Road plane intersection extraction method, device and equipment and storage medium

Abstract

Embodiments of the present disclosure provide a road plane intersection extraction method, device and equipment and a storage medium, which are applied to the technical field of remote sensing image application. The method comprises: acquiring a remote sensing image of a target area; inputting the remote sensing image of the target area into a pre-trained road plane intersection extraction network model for semantic segmentation, and outputting a road segmentation result image; extracting a road center line based on a thinning algorithm for the road segmentation result image; performing raster vectorization processing on the road center line to obtain a starting point and an ending point corresponding to the road center line; performing two-dimensional grid space division on a space range in which the road center line after the raster vectorization processing is located; and if there are more than three starting points or ending points in a same grid, marking the grid as a road plane intersection and outputting the same. In this way, the accuracy of road plane intersection information acquisition can be improved and the acquisition cost can be reduced.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of remote sensing image application, and in particular to a road plane intersection extraction method, device, equipment and storage medium. BACKGROUND

[0002] The road plane intersection mainly refers to the road structure of two or more roads intersecting at the same plane or the road mode of multiple different planes intersecting, which is an important part of the traffic micro level and is an important key link for influencing traffic safety and improving traffic flow. Related research shows that about 70% of traffic accidents occur at plane intersections, and therefore, accurately obtaining road intersections is of great significance for eliminating road safety hazards and improving intersection traffic efficiency.

[0003] At present, the method for obtaining the geographical position of the road plane intersection includes an extraction method based on road network basic data, an extraction method based on remote sensing images, and an extraction method based on road traffic data. The extraction of the road plane intersection based on the road network basic data requires obtaining the road network basic data, and the road intersection data cannot be obtained in the absence of the road network basic data. The extraction of the road plane intersection based on the traffic data requires a large amount of vehicle trajectory point information, and the extraction efficiency is relatively low. With the popularization of remote sensing images in the field of transportation, the objectivity and extensive coverage of remote sensing images can greatly improve the accuracy and cost of obtaining road plane intersection information. Therefore, the extraction of the road plane intersection from remote sensing images has the characteristics of high efficiency and low cost, and the research on the automatic extraction technology is of great significance. The automatic extraction method of the road plane intersection based on remote sensing images mainly includes a method based on deep learning target detection and a method based on deep learning road network segmentation. Among them, the road plane intersection extraction method based on deep learning target detection has a relatively low recognition accuracy due to the complexity of the features of the road plane intersection in the image. The road plane intersection extraction method based on remote sensing image semantic segmentation has a relatively high road recognition accuracy, but requires higher cost for fine sample labeling. However, the current deep learning network model based on remote sensing image semantic segmentation mainly uses the cross entropy loss function CrossEntropy Loss or its variant focal loss function Focal Loss as the loss function. Such loss functions only consider the probability similarity between the predicted image and the label, and ignore the influence of image texture information on the model. However, the road has a typical texture feature of being long and uniform inside, which is a very typical feature, and ultimately affects the accuracy of the road plane intersection extraction. SUMMARY

[0004] The present disclosure provides a road plane intersection extraction method, device, equipment and storage medium.

[0005] According to a first aspect of the present disclosure, a road plane intersection extraction method is provided. The method comprises:

[0006] obtaining a remote sensing image of a target area;

[0007] inputting the remote sensing image of the target area into a pre-trained road plane intersection extraction network model for semantic segmentation, and outputting a road segmentation result image;

[0008] extracting a road centerline from the road segmentation result image based on a thinning algorithm;

[0009] performing raster vectorization processing on the road centerline to obtain a starting point and an ending point corresponding to the road centerline;

[0010] performing two-dimensional grid space division on a space range in which the road centerline after raster vectorization processing is located;

[0011] if there are more than three starting points or ending points in the same grid, the grid is marked as a road plane intersection and output.

[0012] Further, the training process of the road plane intersection extraction network model includes the production process of the sample image set and the label image set, including:

[0013] obtaining a remote sensing image of a typical area to obtain an initial sample image set;

[0014] performing pixel-by-pixel labeling on road regions in the initial sample image set using an image labeling tool to obtain an initial label image set;

[0015] performing enhancement operations on the initial sample image set and the initial label image set to obtain corresponding sample image sets and label image sets; wherein the enhancement operations include rotation, flipping, random cropping, and center cropping.

[0016] Further, the training process of the road plane intersection extraction network model includes:

[0017] training a semantic segmentation network model using the sample image set, and using the trained model to infer the sample image set to obtain a semantic segmentation result image set;

[0018] calculating a loss of the semantic segmentation network model according to the semantic segmentation result image set and the label image set;

[0019] using the semantic segmentation network model and the loss as a road plane intersection extraction network model.

[0020] Further, the loss of the semantic segmentation network model is calculated according to the semantic segmentation result image set and the label image set, and the loss calculation comprises:

[0021] Gram matrices of the semantic segmentation result image set and the label image set are constructed, and a texture loss is calculated according to the Gram matrices;

[0022] A focus loss is calculated according to coordinates of each pixel point in the semantic segmentation result image set and a corresponding predicted correct probability value, wherein the predicted correct probability value is a probability that each pixel point in the semantic segmentation result image is equal to a corresponding pixel point in the label image;

[0023] The loss of the semantic segmentation network model is calculated according to the texture loss and the focus loss.

[0024] Further, a formula for calculating the texture loss according to the Gram matrices is as follows:

[0025]

[0026]

[0027]

[0028] wherein O i , O j respectively represent a vector composed of i-th row elements of a pixel-by-pixel probability prediction map output by a model and a vector composed of j-th row elements of the pixel-by-pixel probability prediction map; L i , L j respectively represent a vector composed of i-th row elements of a label image set L and a vector composed of j-th row elements of the label image set L; G O , G L respectively represent Gram matrices of O and L; N×N and M×M respectively represent sizes of G O , G L , N and M respectively represent lengths of edges of O and L; wherein O is a semantic segmentation result image set, and L is a label image set.

[0029] Further, a formula for calculating the focus loss according to coordinates of each pixel point in the semantic segmentation result image set and a corresponding predicted correct probability value is as follows:

[0030]

[0031] wherein a and b represent coordinates of pixel points in O, represents a predicted correct probability value of the pixel points a and b, α represents a balance factor, and γ represents an adjustment factor; wherein O is a semantic segmentation result image set.

[0032] Further, the formula for calculating the loss of the semantic segmentation network model according to the texture loss and the focus loss is as follows:

[0033] L total =βL focal +(1-β)L style

[0034] wherein β represents a loss weight, L focal represents the focus loss, and L style represents the texture loss.

[0035] According to a second aspect of the present disclosure, a road plane intersection extraction device is provided. The device comprises:

[0036] an image acquisition module configured to acquire a remote sensing image of a target region;

[0037] an image segmentation inference module configured to input the remote sensing image of the target region into a pre-trained road plane intersection extraction network model for semantic segmentation, and output a road segmentation result image;

[0038] a center line extraction module configured to extract a road center line based on a thinning algorithm for the road segmentation result image;

[0039] a processing module configured to perform raster vectorization processing on the road center line to obtain a starting point and an ending point corresponding to the road center line;

[0040] a space division module configured to perform two-dimensional grid space division on a space range in which the road center line after the raster vectorization processing is located;

[0041] an intersection output module configured to, if there are more than three starting points or ending points in a same grid, mark and output the grid as a road plane intersection.

[0042] According to a third aspect of the present disclosure, an electronic device is provided. The electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the method as described above when executing the program.

[0043] According to a fourth aspect of the present disclosure, a computer readable storage medium is provided, which stores a computer program, and the program is executed by a processor to implement the method according to the first aspect of the present disclosure.

[0044] The road plane intersection extraction method, device, equipment and storage medium provided by the embodiments of the present disclosure can greatly improve the accuracy of road plane intersection information acquisition and reduce the acquisition cost by using the pre-trained road plane intersection extraction network model to segment the road image.

[0045] It is to be understood that the description in the summary is not intended to identify key or essential features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the disclosure will be apparent from review of the description below and the associated drawings. BRIEF DESCRIPTION OF DRAWINGS

[0046] The above and other features, aspects and advantages of embodiments of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings. The drawings are intended to illustrate but not to limit the present disclosure. In the drawings:

[0047] Figure 1 A flowchart of a road plane intersection extraction method according to an embodiment of the present disclosure is shown;

[0048] Figure 2 A flowchart of a road plane intersection extraction method according to another embodiment of the present disclosure is shown;

[0049] Figure 3 A flowchart of a road plane intersection extraction method according to another embodiment of the present disclosure is shown;

[0050] Figure 4 A flowchart of a road plane intersection extraction method according to another embodiment of the present disclosure is shown;

[0051] Figure 5 A block diagram of a road plane intersection extraction apparatus according to an embodiment of the present disclosure is shown;

[0052] Figure 6 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0053] In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present disclosure.

[0054] In addition, the term "and / or" herein is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects.

[0055] In the present disclosure, in order to make full use of the texture characteristics of the road in the remote sensing image, an image texture information supervised model training is introduced, a new loss function is proposed, a remote sensing image road plane intersection extraction method based on texture information supervision is realized, and the accuracy of road plane intersection extraction is improved.

[0056] Figure 1 A flowchart of a road plane intersection extraction method 100 according to an embodiment of the present disclosure is shown. The method 100 includes:

[0057] Step 110, acquiring a remote sensing image of a target area.

[0058] In some embodiments, due to the application and promotion of remote sensing images in the field of transportation, the objectivity and extensive coverage of remote sensing images are beneficial to the extraction of road plane intersections in the target area.

[0059] Step 120, inputting the remote sensing image of the target area into a pre-trained road plane intersection extraction network model for semantic segmentation, and outputting a road segmentation result image.

[0060] In some embodiments, as shown in the schematic diagram of the training process of the road plane intersection extraction network model, the training process includes the following steps: Figure 2

[0061] Step 210, training the semantic segmentation network model using the sample image set, and using the trained model to infer the sample image set to obtain a semantic segmentation result image set.

[0062] Step 220, calculating the loss of the semantic segmentation network model according to the semantic segmentation result image set and the label image set.

[0063] Step 230, taking the semantic segmentation network model and the loss as a road plane intersection extraction network model

[0064] ​In some embodiments, any semantic segmentation network model is selected, such as SGCN (Split Depth-wise Separable Graph Convolution Network) for road extraction in remote sensing images, as the segmentation model M. SGCN uses ResNet50 as its backbone, and the output of the backbone undergoes depthwise separable convolution processing. Then, an upsampling module gradually restores the feature maps to the original image size. Finally, a softmax function is used to form a pixel-wise category prediction map, resulting in a semantic segmentation image set. Based on the semantic segmentation image set and the label image set, the loss of the semantic segmentation network model is calculated, and this loss is used as the loss of the road intersection extraction network model when extracting road intersections.

[0065] In some embodiments, such as Figure 3 The diagram shown in step 210 illustrates the process of creating the sample image set for the road intersection network model, which involves extracting samples. The process includes the following steps:

[0066] Step 310: Obtain remote sensing images of typical areas to obtain an initial sample image set.

[0067] Step 320: Use an image annotation tool to annotate the road areas in the initial sample image set pixel by pixel to obtain an initial label image set.

[0068] Step 330: Perform enhancement operations on the initial sample image set and the initial label image set to obtain the corresponding sample image set and label image set.

[0069] The enhanced operations include rotation, flipping, random cropping, and center cropping.

[0070] In some embodiments, remote sensing images of typical areas (with similar features to the target area, such as image resolution, color space, etc.) are acquired to generate training samples for a road intersection extraction network model, serving as an initial sample image set. A sample labeling tool, such as Labelme, is used to label road areas in the remote sensing images pixel-by-pixel with an envelope, and the image labels are converted to PNG format to obtain an initial label image set. Enhancement operations, including rotation, flipping, random cropping, and center cropping, are then performed on the initial sample image set and the initial label image set to obtain sample image set I and label image set L.

[0071] In some embodiments, such as Figure 4 The diagram shown in step 220 illustrates the loss generation process of the semantic segmentation network model, which includes the following steps:

[0072] Step 410, construct a Gram matrix of the semantic segmentation result image set and the label image set, and calculate a texture loss according to the Gram matrix.

[0073] Step 420, calculate a focus loss according to the coordinates of each pixel point in the semantic segmentation result image set and the corresponding predicted correct probability value.

[0074] The predicted correct probability value is the probability that each pixel point in the semantic segmentation result image is equal to the corresponding pixel point in the label image.

[0075] Step 430, calculate a loss of the semantic segmentation network model according to the texture loss and the focus loss.

[0076] In some embodiments, then the sample image set I is input into the segmentation model M to obtain a semantic segmentation result image set O. The Gram matrix (Gram matrix) of the semantic segmentation result O and the label image set L is calculated, respectively denoted as For any matrix F, the Gram matrix is calculated as follows:

[0077] G ij =F i F j

[0078] Where F represents any matrix, F i , F j is the value of the i, j element of the Gram matrix, and G ij is the inner product of the i-th row and the j-th row of F. Then, the texture loss is calculated based on the following formula:

[0079]

[0080]

[0081]

[0082] Where O i , O j respectively represent the i-th row element vector and the j-th row element vector of the pixel-by-pixel probability prediction map output by the model; L i , L j respectively represent the i-th row element vector and the j-th row element vector of the label image set L; G O , G L respectively represent the Gram matrix of O and L; N x N, M x M are respectively the dimensions of G O , G Lthe size of O and L, N and M are the length of O and L respectively; wherein, O is the semantic segmentation result image set, L is the label image set. Then, the focal loss is calculated based on the following formula:

[0083]

[0084] wherein, a, b represent the coordinates of the pixel points in O, represents the predicted correct probability value of the pixel points a, b, and a represents the balance factor, and g represents the adjustment factor, wherein O is the semantic segmentation result image set. Wherein, for any pixel point, for example, the predicted correct probability value of pixel point A is the probability that the A pixel point in the semantic segmentation result image is equal to the pixel point corresponding to the A pixel point in the label image……, and so on for other pixel points. Then, the loss weight b is calculated based on the following formula

[0085]

[0086] wherein, g is an adjustment factor, set to the current round, used to control the information proportion of texture information in the loss function, and epoch_num represents the total training rounds. The value of epoch_num can be set differently for different data sets, first preset a fixed b value, find the round corresponding to the vertex where the mIoU value of the deep learning computer vision image segmentation field reaches the top as epoch_num. For example, g takes the current round number, and epoch_num takes 50. Texture information supervision can accelerate model training and improve model accuracy in the early stage; while after a certain number of training, the pixel-by-pixel class prediction map and the label image are basically consistent in texture, and the texture effect needs to be weakened, otherwise it will mislead the model. Therefore, the value of b needs to be adjusted according to the change of the round, so as to calculate the loss weight b of the loss function according to this formula, used to adjust the proportion weight of texture loss, so that the prediction result is more accurate. Finally, according to the calculated texture loss, focal loss and loss weight, the final loss of the semantic segmentation network model is obtained based on the following formula.

[0087] L total = bL focal + (1-b) L style

[0088] wherein, b represents the loss weight, L focal represents the focal loss, and L style represents the texture loss.

[0089] The loss for remote sensing image road extraction based on texture information auxiliary supervision fully combines the texture characteristics of the road target in the remote sensing image (with typical elongated and internally uniform texture characteristics). On the basis of considering class balance, the structural and texture differences between the output image and the label image are introduced, which improves the performance of the model and enhances the accuracy of road extraction. Among them, the influence factor, i.e., the loss weight and the adjustment factor, is designed to fully utilize the influence degree and proportion of texture information in different rounds of training, so that the texture information supervision accelerates the model training in the early stage and improves the model accuracy.

[0090] In step 130, based on the road segmentation result image, the road center line is extracted based on the thinning algorithm.

[0091] In some embodiments, based on the road segmentation result, the thinning algorithm is used to extract the road skeleton, and after deburring, vector smoothing and other operations, the thinned road center line is formed.

[0092] In step 140, the road center line is subjected to raster vectorization processing to obtain the start point and the end point corresponding to the road center line.

[0093] In some embodiments, the road center line obtained by step 130 is subjected to raster vectorization, and the start point and the end point of each road after vectorization are extracted.

[0094] In step 150, the space range of the road center line after raster vectorization processing is subjected to two-dimensional grid space division.

[0095] In some embodiments, the overall space range of the road vector is divided into a two-dimensional grid with the same horizontal and vertical width.

[0096] In step 160, if there are more than three start points or end points in the same grid, the grid is marked as a road plane intersection and output.

[0097] In some embodiments, since the remote sensing image covers a relatively large area, in order to quickly extract the road intersection, the characteristics of the road intersection (three or more road segments intersecting at the same position) are used to extract the road intersection. If there are more than three road segment start points or end points in a certain grid, it can be determined that there is a branch intersection in the grid. The grid frame (top left corner coordinates, grid width) is marked and output, and the output result is the position of the extracted road intersection. Based on the two-dimensional grid road intersection extraction, the road segmentation result based on the semantic segmentation model can be displayed in the form of a target detection frame, and the extraction accuracy is higher than that of directly using target detection for sample labeling.

[0098] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all described as a combination of a series of actions, but those skilled in the art should know that the present disclosure is not limited by the order of the described actions, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily necessary for the present disclosure.

[0099] The above is the introduction of the method embodiment, and the scheme of the present disclosure is further described through the device embodiment.

[0100] Figure 5 A block diagram of a road plane intersection extraction device 500 according to an embodiment of the present disclosure is shown. As shown in Figure 5 The device 500 includes:

[0101] An image acquisition module 510 is configured to acquire a remote sensing image of a target region.

[0102] An image segmentation inference module 520 is configured to input the remote sensing image of the target region into a pre-trained road plane intersection extraction network model for semantic segmentation, and output a road segmentation result image.

[0103] A center line extraction module 530 is configured to extract a road center line based on a thinning algorithm for the road segmentation result image.

[0104] A processing module 540 is configured to perform raster vectorization processing on the road center line to obtain a starting point and an ending point corresponding to the road center line.

[0105] A space division module 550 is configured to perform two-dimensional grid space division on a space range in which the road center line after raster vectorization processing is located.

[0106] An intersection output module 560 is configured to, if there are more than three starting points or ending points in the same grid, mark and output the grid as a road plane intersection.

[0107] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described modules can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0108] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0109] Figure 6A schematic block diagram of an electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0110] The electronic device 600 includes a computing unit 601 that can perform various appropriate actions and processes in accordance with a computer program stored in a ROM 602 or a computer program loaded from the storage unit 608 into a RAM 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An I / O interface 605 is also connected to the bus 604.

[0111] A plurality of components in the electronic device 600 are connected to the I / O interface 605, including an input unit 606, such as a keyboard, a mouse, and the like, an output unit 607, such as various types of displays, a speaker, and the like, a storage unit 608, such as a magnetic disk, an optical disk, and the like, and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 609 allows the electronic device 600 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0112] The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, and the like. The computing unit 601 performs various methods and processes described above, such as the method 100. For example, in some embodiments, the method 100 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded onto the RAM 603 and executed by the computing unit 601, one or more steps of the method 100 described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform the method 100 by any other suitable means, such as by means of firmware.

[0113] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0114] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0115] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0116] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0117] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0118] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0119] It should be understood that the various forms of flow shown above can be re-ordered, steps added or removed. For example, the steps recited in the present disclosure can be performed in parallel, in series, in a different order, or any combination thereof, so long as the desired results of the technology disclosed in the present disclosure are achieved, which is not limited herein.

[0120] The specific implementation described above does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A road plane intersection extraction method characterized by, The method comprises the following steps: acquiring a remote sensing image of a target region; inputting the remote sensing image of the target region into a pre-trained road plane intersection extraction network model for semantic segmentation, and outputting a road segmentation result image; wherein the training process of the road plane intersection extraction network model comprises a sample image set and a label image set production process, which comprises the following steps: acquiring a remote sensing image of a region to obtain an initial sample image set; performing pixel-by-pixel labeling on road regions in the initial sample image set by using an image labeling tool to obtain an initial label image set; performing an enhancement operation on the initial sample image set and the initial label image set to obtain corresponding sample image sets and label image sets; wherein the enhancement operation comprises rotation, flipping, random cropping, and center cropping; the training process of the road plane intersection extraction network model comprises the following steps: training a semantic segmentation network model by using the sample image set, and obtaining a semantic segmentation result image set by using the trained model to infer the sample image set; constructing a Gram matrix of the semantic segmentation result image set and the label image set, and calculating a texture loss according to the Gram matrix: in , These represent the vectors formed by the elements of the i-th row and the vectors formed by the elements of the j-th row of the pixel-by-pixel probability prediction map output by the model, respectively. Representing the label image set The vector formed by the elements of the i-th row and the vector formed by the elements of the j-th row; This represents the Gram matrix of the segmented image with parameters i and j. This represents the Gram matrix of the labeled image set with parameters i and j; They represent , The order of the matrix; i and j are integers. calculating a focal loss according to the coordinates of each pixel point in the semantic segmentation result image set and the corresponding predicted correct probability value; wherein the predicted correct probability value is the probability that each pixel point in the semantic segmentation result image is equal to the corresponding pixel point in the label image; calculating the loss of the semantic segmentation network model according to the texture loss and the focal loss; the semantic segmentation network model and the loss are used as a road plane intersection extraction network model; extracting a road center line based on a thinning algorithm on the road segmentation result image; performing raster vectorization processing on the road center line to obtain the start point and the end point corresponding to the road center line; performing two-dimensional grid space division on the space range of the road center line after raster vectorization processing; if there are more than three start points or end points in the same grid, the grid is marked as a road plane intersection and output.

2. The method of claim 1, wherein, The formula for calculating the focal loss according to the coordinates of each pixel point in the semantic segmentation result image set and the corresponding predicted correct probability value is as follows: wherein, represents a coordinate of a pixel point, a prediction correct probability value of the pixel point, represents a balance factor, represents an adjustment factor.

3. The method of claim 1, wherein, The formula for calculating the loss of the semantic segmentation network model according to the texture loss and the focal loss is as follows: wherein, represents a loss weight, L focal represents a focal loss, L style represents a texture loss.

4. A road plan intersection extraction apparatus characterized by comprising: The method comprises the following steps: an image acquisition module is configured to acquire a remote sensing image of a target region; an image segmentation inference module is configured to input the remote sensing image of the target region into a pre-trained road plane intersection extraction network model for semantic segmentation, and output a road segmentation result image; wherein the training process of the road plane intersection extraction network model comprises a sample image set and a label image set production process, which comprises the following steps: acquiring a remote sensing image of a region to obtain an initial sample image set; performing pixel-by-pixel labeling on road regions in the initial sample image set by using an image labeling tool to obtain an initial label image set; performing an enhancement operation on the initial sample image set and the initial label image set to obtain corresponding sample image sets and label image sets; wherein the enhancement operation comprises rotation, flipping, random cropping, and center cropping; the training process of the road plane intersection extraction network model comprises the following steps: The semantic segmentation network model is trained by using the sample image set, and a semantic segmentation result image set is obtained by using the trained model to infer the sample image set; constructing a Gram matrix of the semantic segmentation result image set and the label image set, and calculating a texture loss according to the Gram matrix: wherein , respectively denote a vector of the i-th row elements and a vector of the j-th row elements of the pixel-wise probability prediction map output by the model; respectively denote a vector of the i-th row elements and a vector of the j-th row elements of the set of label images ; denotes a segmentation result image Gram matrix parameterized by i, j, denotes a label image set Gram matrix parameterized by i, j; respectively denote , a matrix order; i, j denote integers, respectively. According to the coordinates of each pixel point in the semantic segmentation result image set and the corresponding predicted correct probability value, a focal loss is calculated; wherein the predicted correct probability value is the probability that each pixel point in the semantic segmentation result image is equal to the corresponding pixel point in the label image; According to the texture loss and the focal loss, the loss of the semantic segmentation network model is calculated; The semantic segmentation network model and the loss are used as a road plane intersection extraction network model; a center line extraction module is used to extract a road center line based on a thinning algorithm for the road segmentation result image; A processing module is used to perform raster vectorization processing on the road center line to obtain the start point and end point corresponding to the road center line; A space division module is used to perform two-dimensional grid space division on the space range of the road center line after raster vectorization processing; An intersection output module is used to mark and output a grid as a road plane intersection if there are more than three start points or end points in the same grid.

5. An electronic device, comprising: Comprise: At least one processor; And The memory is connected in communication with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method of any one of claims 1-3.

6. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to execute the method according to any one of claims 1-3.

Images