process, method, product, or apparatus.

[0037] The method for road recognition provided by the embodiments of the present application is realized based on artificial intelligence (AI). Artificial intelligence is a theory, method, technology and application system for using a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive an environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.

[0038] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0039] In the embodiments of the present application, the artificial intelligence technologies mainly involved include the above-mentioned machine learning and other directions. For example, deep learning in machine learning (ML) can be involved, including convolutional neural networks, etc.

[0040] The road recognition method provided in the application can be applied to a road recognition device with data processing capability, such as a terminal device, a server, and the like. The terminal device can include, but is not limited to, a smart phone, a desktop computer, a notebook computer, a tablet computer, a smart speaker, a vehicle-mounted device, a smart watch, and the like. The server can be a stand-alone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and basic cloud computing services such as big data and artificial intelligence platforms, and the like, which are not limited in the application. In addition, the terminal device and the server mentioned above can be directly connected or indirectly connected through wired communication or wireless communication, which is not limited in the application.

[0041] The road recognition device mentioned above can have the ability to implement the computer vision technology mentioned above. The computer vision technology mentioned above is a science that studies how to make machines "see". Further, it is a machine vision that uses cameras and computers to replace human eyes to identify, track and measure targets, and further performs image processing to make computer processing more suitable for image observation or instrument detection by human eyes. As a scientific discipline, computer vision researches related theories and technologies, and attempts to establish an artificial intelligence system that can obtain information from images or multidimensional data. Computer vision technology usually includes image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and the like. It also includes common face recognition, fingerprint recognition and other biometric identification technologies. In the embodiments of the application, the road recognition device can process remote sensing images through the computer vision technology.

[0042] The road recognition device mentioned above can have machine learning capability. Machine learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and the like. It is a subject that studies how a computer simulates or implements human learning behavior to obtain new knowledge or skills, and reorganizes existing knowledge structure to continuously improve its own performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its application is widespread in various fields of artificial intelligence. Machine learning and deep learning usually include neural network technology.

[0043] In the method for road recognition provided in the embodiments of the present application, the artificial intelligence model is mainly related to the application of a neural network, and the road type of a to-be-recognized road region is recognized through the neural network.

[0044] In the mining process of the basic road data of an electronic map, accurately, quickly and stably recognizing the road type and then generating a road network in a corresponding range has always been a popular research direction. Accurately recognizing the road in a track and automatically realizing the segmentation of the road and the background have a key significance for mining the road shape and the road line. In the real world, tracks are complex and diverse, and different levels of roads have different width and shape characteristics, such as national roads, provincial roads, village roads, mountain roads and suburban areas. At the same time, the road region is affected by many factors such as buildings, trees, road central green belts and other objects, so that the accurate extraction of road information is still a research frontier and technical difficulty in the field of track information extraction. At present, in the related scheme, a segformer model or other deep learning model is mainly used to recognize the road of a remote sensing image, and then the type of the road is determined. However, the road structure is complex and is easily blocked by factories, houses or trees and other objects, and the existing way of recognizing the remote sensing image can only recognize the road that is not blocked by objects, and there is a great difficulty in recognizing the road type in the region blocked by objects, which leads to low recognition accuracy and low recall rate.

[0045] Based on this, the embodiments of the present application provide a method for road recognition. The method for road recognition can consider the traffic flow characteristics and the traffic speed characteristics on the basis of the remote sensing image, so that the road type of the to-be-recognized road region finally recognized by the target road recognition model is closer to the actual situation, can be applied to the region scene blocked by objects and can be applied to the region scene not blocked by objects, greatly makes up for the defects brought by only depicting the road type from the remote sensing image, reduces the difficulty of road type recognition, and improves the recognition accuracy and the recall rate. Exemplarily, the method for road recognition provided in the embodiments of the present application can be applied to various scenes such as maps, intelligent transportation, autonomous driving, artificial intelligence, auxiliary driving and Internet of Vehicles.

[0046] It should be noted that the intelligent traffic system (Intelligent Traffic System, ITS) described in the field of intelligent traffic is also called intelligent transportation system (Intelligent Transportation System), which is to effectively integrate advanced science and technology (information technology, computer technology, data communication technology, sensor technology, electronic control technology, automatic control theory, operations research, artificial intelligence, etc.) in transportation, service control and vehicle manufacturing, and strengthen the connection between vehicles, roads and users, so as to form a comprehensive transportation system that ensures safety, improves efficiency, improves environment and saves energy. Or;

[0047] Intelligent vehicle infrastructure cooperative system (Intelligent Vehicle Infrastructure Cooperative Systems, IVICS), simply referred to as vehicle infrastructure cooperative system, is a development direction of intelligent transportation system (ITS). The vehicle infrastructure cooperative system is to use advanced wireless communication and new generation Internet technology to implement dynamic real-time information interaction between vehicles and roads in all directions, and to develop vehicle active safety control and road cooperative management on the basis of full-time and space dynamic traffic information collection and fusion, to fully realize the effective cooperation of man, vehicle and road, to ensure traffic safety and improve traffic efficiency, so as to form a safe, efficient and environmentally friendly road traffic system.

[0048] Next, a method for road recognition provided by an embodiment of the present application will be described with reference to the accompanying drawings. Figure 1 A flowchart of the method for road recognition provided by the embodiment of the present application is shown. As shown in Figure 1 The method for road recognition can include the following steps:

[0049] 101, obtain the traffic flow characteristics and the traffic speed characteristics of the to-be-recognized road region, the traffic flow characteristics are used to indicate the vehicle flow on the to-be-recognized road region, and the traffic speed characteristics are used to indicate the driving speed of the vehicle on the to-be-recognized road region.

[0050] In this example, since the vehicle trajectory includes both spatial distribution characteristics and implicit dynamic characteristics, a trajectory static flow graph can be constructed based on the vehicle trajectory in a certain time period. The trajectory static flow graph includes the traffic flow and speed of the trajectory in the to-be-recognized road region. The spatial distribution characteristics described can include but are not limited to traffic flow characteristics. The dynamic characteristics can include but are not limited to traffic speed characteristics and the like. The traffic flow characteristics described can be used to indicate the vehicle flow on the to-be-recognized road region, and the traffic speed characteristics can indicate the driving speed of the vehicle on the to-be-recognized road region.

[0051] Exemplarily, the described trajectory static flow map can be understood with reference to the schematic diagram shown in FIG. 1. Figure 2 As shown in FIG. 1, the trajectory static flow map is constructed by a road recognition device. Figure 2 As shown in FIG. 1, the trajectory static flow map is constructed by a road recognition device.

[0052] The R channel layer can be understood as the actual flow value of the to-be-identified road region in the geographical space region. Exemplarily, the road recognition device can traverse each trajectory point on the to-be-identified road region and perform cumulative summation processing on each trajectory point to obtain the traffic flow feature of the to-be-identified road region. It should be noted that each trajectory point, also known as a GPS point, can be used to indicate the positioning position of the vehicle passing through the to-be-identified road region, such as p1 to p7 shown in FIG. 1. Figure 2 Therefore, after obtaining the traffic flow feature of the to-be-identified road region, the traffic flow feature can be regarded as the R channel layer. In addition, the greater the value of the R channel layer (such as 0 to 255), the greater the traffic flow of the to-be-identified road region, and the visual effect presents a large proportion of red components.

[0053] The B1 channel layer can be understood as the projection speed of the actual driving speed of the vehicle passing through the to-be-identified road region in the first direction in the geographical space region. The G1 channel layer can be understood as the projection speed of the actual driving speed of the vehicle passing through the to-be-identified road region in the second direction in the geographical space region. The first direction and the second direction are perpendicular, for example, if the first direction is the north direction, the second direction can be the east direction, which is not limited in the embodiments of the present application. In addition, the greater the value of the B1 channel layer (such as 0 to 255), the greater the component of the traffic speed of the vehicle in the current position along the first direction, and the visual effect presents a large proportion of blue components. Similarly, the greater the value of the B1 channel layer (such as 0 to 255), the greater the component of the traffic speed of the vehicle in the current position along the first direction, and the visual effect presents a large proportion of blue components. In addition, since the B1 channel layer and the G1 channel layer here are used to represent the speed of the vehicle, the value of the B1 channel layer and the value of the G1 channel layer can also be normalized to the final B channel layer.

[0054] In some examples, the road recognition device can traverse each trajectory point on the road region to be identified, and calculate the traffic speed feature according to the projection speed of each trajectory point in at least two directions of the road region to be identified. For example, by calculating the projection speed of each trajectory point in a first direction of the road region to be identified, and the projection of each trajectory point in a second direction of the road region to be identified. Then, the road recognition device respectively averages and accumulates the projection speed of each trajectory point in the first direction of the road region to be identified, and the projection speed of each trajectory point in the second direction of the road region to be identified, to obtain a first projection speed and a second projection speed, the first projection speed indicating a traffic speed component of the vehicle in the first direction, and the second projection speed indicating a traffic speed component of the vehicle in the second direction. Finally, the road recognition device normalizes the first projection speed and the second projection speed to obtain the traffic speed feature. It should be noted that after obtaining the traffic speed feature of the road region to be identified, the traffic speed feature can be regarded as a B channel layer in the subsequent modeling process.

[0055] 102. Obtain a remote sensing image of the road region to be identified.

[0056] In this example, the remote sensing image can also be referred to as a remote sensing image, which is not limited in the embodiments of the present application. The remote sensing image of the road region to be identified can include the remote sensing image of each road in the road region to be identified.

[0057] The described remote sensing image can include but is not limited to multispectral remote sensing image, hyperspectral remote sensing image, low spectral remote sensing image, etc., which is not limited in the embodiments of the present application.

[0058] 103. Perform feature extraction processing on the remote sensing image to obtain a road network feature of the road region to be identified.

[0059] In this example, if the original color remote sensing image is directly selected as the road network feature of the road region to be identified, the road information in the remote sensing image is weakened due to the existence of many other objects such as buildings, roads, farms, forests, rivers, etc. in the remote sensing image, and a lot of noise data is generated in the labeling process. Therefore, after obtaining the remote sensing image of the road region to be identified, it is necessary to perform binaryzation processing on the remote sensing image to extract the corresponding feature. For example, after obtaining the remote sensing image of the road region to be identified, feature extraction processing is performed on the remote sensing image to extract the road network feature of the road region to be identified.

[0060] In some examples, the road recognition device performs feature extraction processing on the remote sensing image to obtain road network features of the to-be-identified road region, which can be implemented by using a deep learning technology. Illustratively, after obtaining the remote sensing image, the road recognition device can perform road recognition processing on the remote sensing image by using a preset recognition model, and then obtain a road segmentation map of the to-be-identified road region. Then, the road recognition device can take the road segmentation map of the to-be-identified road region as the road network features of the to-be-identified road region. The described road segmentation map can be understood as a segmentation map formed after all roads in the to-be-identified road region are divided. In addition, the described preset recognition model can include but is not limited to a segformer model, and the present application embodiment is not limited to the description.

[0061] Illustratively, after extracting the road network features of the to-be-identified road region, the road recognition device can also take the road network features as a G channel layer.

[0062] 104. The road recognition device performs road recognition processing on the traffic flow features, the traffic speed features, and the road network features based on a target road recognition model to obtain a target classification result, and the target classification result is used to indicate a road type of a road in the to-be-identified road region.

[0063] In this example, after obtaining the traffic flow features and the traffic speed features of the to-be-identified road region and extracting the corresponding road network features, the road recognition device can take the traffic flow features, the traffic speed features, and the road network features as inputs of the target road model, and then perform road recognition processing on the traffic flow features, the traffic speed features, and the road network features by using the target road model, so as to obtain a target classification result. The target classification result can indicate the road type of the road in the to-be-identified road region.

[0064] Exemplarily, the target road recognition model is used to identify the traffic flow feature, the traffic speed feature, and the road network feature to obtain a target classification result. The target classification result can be obtained by the following method: the traffic flow feature, the traffic speed feature, and the road network feature are fused to obtain a target fusion feature; and the target fusion feature is taken as an input of the target road recognition model to obtain the target classification result. Exemplarily, the traffic flow feature can be taken as a first channel layer, the traffic speed feature can be taken as a second channel layer, and the road network feature can be taken as a third channel layer. Then, the first channel layer, the second channel layer, and the third channel layer are modeled to obtain the target fusion feature. It should be noted that the first channel layer can be understood as the R channel layer described above, the second channel layer can be understood as the B channel layer described above, and the third channel layer can be understood as the G channel layer. The R channel layer (i.e., the traffic flow feature), the B channel layer (i.e., the traffic speed feature), and the G channel layer (i.e., the road network feature) are modeled in three channels, so that the traffic flow feature, the traffic speed feature, and the road network feature can be better fused. For example, the density map obtained after modeling can be understood with reference to the schematic diagram described above. Figure 3 As can be seen from Figure 3 The density map not only includes the space-time information and the speed information of the road, but also includes rich remote sensing spatial information.

[0065] It should be noted that the first channel layer is taken as the R channel layer, the second channel layer is taken as the B channel layer, and the third channel layer is taken as the G channel layer in the modeling description. In actual application, the first channel layer can also be the G channel layer or the B channel layer, the second channel layer can also be the R channel layer or the G channel layer, and the third channel layer can also be the R channel layer or the B channel layer. In the embodiment of the present application, the first channel layer, the second channel layer, and the third channel layer constitute the RGB channel layer, and the specific type of the channel layer is not limited in the present application.

[0066] The training process of the target road recognition model described above can be understood with reference to the content of the training process shown in Figure 4 As shown in Figure 4 The model training process of the target road recognition model includes at least the following steps:

[0067] 401、obtain a traffic flow feature representation, a traffic speed feature representation, and a road network feature representation of a training sample. The training sample is a sample in which a preset road type has been labeled on a fusion feature obtained by fusing the traffic flow feature representation, the traffic speed feature representation, and the road network feature representation. The road network feature representation is obtained from a remote sensing image of the training sample.

[0068] In this example, the process of generating training samples can be referenced. Figure 5 Use the described diagram to understand. Figure 5 As shown, the road recognition device can acquire training samples, including trajectory data and remote sensing imagery. Then, by analyzing the trajectory data, the road recognition device can extract traffic flow feature representations and traffic speed feature representations. Furthermore, by using remote sensing imagery as input to models such as the Segformer model, the road recognition device can extract relevant road network feature representations. It should be understood that, to achieve better training results, this embodiment selects road samples from different regions, such as sparsely traversed suburbs, densely traversed urban areas, and special areas like some intersections, as training samples. Moreover, the road recognition device needs to fuse the traffic flow feature representations, traffic speed feature representations, and road network feature representations, then manually annotate the fused features, and add the manually annotated samples with preset road types to the training set to obtain complete training samples.

[0069] 402. The traffic flow feature representation, traffic speed feature representation, and road network feature representation of the training samples are used as inputs to the initial road recognition model to obtain the target road type.

[0070] In this example, after obtaining the traffic flow feature representation, traffic speed feature representation, and road network feature representation, these features can be fused to obtain a fused feature. Then, the fused feature is used as input to the initial road recognition model, which identifies the corresponding target road type.

[0071] 403. Calculate the difference between the target road type and the preset road type to obtain the target loss value.

[0072] In this example, because the output of the deep neural network (such as the initial road recognition model described above) is expected to be as close as possible to the value that is actually intended to be predicted, the weight vector of each layer of the neural network can be updated by comparing the predicted value of the current network with the target value that is actually intended to be predicted, and then adjusting the weight vector according to the difference between the two. (Of course, before the first update, there is usually an initialization process, that is, the parameters of each layer of the deep neural network are pre-configured.) For example, if the predicted value of the network is too high, the weight vector is adjusted to make it predict lower, and the adjustment is continuously made until the neural network can predict the target value that is actually intended to be predicted. Therefore, it is necessary to define "how to compare the difference between the predicted value and the target value", which is the loss function or the objective function, which is an important equation for measuring the difference between the predicted value and the target value. Taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, and then the training of the deep neural network becomes a process of trying to minimize this loss.

[0073] Therefore, after the road recognition device predicts the target road type, it needs to calculate the difference between the target road type and the preset road type that has been labeled, and then obtain the corresponding target loss value.

[0074] 404、based on the target loss value, updating the model parameters of the initial road recognition model to obtain a target road recognition model.

[0075] In this example, after the road recognition device calculates the target loss value, it can continuously adjust and update the model parameters of the initial road recognition model based on the target loss value until the initial road recognition model converges iteratively, thereby training the target road recognition model.

[0076] In this way, after the road recognition device trains the target road recognition model, it can take the traffic flow feature, the traffic speed feature, and the road network feature as the input of the target road model, and then perform road recognition processing through the target road model to obtain a target classification result. Illustratively, based on the target road recognition model, the road recognition device performs road recognition processing on the traffic flow feature, the traffic speed feature, and the road network feature to obtain a target classification result, which can be achieved by the following method: fusing the traffic flow feature, the traffic speed feature, and the road network feature to obtain a target fusion feature; taking the target fusion feature as the input of the target road recognition model to obtain the target classification result.

[0077] It should be noted that the target road recognition model described can include but is not limited to a D-linkNet network, and the embodiments of the present application are not limited to the description.

[0078] Referring toFigure 6 This is a schematic diagram of the target road recognition model provided in an embodiment of this application. Figure 6 As shown, taking the D-linkNet network as the target road recognition model as an example, the D-linkNet network includes an encoder subnetwork, a feature extraction subnetwork, and a decoder subnetwork. The described encoder subnetwork may include, but is not limited to, the ResNet34 network. After fusing traffic flow features, traffic speed features, and road network features to obtain target fused features, the road recognition device can input the target fused features into the D-linkNet network to encode them through the encoder subnetwork, obtaining the first output feature. For example, the encoder subnetwork first uses a convolutional layer with a kernel size of 7×7, 64 filters, and a downsampling stride of 2 to perform a convolution operation on the target fused feature, and then performs pooling processing on the result of the convolution. The pooling window is 3×3 and the downsampling stride is 2. Next, the processed features are input into four encoding units containing residual blocks for processing, thereby obtaining the first output feature. The number of Res-blocks included in each encoding unit is 3, 4, 6, and 3, respectively.

[0079] Then, the road recognition device can perform feature extraction processing on the first output features through the feature extraction subnetwork in the D-linkNet network to obtain the second output features. The feature extraction subnetwork consists of dilated convolutions and attention modules. For example, Figure 7 A schematic diagram of the feature extraction subnetwork in the D-linkNet network is shown. Figure 7 As shown, this feature extraction sub-network can be composed of dilated convolutions and convolutional block attention modules, forming a parallel network structure. Through this dilated convolutional layer, the spatial resolution of the features can be maintained, while increasing the receptive range of features in the central part of the network and preserving detailed information. It should be noted that the dilated convolutional layers in the feature extraction sub-network used in the D-linkNet network have skip-connections. Figure 7As shown, the feature extraction subnetwork is divided into 5 extraction units, which perform the following operations respectively: ① the first output feature of the feature extraction subnetwork is directly processed and output; ② the first output feature of the feature extraction subnetwork is processed by a dilated convolution operation with a dilation rate of 1 and then processed and output; ③ the first output feature of the feature extraction subnetwork is processed by dilated convolution operations with dilation rates of 1 and 2 in turn and then processed and output; ④ the first output feature of the feature extraction subnetwork is processed by dilated convolution operations with dilation rates of 1, 2 and 4 and then processed and output; ⑤ the first output feature of the feature extraction subnetwork is processed by dilated convolution operations with dilation rates of 1, 2, 4 and 8 and then processed and output. The processes of the five extraction units are parallel to each other, and the outputs of each are added at the end of the feature extraction subnetwork to serve as the overall output of the feature extraction subnetwork, i.e., the second output feature is obtained.

[0080] Finally, after obtaining the first output feature and the second output feature, the road recognition device inputs the first output feature and the second output feature into the decoder subnetwork in the D-linkNet network to perform decoding processing by the decoder subnetwork in the D-linkNet network to obtain the target classification result.

[0081] It should be noted that the above only takes the D-linkNet network as an example to illustrate the target road recognition model, which is not specifically limited in the embodiments of the present application.

[0082] In the embodiments of the present application, the remote sensing image of the to-be-recognized road region is obtained, and feature extraction processing is performed on the remote sensing image to obtain the road network features of the to-be-recognized road region. Moreover, since the traffic flow feature can indicate the vehicle flow on the to-be-recognized road region, and the traffic speed feature can indicate the driving speed of the vehicle on the to-be-recognized road region, the traffic flow feature and the traffic speed feature of the to-be-recognized road region can also be obtained. Then, the target road recognition model is used to perform road recognition processing on the traffic flow feature, the traffic speed feature and the road network feature, and further obtain the target classification result, so as to indicate the road type of the to-be-recognized road region through the target classification result. Through the above manner, since the traffic flow feature and the traffic speed feature can depict the trajectory data of the vehicle, the influence of the road by other objects when the road recognition is performed based on the remote sensing image can be eliminated. Therefore, on the basis of the remote sensing image, the traffic flow feature and the traffic speed feature are taken into account, so that the road type of the to-be-recognized road region finally recognized by the target road recognition model is closer to the actual situation, which can be applied to the region scene blocked by objects and the region scene not blocked by objects, greatly makes up for the defects caused by depicting the road type only by the remote sensing image, reduces the difficulty of road type recognition, and improves the recognition accuracy and recall rate.

[0083] For example, Figure 8 and Figure 9 both show the comparison diagram of the identification result of the present application and the identification result of the prior art. As shown in Figure 8 and Figure 9 After comparing the results obtained by only performing road identification based on remote sensing images in the prior art, it can be found that: after adding trajectory features (i.e. the aforementioned traffic flow features and traffic speed features) on the basis of remote sensing image identification, and after the adaptation and training of the target road identification model, the road predicted by the target road identification model can greatly solve the problem that the corresponding road cannot be identified due to dense vegetation, factories and clouds covering the road, and also solve the problem that the road cannot be identified due to the lack of trajectory.

[0084] The above mainly introduces the scheme provided by the embodiments of the present application from the perspective of the method. It can be understood that in order to realize the above functions, the corresponding hardware structure and / or software module for executing each function is included. Those skilled in the art should easily realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0085] The embodiments of the present application can divide the functional modules of the device according to the above method examples, for example, each functional module can be divided according to each function, or two or more functions can be integrated in one processing module. The integrated module can be realized in the form of hardware or software functional module. It should be noted that the division of modules in the embodiments of the present application is illustrative, and is only a logical functional division. There can be another division way when actually implemented.

[0086] The road identification device in the embodiments of the present application will be described in detail below, Figure 10 is a schematic diagram of one embodiment of the road identification device provided in the embodiments of the present application. As shown in Figure 10 The road identification device can include an acquisition unit 1001 and a processing unit 1002.

[0087] The acquisition unit 1001 is configured to acquire a traffic flow feature and a traffic speed feature of a to-be-identified road region, the traffic flow feature being used to indicate a vehicle flow passing through the to-be-identified road region, and the traffic speed feature being used to indicate a driving speed of a vehicle passing through the to-be-identified road region. The acquisition unit 1001 is configured to acquire a remote sensing image of the to-be-identified road region. The processing unit 1002 is configured to perform feature extraction processing on the remote sensing image to obtain a road network feature of the to-be-identified road region. The processing unit 1002 is configured to perform road identification processing on the traffic flow feature, the traffic speed feature and the road network feature based on a target road identification model to obtain a target classification result, the target classification result being used to indicate a road type of a road in the to-be-identified road region.

[0088] In some optional embodiments, the processing unit 1002 is configured to: traverse each trajectory point on the to-be-identified road region, each trajectory point being used to indicate a positioning position of a vehicle passing through the to-be-identified road region; and perform cumulative summation processing on each trajectory point to obtain the traffic flow feature of the to-be-identified road region.

[0089] In some optional embodiments, the processing unit 1002 is configured to: traverse each trajectory point on the to-be-identified road region, each trajectory point being used to indicate a positioning position of a vehicle passing through the to-be-identified road region; and calculate the traffic speed feature based on a projection speed of each trajectory point in at least two directions of the to-be-identified road region.

[0090] In some optional embodiments, the processing unit 1002 is configured to: calculate a projection speed of each trajectory point in a first direction of the to-be-identified road region and a projection speed of each trajectory point in a second direction of the to-be-identified road region, the first direction being perpendicular to the second direction, and the first direction and the second direction being any two of the at least two directions; perform average cumulative calculation on the projection speed of each trajectory point in the first direction of the to-be-identified road region and the projection speed of each trajectory point in the second direction of the to-be-identified road region, respectively, to obtain a first projection speed and a second projection speed, the first projection speed being used to indicate a traffic speed component of the vehicle in the first direction, and the second projection speed being used to indicate a traffic speed component of the vehicle in the second direction; and perform normalization processing on the first projection speed and the second projection speed to obtain the traffic speed feature.

[0091] In some optional embodiments, the obtaining unit 1001 is further configured to obtain a traffic flow feature representation, a traffic speed feature representation, and a road network feature representation of a training sample, the training sample being a sample in which a preset road type has been labeled on a fusion feature obtained after the traffic flow feature representation, the traffic speed feature representation, and the road network feature representation are fused, and the road network feature representation being obtained from a remote sensing image of the training sample. The processing unit 1002 is further configured to take the traffic flow feature representation, the traffic speed feature representation, and the road network feature representation as inputs of an initial road recognition model, to obtain a target road type; to calculate a difference between the target road type and the preset road type, to obtain a target loss value; and to update model parameters of the initial road recognition model based on the target loss value, to obtain the target road recognition model.

[0092] In some optional embodiments, the processing unit 1002 is configured to perform fusion processing on the traffic flow feature, the traffic speed feature, and the road network feature, to obtain a target fusion feature; and to take the target fusion feature as an input of the target road recognition model, to obtain the target classification result.

[0093] In some optional embodiments, the target road recognition model includes a D-linkNet network. The processing unit 1002 is configured to input the target fusion feature into the D-linkNet network, to perform encoding processing on the target fusion feature by an encoder subnetwork in the D-linkNet network, to obtain a first output feature; to input the first output feature into a feature extraction subnetwork in the D-linkNet network, to perform feature extraction processing, to obtain a second output feature, the feature extraction subnetwork being composed of a dilated convolution and a convolution block attention module; and to input the first output feature and the second output feature into a decoder subnetwork in the D-linkNet network, to perform decoding processing by the decoder subnetwork in the D-linkNet network, to obtain the target classification result.

[0094] In some optional embodiments, the processing unit 1002 is configured to take the traffic flow feature as a first channel layer, the traffic speed feature as a second channel layer, and the road network feature as a third channel layer, the first channel layer, the second channel layer, and the third channel layer constituting an RGB channel; and to perform modeling processing on the first channel layer, the second channel layer, and the third channel layer, to obtain a target fusion feature.

[0095] In some alternative embodiments, the processing unit 1002 is configured to perform road recognition processing on the remote sensing image based on a preset recognition model to obtain a road segmentation map of the to-be-recognized road region; and use the road segmentation map of the to-be-recognized road region as a road network feature of the to-be-recognized road region.

[0096] In some alternative embodiments, the remote sensing image includes a multispectral remote sensing image, a hyperspectral remote sensing image, or a low-spectrum remote sensing image.

[0097] The road recognition device in the embodiments of the present application is described above from the perspective of a modular functional entity. The road recognition device in the embodiments of the present application is described below from the perspective of hardware processing. The described road recognition device can be a terminal device, a server, or the like. Figure 11 FIG. 1 is a structural schematic diagram of a road recognition device provided by an embodiment of the present application. The road recognition device can have great differences due to different configurations or performances. The road recognition device can include at least one processor 1101, a communication line 1107, a memory 1103, and at least one communication interface 1104.

[0098] The processor 1101 can be a general central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling program execution of the present application.

[0099] The communication line 1107 can include a path for transmitting information between the above components.

[0100] The communication interface 1104 uses any transceiver-like device for communicating with other devices or communication networks, such as an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), or the like.

[0101] The memory 1103 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, and the memory can exist independently or be connected to the processor through the communication line 1107. The memory can also be integrated with the processor.

[0102] The memory 1103 is configured to store computer-executable instructions for implementing the solutions of the present application, and the processor 1101 is configured to execute the computer-executable instructions stored in the memory 1103. The processor 1101 is configured to execute the computer-executable instructions stored in the memory 1103, so as to implement the methods provided in the above-described embodiments of the present application.

[0103] Optionally, the computer-executable instructions in the embodiments of the present application can also be referred to as application codes, which are not limited in the embodiments of the present application.

[0104] In a specific implementation, as an example, the road recognition apparatus can include a plurality of processors, for example, the processor 1101 and the processor 1102 in the apparatus 1100. Each of the processors can be a single-CPU processor or a multi-CPU processor. The processor herein can refer to one or more apparatuses, circuits, and / or processing cores for processing data (for example, computer program instructions). Figure 11

[0105] In a specific implementation, as an example, the road recognition apparatus can further include an output device 1105 and an input device 1106. The output device 1105 is in communication with the processor 1101, and can display information in various ways. The input device 1106 is in communication with the processor 1101, and can receive input of a target object in various ways. For example, the input device 1106 can be a mouse, a touch screen apparatus, a sensor apparatus, or the like.

[0106] The road recognition apparatus described above can be a general-purpose apparatus or a special-purpose apparatus. In a specific implementation, the road recognition apparatus can be a server, a terminal device, or the like, or an apparatus having a similar structure in the apparatus 1100. The embodiments of the present application do not limit the type of the road recognition apparatus. Figure 11

[0107] It should be noted that the processor 1101 in the apparatus 1100 can cause the road recognition apparatus to perform the methods in the method embodiments corresponding to the apparatus 1100 and the apparatus 1100 by invoking the computer-executable instructions stored in the memory 1103. Figure 11 Figure 1 Figure 4

[0108] Specifically, the functions / implementation processes of the processing unit 1002 in the apparatus 1000 can be implemented by the processor 1101 in the apparatus 1100 invoking the computer-executable instructions stored in the memory 1103. Figure 10 Figure 11 The functions / implementation processes of the acquisition unit 1001 in the apparatus 1000 can be implemented by the communication interface 1104 in the apparatus 1100. Figure 10 Figure 11

[0109] ​​​​​​​​The embodiment of the present application further provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, and the computer program causes a computer to execute part or all of steps of any one of the road recognition methods described in the above method embodiments.

[0110] The embodiment of the present application further provides a computer program product, which comprises a non-transitory computer readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of steps of any one of the road recognition methods described in the above method embodiments.

[0111] In the above embodiment, the system, device and unit can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the system, device and unit can be realized in the form of a computer program product, in whole or in part.

[0112] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the above method embodiments, which will not be described here.

[0113] In the several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of the units is only a logical function division. There can be another division manner for actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0114] The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0115] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software functional unit.

[0116] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0117] The above embodiments can be implemented wholly or partially by software, hardware, firmware, or any combination thereof. When implemented by software, the embodiments can be implemented wholly or partially in the form of a computer program product.

[0118] The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that the computer can store or the data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, an SSD), etc.

[0119] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of road recognition, characterized in that, The method comprises the following steps: obtaining traffic flow characteristics and traffic speed characteristics of a to-be-identified road region, wherein the traffic flow characteristics are used to indicate the vehicle flow on the to-be-identified road region, and the traffic speed characteristics are used to indicate the driving speed of the vehicles on the to-be-identified road region; obtaining remote sensing images of each road in the to-be-identified road region; performing feature extraction processing on the remote sensing images to obtain road network characteristics of the to-be-identified road region; taking the traffic flow characteristics as a first channel layer, taking the traffic speed characteristics as a second channel layer, and taking the road network characteristics as a third channel layer, wherein the first channel layer, the second channel layer, and the third channel layer constitute an RGB channel; performing modeling processing on the first channel layer, the second channel layer, and the third channel layer to obtain target fusion characteristics; inputting the target fusion characteristics into a D-linkNet network in a target road identification model to perform encoding processing on the target fusion characteristics through an encoder subnetwork in the D-linkNet network to obtain first output characteristics; the training process of the target road identification model comprises the following steps: obtaining traffic flow characteristic representations, traffic speed characteristic representations, and road network characteristic representations of training samples, wherein the training samples are samples in which a preset road type has been labeled on fusion characteristics obtained after the traffic flow characteristic representations, the traffic speed characteristic representations, and the road network characteristic representations are fused, and the road network characteristic representations are obtained from remote sensing images of the training samples; taking the traffic flow characteristic representations, the traffic speed characteristic representations, and the road network characteristic representations as inputs of an initial road identification model to obtain a target road type; calculating the difference between the target road type and the preset road type to obtain a target loss value; updating model parameters of the initial road identification model based on the target loss value to obtain the target road identification model; inputting the first output characteristics into a feature extraction subnetwork in the D-linkNet network to perform feature extraction processing to obtain second output characteristics, wherein the feature extraction subnetwork is composed of a hollow convolution and a convolution block attention module; the feature extraction subnetwork comprises a plurality of extraction units, the processes of the extraction units are parallel to each other, and the outputs of the extraction units are added at the end of the feature extraction subnetwork to serve as the overall output of the feature extraction subnetwork, thereby obtaining the second output characteristics; inputting the first output characteristics and the second output characteristics into a decoder subnetwork in the D-linkNet network to perform decoding processing through the decoder subnetwork to obtain a target classification result, wherein the target classification result is used to indicate the road type of the road in the to-be-identified road region.

2. The method of claim 1, wherein, The method comprises the following steps: traversing each trajectory point on the to-be-identified road region, wherein each trajectory point is used to indicate the positioning position of a vehicle on the to-be-identified road region; The traffic flow feature of the to-be-identified road region is obtained by performing cumulative summation processing on each trajectory point.

3. The method according to claim 1 or 2, characterized in that, The traffic speed feature is obtained by: traversing each trajectory point on the to-be-identified road region, each trajectory point being used to indicate the positioning position of a vehicle passing through the to-be-identified road region; calculating the traffic speed feature based on the projection speed of each trajectory point in at least two directions of the to-be-identified road region.

4. The method of claim 3, wherein, The traffic speed feature is calculated based on the projection speed of each trajectory point in at least two directions of the to-be-identified road region, including: calculating the projection speed of each trajectory point in a first direction of the to-be-identified road region and the projection speed of each trajectory point in a second direction of the to-be-identified road region, the first direction being perpendicular to the second direction, the first direction and the second direction being any two of the at least two directions; performing average cumulative calculation on the projection speed of each trajectory point in the first direction of the to-be-identified road region and the projection speed of each trajectory point in the second direction of the to-be-identified road region respectively to obtain a first projection speed and a second projection speed, the first projection speed being used to indicate the passing speed component of the vehicle in the first direction, and the second projection speed being used to indicate the passing speed component of the vehicle in the second direction; performing normalization processing on the first projection speed and the second projection speed to obtain the traffic speed feature.

5. The method according to claim 1 or 2, characterized in that, The feature extraction processing is performed on the remote sensing image to obtain the road network feature of the to-be-identified road region, including: performing road identification processing on the remote sensing image based on a preset identification model to obtain a road segmentation map of the to-be-identified road region; taking the road segmentation map of the to-be-identified road region as the road network feature of the to-be-identified road region.

6. The method of claim 1 or 2, wherein, The remote sensing image includes multispectral remote sensing image, hyperspectral remote sensing image or low-spectral remote sensing image.

7. A road recognition apparatus characterized by comprising: including: an acquisition unit, configured to acquire a traffic flow feature and a traffic speed feature of a to-be-identified road region, the traffic flow feature being used to indicate the vehicle flow passing through the to-be-identified road region, and the traffic speed feature being used to indicate the driving speed of the vehicle passing through the to-be-identified road region; the acquisition unit is configured to acquire a remote sensing image of each road in the to-be-identified road region; a processing unit, configured to perform feature extraction processing on the remote sensing image to obtain a road network feature of the to-be-identified road region; the processing unit is configured to perform road identification processing on the traffic flow feature, the traffic speed feature and the road network feature based on a target road identification model to obtain a target classification result, the target classification result being used to indicate the road type of the road in the to-be-identified road region; The training process of the target road recognition model comprises: obtaining traffic flow feature representation, traffic speed feature representation and road network feature representation of a training sample, the training sample being a sample having a preset road type labeled on a fusion feature obtained after the traffic flow feature representation, the traffic speed feature representation and the road network feature representation are fused, the road network feature representation being obtained from a remote sensing image of the training sample; taking the traffic flow feature representation, the traffic speed feature representation and the road network feature representation as inputs of an initial road recognition model to obtain a target road type; calculating a difference between the target road type and the preset road type to obtain a target loss value; updating model parameters of the initial road recognition model based on the target loss value to obtain the target road recognition model. The processing unit is configured to: take the traffic flow feature as a first channel layer, take the traffic speed feature as a second channel layer, and take the road network feature as a third channel layer, the first channel layer, the second channel layer and the third channel layer constituting an RGB channel; perform modeling processing on the first channel layer, the second channel layer and the third channel layer to obtain a target fusion feature; take the target fusion feature as an input of the target road recognition model to obtain the target classification result. The target road recognition model comprises a D-linkNet network; the processing unit is configured to: input the target fusion feature into the D-linkNet network to perform encoding processing on the target fusion feature through an encoder subnetwork in the D-linkNet network to obtain a first output feature; input the first output feature into a feature extraction subnetwork in the D-linkNet network to perform feature extraction processing to obtain a second output feature, the feature extraction subnetwork being composed of a dilated convolution and a convolution block attention module; input the first output feature and the second output feature into a decoder subnetwork in the D-linkNet network to perform decoding processing through the decoder subnetwork in the D-linkNet network to obtain the target classification result; the feature extraction subnetwork comprises a plurality of extraction units, the process of each extraction unit being parallel to each other, the output of each extraction unit being added after the end of the feature extraction subnetwork to serve as the overall output of the feature extraction subnetwork, thereby obtaining the second output feature.

8. The road recognition apparatus according to claim 7, characterized by The processing unit is configured to: traverse each trajectory point on the to-be-identified road region, each trajectory point being used to indicate a positioning position of a vehicle passing through the to-be-identified road region; perform cumulative summation processing on each trajectory point to obtain a traffic flow feature of the to-be-identified road region.

9. The road recognition apparatus according to claim 7 or 8, characterized by The processing unit is configured to: traverse each trajectory point on the to-be-identified road region, each trajectory point being used to indicate a positioning position of a vehicle passing on the to-be-identified road region; and calculate the passing speed feature based on a projection speed of each trajectory point in at least two directions of the to-be-identified road region.

10. The road recognition apparatus according to claim 9, characterized by The processing unit is configured to: calculate a projection speed of each trajectory point in a first direction of the to-be-identified road region and a projection speed of each trajectory point in a second direction of the to-be-identified road region, the first direction being perpendicular to the second direction, and the first direction and the second direction being any two of the at least two directions. The projection speed of each trajectory point in the first direction of the to-be-identified road region and the projection speed of each trajectory point in the second direction of the to-be-identified road region are respectively averaged and accumulated to obtain a first projection speed and a second projection speed, the first projection speed being used to indicate a passing speed component of the vehicle in the first direction, and the second projection speed being used to indicate a passing speed component of the vehicle in the second direction. The first projection speed and the second projection speed are normalized to obtain the passing speed feature.

11. The road recognition apparatus according to claim 7 or 8, characterized by The processing unit is configured to: perform road identification processing on the remote sensing image based on a preset identification model to obtain a road segmentation map of the to-be-identified road region; and use the road segmentation map of the to-be-identified road region as a road network feature of the to-be-identified road region.

12. The road recognition apparatus according to claim 7 or 8, characterized by The remote sensing image includes a multispectral remote sensing image, a hyperspectral remote sensing image, or a low-spectral remote sensing image.

13. A road recognition apparatus characterized by comprising: The road identification device includes an input / output (I / O) interface, a processor, and a memory, and the memory stores program instructions. The processor is configured to execute the program instructions stored in the memory to perform the method in any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions, which, when executed on a computer device, cause the computer device to perform the method in any one of claims 1 to 6.

15. A computer program product, characterised in that, The computer program product includes instructions, which, when executed on a computer device or a processor, cause the computer device or the processor to perform the method in any one of claims 1 to 6.

Images