CN115641512B - Satellite remote sensing image road identification method, device, equipment and medium - Google Patents
Satellite remote sensing image road identification method, device, equipment and medium Download PDFInfo
- Publication number
- CN115641512B CN115641512B CN202211672238.XA CN202211672238A CN115641512B CN 115641512 B CN115641512 B CN 115641512B CN 202211672238 A CN202211672238 A CN 202211672238A CN 115641512 B CN115641512 B CN 115641512B
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- road
- backbone
- prediction result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The application discloses a method, a device, equipment and a medium for identifying a satellite remote sensing image road, and relates to the technical field of image data road identification. The method comprises the steps that a primary remote sensing image of a target object is input into a trained road segmentation model to obtain a segmentation result remote sensing image; then, splicing the three-channel matrix of the original remote sensing image with the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result; then inputting the splicing result remote sensing image into a trained road connection model to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; and finally, identifying the road of the target object based on the remote sensing image of the prediction result. By the technical scheme, the road in the remote sensing image of the target object can be more accurately identified, so that the road prediction precision of the target object is improved.
Description
Technical Field
The application relates to the technical field of image data road identification, in particular to a satellite remote sensing image road identification method, device, equipment and medium.
Background
The remote sensing image is a film or a photo for recording the size of electromagnetic waves of various ground objects, and is mainly divided into an aerial photo and a satellite photo.
In the prior art, a method for identifying a road by using a satellite remote sensing image comprises the steps of obtaining a remote sensing image of a target object, processing the remote sensing image to identify the road in the remote sensing image, and accordingly identifying a real road of the target object. However, when the remote sensing image of the target road is shot, the remote sensing image is limited by objective factors and a satellite shooting angle, for example, buildings and trees beside the road block the road, so that the road in the remote sensing image is disconnected, the road in the remote sensing image of the target road cannot be accurately identified, and the prediction effect of the model is further influenced.
Disclosure of Invention
The application mainly aims to provide a method, a device, equipment and a medium for identifying roads in satellite remote sensing images, and aims to solve the technical problem that roads in remote sensing images of target objects cannot be accurately identified in the prior art.
In order to achieve the above object, a first aspect of the present application provides a method for identifying a satellite remote sensing image road, where the method includes:
inputting the original remote sensing image of the target object into the trained road segmentation model to obtain a segmentation result remote sensing image;
splicing the three-channel matrix of the native remote sensing image with the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result;
inputting the splicing result remote sensing image into a trained road connection model to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information simulating that a road in the native remote sensing image is blocked to disconnect the road;
identifying a road of the target object based on the prediction result remote sensing image;
wherein the identifying the road of the target object based on the remote sensing image of the prediction result comprises:
obtaining a backbone remote sensing image based on the prediction result remote sensing image; the backbone remote sensing image comprises a plurality of backbone data; the backbone data are used for representing data of a road in the prediction result remote sensing image;
extending target backbone data in the backbone remote sensing image to obtain a final remote sensing image;
and identifying the road of the target object based on the final remote sensing image.
Optionally, before the step of inputting the remote sensing image of the stitching result into the trained road connection model to obtain the remote sensing image of the prediction result, the method further includes:
acquiring a training set remote sensing image;
loading prior information to the training set remote sensing image to generate a prior information matrix;
splicing the three-dimensional matrix of the training set remote sensing image with the prior information matrix to obtain a four-dimensional input matrix;
and training a first basic network structure based on the four-dimensional input matrix to obtain a road connection model.
Optionally, before the step of loading the prior information to the training set remote sensing image to generate the prior information matrix, the method further includes:
preprocessing the training set remote sensing images to unify color features of the training set remote sensing images;
filtering pictures with background specific gravity larger than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a first remote sensing image;
performing data enhancement on the first remote sensing image to obtain a second remote sensing image;
the loading of prior information to the training set remote sensing image to generate a prior information matrix comprises:
and loading prior information to the second remote sensing image to generate a prior information matrix.
Optionally, before the step of inputting the native remote sensing image of the target object into the trained road segmentation model to obtain the segmentation result remote sensing image, the method further includes:
preprocessing the training set remote sensing images to unify color features of the training set remote sensing images;
filtering pictures with background specific gravity greater than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a third remote sensing image;
performing data enhancement on the third remote sensing image to obtain a fourth remote sensing image;
and training a second basic network structure based on the fourth remote sensing image to obtain a road segmentation model.
Optionally, the obtaining a backbone remote sensing image based on the prediction result remote sensing image includes:
carrying out binarization processing on the remote sensing image of the prediction result;
and extracting a plurality of backbone data of the remote sensing image of the prediction result after binarization processing to obtain a backbone remote sensing image.
Optionally, the extending the target backbone data in the backbone remote sensing image to obtain a final remote sensing image includes:
performing convolution operation on the backbone remote sensing image to obtain a characteristic remote sensing image;
restoring the size of the characteristic remote sensing image to the size of the native remote sensing image;
obtaining a plurality of endpoints in the reduced feature remote sensing image; wherein the end point is a point with a pixel value of 11 in the characteristic remote sensing image;
based on a first recursive function, obtaining the length of a line segment where an endpoint in the characteristic remote sensing image is located;
screening out line segments of which the lengths of the line segments of the endpoints in the characteristic remote sensing image are greater than a preset length;
obtaining the gradient of the end point of the line segment with the length larger than the preset length;
based on a second recursive function, emitting rays along the direction of the gradient to obtain a final remote sensing image; wherein the ray is used for generating a road in the occluded characteristic remote sensing image; the gradient direction is the direction of the road.
In a second aspect, the present application provides a satellite remote sensing image road recognition device, the device includes:
the first obtaining module is used for inputting the original remote sensing image of the target object into the trained road segmentation model so as to obtain a segmentation result remote sensing image;
the second obtaining module is used for splicing the three-channel matrix of the native remote sensing image and the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result;
the third obtaining module is used for inputting the splicing result remote sensing image into a trained road connection model so as to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information for simulating that a road in the original remote sensing image is blocked so as to disconnect the road;
the identification module is used for identifying the road of the target object based on the prediction result remote sensing image;
wherein the identifying the road of the target object based on the remote sensing image of the prediction result comprises:
obtaining a backbone remote sensing image based on the prediction result remote sensing image; the backbone remote sensing image comprises a plurality of backbone data; the backbone data are used for representing data of a road in the prediction result remote sensing image;
extending target backbone data in the backbone remote sensing image to obtain a final remote sensing image;
and identifying the road of the target object based on the final remote sensing image.
In a third aspect, the present application provides a computer device, which includes a memory and a processor, where the memory stores a computer program, and the processor executes the computer program to implement the method described in the embodiment.
In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein a processor executes the computer program to implement the method described in the embodiments.
Through above-mentioned technical scheme, this application has following beneficial effect at least:
the embodiment of the application provides a method, a device, equipment and a medium for identifying a satellite remote sensing image road, which comprise the following steps: inputting a primary remote sensing image of a target object into a trained road segmentation model to obtain a segmentation result remote sensing image; then, splicing the three-channel matrix of the original remote sensing image with the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result; then inputting the splicing result remote sensing image into a trained road connection model to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information for simulating that a road in the original remote sensing image is blocked so as to disconnect the road; and finally, identifying the road of the target object based on the remote sensing image of the prediction result.
When a road of a target object needs to be identified, an unprocessed remote sensing image of the target object is input into a road segmentation model trained in advance, a segmentation result remote sensing image is obtained after the road segmentation model is processed, then a three-channel matrix of the unprocessed remote sensing image is spliced with the segmentation result remote sensing image to obtain a splicing result remote sensing image after splicing, the splicing result remote sensing image is input into a trained road connection model to obtain a prediction result remote sensing image, and finally the real road of the target object is identified through the prediction result remote sensing image.
Namely, the road connection model is obtained based on prior information training, the prior information comprises information of the broken connection of the road caused by the fact that the road in the simulated native remote sensing image is shielded, namely prior knowledge is added to the effect of simulating the shielding of the road, the real value of the label is selected to be randomly shielded, and then the real value which is randomly shielded is used as the fourth-dimensional training road connection model of the input image. When the remote sensing image of the segmentation result is input into the road connection model, the road connection model can connect roads in the remote sensing image of the segmentation result which is predicted by the road segmentation model and is disconnected due to occlusion with the help of the prior knowledge. Because the roads which are disconnected in the remote sensing image of the segmentation result due to shielding of buildings, trees and the like in practice can be automatically connected through the road connection model, compared with the prior art that the disconnected roads cannot be identified by the remote sensing image, the method can more accurately identify the roads in the remote sensing image of the target object and has universality, thereby improving the connectivity of the whole road and improving the road prediction precision of the target object. Specifically, under the matching of the road connection model and the road segmentation model, the road prediction precision of the method is improved by about 7% -8% compared with that of a conventional segmentation model.
Drawings
FIG. 1 is a schematic diagram of a computer device in a hardware operating environment according to an embodiment of the present application;
fig. 2 is a flowchart of a method for identifying a satellite remote sensing image road according to an embodiment of the present disclosure;
FIG. 3 is a flowchart of a method for training a road connection model according to an embodiment of the present disclosure;
FIG. 4 is a flowchart of a method for training a road segmentation model according to an embodiment of the present application;
fig. 5 is a schematic flowchart of a specific implementation method of step S13 provided in this embodiment;
fig. 6 is a flowchart illustrating a specific implementation method of step S131 provided in this embodiment;
fig. 7 is a schematic diagram of a remote sensing image of a prediction result after binarization processing provided in an embodiment of the application;
fig. 8 is a flowchart illustrating a specific implementation method of step S132 provided in this embodiment;
FIG. 9 is a schematic diagram of a plurality of endpoints in a feature remote sensing image obtained according to an embodiment of the present application;
fig. 10 is a schematic diagram of obtaining the length of a line segment where an endpoint in a feature remote sensing image is located according to an embodiment of the present application;
FIG. 11 is a schematic diagram of calculating a gradient at an endpoint provided by an embodiment of the present application;
FIG. 12 is a schematic diagram of a final remote sensing image obtained by emitting rays along the direction of the gradient according to an embodiment of the present application;
FIG. 13 is a schematic diagram of a prior art remote sensing image obtained according to an embodiment of the present application;
FIG. 14 is a schematic illustration of a remote sensing image obtained by the present application;
fig. 15 is a schematic view of a satellite remote sensing image road recognition device according to an embodiment of the present application.
The implementation, functional features and advantages of the objectives of the present application will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In the current remote sensing satellite image road prediction algorithm, prediction is performed based on a basic model, and various parameters of the model are optimized according to the quality of the prediction effect, such as learning rate, optimizer, loss function and the like, or the structure of a network is modified, the connectivity of the road is increased, or the MIoU (Mean interaction of Unit) of the model is violently improved by a method of adding data. Firstly, the conventional methods are not universal and need to adjust parameters thereof according to each special model; secondly, stable output cannot be obtained after a large amount of time is used in the tuning process, and labor consumption is high; finally, the output results cannot be further improved due to objective factors, such as roads covered by trees, and roads covered by buildings due to the angle of the satellite images are mostly unconnected. In summary, at present, the road in the remote sensing image of the target object cannot be accurately identified, so that the road prediction accuracy of the target object is not high enough.
In order to solve the technical problems, the present application provides a method, an apparatus, a device and a medium for identifying a satellite remote sensing image road, and before introducing a specific technical scheme of the present application, a hardware operating environment related to the scheme of the embodiment of the present application is introduced.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a computer device in a hardware operating environment according to an embodiment of the present application.
As shown in fig. 1, the computer apparatus may include: a processor 1001, such as a Central Processing Unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement connection communication among these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a WIreless interface (e.g., a WIreless-FIdelity (WI-FI) interface). The Memory 1005 may be a Random Access Memory (RAM) Memory, or may be a Non-Volatile Memory (NVM), such as a disk Memory. The memory 1005 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the configuration shown in FIG. 1 does not constitute a limitation of a computer device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, a memory 1005, which is a storage medium, may include therein an operating system, a data storage module, a network communication module, a user interface module, and an electronic program.
In the computer device shown in fig. 1, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 of the computer device of the present application may be provided in the computer device, and the computer device calls the satellite remote sensing image road recognition device stored in the memory 1005 through the processor 1001 and executes the satellite remote sensing image road recognition method provided by the embodiment of the present application.
Referring to fig. 2, based on the hardware environment of the foregoing embodiment, an embodiment of the present application provides a method for identifying a satellite remote sensing image road, where the method includes:
s10: and inputting the primary remote sensing image of the target object into the trained road segmentation model to obtain a segmentation result remote sensing image.
In the specific implementation process, the target object refers to an object of which the road distribution condition needs to be obtained through a remote sensing image; the native remote sensing image is a remote sensing image which is shot and is not processed, and can be obtained through a satellite; the road segmentation model can be constructed in advance and trained well, the road segmentation model can be used for preliminarily identifying roads in the original remote sensing image, and the original remote sensing image preliminarily identified by the road segmentation model is a segmentation result remote sensing image.
S11: and splicing the three-channel matrix of the native remote sensing image and the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result.
In the specific implementation process, the three-channel matrix of the native remote sensing image refers to an RGB (red, green and blue) three-channel matrix of the native remote sensing image, and the result remote sensing image obtained by splicing simultaneously comprises three-channel matrix information of the native remote sensing image and remote sensing image information of the segmentation result. In addition, the way of stitching is known to those skilled in the art, for example, stitching is performed through a concat function, so that it is more convenient to identify the road in the remote sensing image through the road connection model at a later stage.
S12: inputting the splicing result remote sensing image into a trained road connection model to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information simulating that a road in the native remote sensing image is blocked to disconnect the road.
In the specific implementation process, the road connection model can further identify roads in the splicing result remote sensing images to obtain predicted result remote sensing images, and the road connection model mainly identifies roads which are in an off-link state on the splicing result remote sensing images because the roads are shielded by external factors such as buildings, trees and the like. Specifically, when the road connection model is trained, the remote sensing image with the prior information is used for training, namely the road connection model is trained by simulating the disconnected information of the road caused by the fact that the road in the original remote sensing image is shielded, and the disconnected road in the remote sensing image of the splicing result can be identified through the trained road connection model.
S13: and identifying the road of the target object based on the prediction result remote sensing image.
In the specific implementation process, just as the road connection model obtained through the prior information training can identify the blocked and disconnected road, the road in the blocked and disconnected remote sensing image can be identified more accurately based on the prediction result remote sensing image processed by the road connection model, so that the road in the target object can be identified more accurately.
In this embodiment, when a road of a target object needs to be identified, an unprocessed remote sensing image of the target object is input into a road segmentation model trained in advance, a segmentation result remote sensing image is obtained after the road segmentation model is processed, then a three-channel matrix of the unprocessed remote sensing image is spliced with the segmentation result remote sensing image, a splicing result remote sensing image is obtained after the splicing, then the splicing result remote sensing image is input into a trained road connection model, a prediction result remote sensing image is obtained, and finally a real road of the target object is identified through the prediction result remote sensing image.
Namely, the road connection model is obtained based on prior information training, the prior information comprises information of the broken connection of the road caused by the fact that the road in the simulated native remote sensing image is shielded, namely prior knowledge is added to the effect of simulating the shielding of the road, the real value of the label is selected to be randomly shielded, and then the real value which is randomly shielded is used as the fourth-dimensional training road connection model of the input image. When the remote sensing image of the segmentation result is input into the road connection model, the road connection model can connect roads in the remote sensing image of the segmentation result which is predicted by the road segmentation model and is disconnected due to occlusion with the help of the prior knowledge. The road connection model can automatically connect roads which are disconnected in the remote sensing image of the segmentation result in practice due to shielding of buildings, trees and the like. Therefore, compared with the prior art that the remote sensing image cannot identify the disconnected road, the method can more accurately identify the road in the remote sensing image of the target object and has universality, thereby improving the connectivity of the whole road and improving the road prediction precision of the target object. Specifically, the road prediction accuracy of the method is improved by about 7% -8% compared with a conventional segmentation model under the matching of the road connection model and the road segmentation model, and the road prediction accuracy of the method can be verified by a conventional method in a mode of improving by about 7% -8% compared with the conventional segmentation model, for example, the higher the parameter value is, the better the parameter value is, and the quality of the model can be measured from different angles through the parameter value.
In some embodiments, as shown in fig. 3, a preferred way of specifically training a road connection model is given, that is, before the step of inputting the stitching result remote sensing image into the trained road connection model to obtain a prediction result remote sensing image, the method further includes:
s20: and acquiring a training set remote sensing image.
In a specific implementation process, the training set remote sensing images can be native remote sensing images with blocked disconnected roads, because only the remote sensing images with the disconnected roads can train a road connection model capable of identifying the disconnected roads; the training set remote sensing images comprise as many remote sensing images as possible, so that the trained road connection model is more accurate.
S21: and loading prior information to the training set remote sensing image to generate a prior information matrix.
In the specific implementation process, the prior information is information for simulating a fracture sample presented in prediction by a tree or a blocked road of a building, so that after the training set remote sensing image is loaded, random blocking of graphs such as circles and squares is performed on the road disconnected labels to generate a required prior information matrix.
S23: and splicing the three-dimensional matrix of the training set remote sensing image with the prior information matrix to obtain a four-dimensional input matrix.
In the specific implementation process, the three-dimensional matrix of the training set remote sensing images is 3 x h x w, wherein 3 refers to three channels RGB of the training set remote sensing images, h refers to the height of the training set remote sensing images, and w refers to the width of the training set remote sensing images; the priori information matrix is a matrix of 1 × h × w, a three-dimensional picture information matrix of 3 × h × w and the priori information matrix of 1 × h × w are spliced to obtain a four-dimensional input matrix of 4 × h × w, and the four-dimensional input matrix can be spliced through a concat function. Therefore, the spliced four-dimensional input matrix not only comprises the information of the three-dimensional matrix of the remote sensing image of the training set, but also comprises the information of the prior information matrix, thereby being more beneficial to training to obtain a road connection model.
S24: and training the first basic network structure based on the four-dimensional input matrix to obtain a road connection model.
In a specific implementation, the first infrastructure network architecture may use an existing network architecture, such as HRnet-OCR. The first basic network structure is trained through a large number of remote sensing images comprising four-dimensional input matrix information, and in the training process, after certain adjustment is carried out on the hyper-parameters of the first basic network structure, such as batch training amount, an optimizer, an optimization method, learning rate, loss and the like, a road connection model capable of identifying disconnected roads can be obtained.
In this embodiment, prior information is loaded to a training set remote sensing image to obtain a prior information matrix with prior knowledge, so that the prior information matrix has prior information of a road disconnected due to occlusion, then the prior information matrix with the prior information and a three-dimensional matrix of the training set remote sensing image are spliced into a four-dimensional input matrix, and then a first basic network structure is trained based on the remote sensing image including the four-dimensional input matrix information to obtain a road connection model. Because the road connection model is formed by training a large number of training set remote sensing images with priori knowledge (priori information), the trained road connection model can identify disconnected roads more accurately, and therefore more accurate reference is provided for related personnel.
In order to obtain a more accurate prior information matrix, in some embodiments, the following preferred scheme is given: the step of loading prior information to the training set remote sensing image to generate a prior information matrix further comprises: preprocessing the training set remote sensing images to unify color characteristics of the training set remote sensing images; then, filtering pictures with background specific gravity larger than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a first remote sensing image; and finally, performing data enhancement on the first remote sensing image to obtain a second remote sensing image. The preset specific gravity can be set according to the requirement of the filter precision of a person skilled in the art, for example, the preset specific gravity is set to 80% in the present embodiment.
Therefore, the step of loading the prior information to the training set remote sensing image to generate a prior information matrix includes: and loading prior information to the second remote sensing image to generate a prior information matrix.
In this embodiment, some preprocessing operations are performed on the training set remote sensing images, and since the shooting of the remote sensing images is greatly affected by the illumination, the overall color style of the training set remote sensing images needs to be unified first, and the style of all the images needs to be transformed by using the network with the migrated style. And filtering the picture with the background specific gravity exceeding the preset specific gravity to obtain a first remote sensing image, and then performing data enhancement operation (such as rotation, overturning, noise adding and the like) on the filtered first remote sensing image to obtain a second remote sensing image. Therefore, the obtained second remote sensing image can better embody the characteristics favorable for training the road segmentation model, and the characteristics which are useless and influence the training road segmentation model are less, so that the prior information is loaded to the second remote sensing image, the generated prior information matrix is more accurate, and the more accurate road segmentation model can be trained.
In some embodiments, as shown in fig. 4, a preferred way to train the road segmentation model is given as follows: before the step of inputting the native remote sensing image of the target object into the trained road segmentation model to obtain the remote sensing image of the segmentation result, the method further comprises the following steps:
s30: preprocessing the training set remote sensing images to unify the color characteristics of the training set remote sensing images.
In the specific implementation process, some preprocessing operations are performed on the training set remote sensing images, and the shooting of the remote sensing images is greatly influenced by illumination, so that the overall color style of the training set remote sensing images needs to be unified, and specifically, the style of all the training set remote sensing images needs to be transformed by using a style migration network.
S31: and filtering pictures with background specific gravity greater than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a third remote sensing image.
In a specific implementation process, the pictures after the style transformation, namely the training set remote sensing images after the unified color features, are filtered to obtain a third remote sensing image after the picture with the background specific gravity exceeding a preset specific gravity (such as 80%).
S32: and performing data enhancement on the third remote sensing image to obtain a fourth remote sensing image.
In a specific implementation process, data enhancement operation is performed on the filtered third remote sensing image, for example, rotation, turning, noise addition and the like are performed, and then a fourth remote sensing image is obtained.
S33: and training a second basic network structure based on the fourth remote sensing image to obtain a road segmentation model.
In a specific implementation, the second infrastructure network may use an existing network structure, such as HRnet-OCR. And training the second basic network structure through a large number of fourth remote sensing images, and in the training process, after certain adjustment is carried out on the hyper-parameters (such as batch training amount, an optimizer, an optimization method, learning rate, loss and the like) of the second basic network structure, obtaining a road segmentation model capable of primarily identifying a road.
In the embodiment, the color features of the training set remote sensing images can be unified by preprocessing the color features of the training set remote sensing images, so that the influence degree of illumination on the training set remote sensing images during shooting can be reduced; then filtering the training set remote sensing images, and filtering out pictures with background specific gravity exceeding the preset specific gravity to obtain a third remote sensing image, so that the influence degree of the pictures with the background specific gravity exceeding the preset specific gravity on the overall quality of the training set remote sensing images can be reduced, and the training effect on the road segmentation model can be further provided; and finally, performing data enhancement on the third remote sensing image to obtain a fourth remote sensing image, wherein the fourth remote sensing image subjected to data enhancement can avoid overfitting when a road segmentation model is trained, and the generalization capability of the road segmentation model can be improved. And finally, the second basic network structure is trained through the fourth remote sensing image, and the road segmentation model obtained after training is more accurate, so that the road in the remote sensing image can be more accurately identified.
In some embodiments, as shown in fig. 5, the specific steps of step S13 are: the step of identifying the road of the target object based on the prediction result remote sensing image comprises the following steps:
s131: obtaining a backbone remote sensing image based on the prediction result remote sensing image; the backbone remote sensing image comprises a plurality of backbone data; and the backbone data is used for representing the data of the road in the remote sensing image of the prediction result.
In the specific implementation process, the prediction result remote sensing image comprises a plurality of roads, and the remote sensing image formed by extracting the roads is the backbone remote sensing image.
S132: and extending the target backbone data in the backbone remote sensing image to obtain a final remote sensing image.
In the specific implementation process, based on the backbone remote sensing image obtained in step S131, it can be known that there are several backbones, i.e., roads, that are disconnected, and these disconnected roads may be occluded and not reflected in the remote sensing image, so that disconnection occurs. All backbones (target backbone data) disconnected due to shielding need to be extended, the shielded parts of the road can be connected after the backbones are extended, and the remote sensing image obtained after the target backbone data are extended is the final remote sensing image.
S133: and identifying the road of the target object based on the final remote sensing image.
In the specific implementation process, the final remote sensing image comprises roads disconnected due to occlusion, and the roads are extended, so that the final remote sensing image can better embody the real road of the target object, and the road of the target object can be identified more accurately based on the final remote sensing image.
In this embodiment, a plurality of backbones of the remote sensing image with the prediction result are extracted to obtain a backbone remote sensing image, then target backbone data in the backbone remote sensing image is found out, wherein the target backbone data is the backbone data disconnected due to shielding, and then the target backbone data is extended, that is, the backbones are completely connected to obtain the final remote sensing image. Because the target backbone data in the final remote sensing image is completely extended, the obtained final remote sensing image can represent the blocked and disconnected road, and therefore the road of a target object can be identified more accurately based on the final remote sensing image.
In order to obtain a more accurate backbone remote sensing image, in some embodiments, as shown in fig. 6, the specific steps of step S131 are given: the step of obtaining the backbone remote sensing image based on the prediction result remote sensing image comprises the following steps:
s1311: and firstly, carrying out binarization processing on the remote sensing image of the prediction result.
In a specific implementation process, the remote sensing image of the prediction result is binarized in a conventional manner to obtain fig. 7, and fig. 7 is a schematic diagram of the remote sensing image of the prediction result after binarization processing provided by the embodiment of the application.
S1312: and then extracting a plurality of backbone data of the remote sensing image of the prediction result after binarization processing to obtain a backbone remote sensing image.
In the specific implementation process, a plurality of backbone data of the remote sensing image of the prediction result after binarization processing are extracted by using the sketch to obtain the backbone remote sensing image.
In order to obtain a more accurate final remote sensing image, in some embodiments, as shown in fig. 8, the specific steps of step S132 are given as follows: the step of extending the target backbone data in the backbone remote sensing image to obtain a final remote sensing image comprises the following steps:
s1321: and carrying out convolution operation on the backbone remote sensing image to obtain a characteristic remote sensing image.
In the specific implementation process, a customized 3 x 3 convolution core is used for carrying out convolution operation on the backbone remote sensing image to obtain the characteristic remote sensing image.
S1322: and reducing the size of the characteristic remote sensing image into the size of the native remote sensing image.
In a specific implementation process, the size of the characteristic remote sensing image is reduced to the size of the native remote sensing image through Padding2d [1,1 ].
S1323: obtaining a plurality of endpoints in the reduced feature remote sensing image; and the end point is a point with a pixel value of 11 in the characteristic remote sensing image.
In a specific implementation process, as shown in fig. 9, fig. 9 is a schematic diagram of a plurality of end points in an obtained feature remote sensing image provided in the embodiment of the present application, where a circle in the diagram includes the end points. Since it is guaranteed that all the values having the pixel value of 11 are end points after the convolution operation in step S1322 (the 3 × 3 convolution is initialized to:
[ [1 1 1]
[1 10 1]
[1 1 1] Where the calculated pixel value is 11 is an end point), the value of the retained pixel value 11 is assigned 2 (the reason for assigning 2 is that the pixel value is 0 to 255, and the calculated pixel value is within the interval for convenience), and the other values are assigned 1, and when the traversal result is 2, the point is the end point.
S1324: and obtaining the length of a line segment where the endpoint in the characteristic remote sensing image is located based on the first recursive function.
In a specific implementation process, as shown in fig. 10, fig. 10 is a schematic diagram of a length of a line segment where an endpoint in an obtained feature remote sensing image is located, according to an embodiment of the present application. And obtaining the length of the line segment where the endpoint in the characteristic remote sensing image is located through a recursive function.
S1325: and screening out the line segments of which the lengths of the end points in the characteristic remote sensing image are greater than the preset length.
In a specific implementation process, since a disconnected line segment (road) in a prediction result remote sensing image obtained after the road connection model processing is possibly caused by an error, the disconnected line segment (road) caused by the error is not disconnected due to the fact that the road is shielded by actual buildings, trees and the like, and the line segment is not a real road of a target object, the part of the road (the line segment where the characteristic remote sensing image is an endpoint) needs to be screened out. The screening method specifically comprises the steps of setting a preset length according to experience, wherein the preset length represents the length from the end point to the last intersection, and if the length of the line segment where the end point is located is larger than the preset length, the disconnection of the line segment where the end point is located is caused by shielding, and the road is a real road.
S1326: a gradient of the end points of the line segment greater than a preset length is obtained.
In a specific implementation process, after the step S1325 is performed to screen out the line segment representing the real road, the gradient of the endpoint is calculated in a conventional manner, as shown in fig. 11, where fig. 11 is a schematic diagram of calculating the gradient of the endpoint provided in the embodiment of the present application.
S1327: based on a second recursive function, emitting rays along the direction of the gradient to obtain a final remote sensing image; wherein the ray is used for generating a road in the occluded characteristic remote sensing image; the gradient direction is the direction of the road.
In a specific implementation process, as shown in fig. 12, fig. 12 is a schematic diagram of obtaining a final remote sensing image by emitting rays along the direction of the gradient according to an embodiment of the present application. And based on a second recursive function, emitting rays along the direction of the gradient, wherein the direction of the gradient is the direction of the road, and emitting rays to represent that the screened disconnected road is subjected to recursive extension until the recursive extension is stopped when the existing road is met, so that the screened line segment can be completely extended, and the road of the target object disconnected due to shielding is extended.
In the embodiment, the remote sensing image of the prediction result is subjected to binarization processing, and the remote sensing image of the prediction result after binarization processing is simpler, so that the remote sensing image of the prediction result can be further processed in the later period; then, extracting backbone data of the remote sensing image of the prediction result, namely extracting a road in the remote sensing image of the prediction result to obtain a backbone remote sensing image, performing convolution operation on the backbone remote sensing image to obtain a characteristic remote sensing image with more useful characteristics, and reducing the size of the characteristic remote sensing image to the size of the original remote sensing image, so that the consistency of the characteristic remote sensing image and the original remote sensing image can be better embodied, and the characteristics on the characteristic remote sensing image are more accurate; finding a plurality of end points in the characteristic remote sensing image, calculating the length of a line segment where the end point is located on the basis of the end points, and screening out line segments needing to be extended through preset length, wherein the line segments needing to be extended are line segments (roads) disconnected due to shielding; after finding the line segment needing to be extended, finding the extending direction of the line segment, namely, extending the line segment until the line segment is completely extended. Therefore, roads (line segments) which are occluded by buildings and the like can be connected completely, and a more accurate remote sensing image which can reflect the occluded roads can be obtained.
In summary, as shown in fig. 13 to fig. 14, fig. 13 is a schematic diagram of a remote sensing image obtained in the prior art according to an embodiment of the present application; fig. 14 is a schematic diagram of a remote sensing image obtained by the present application, and a connecting line in the way represents an actual road in a target object. The prediction results of the road segmentation model are optimized by using a brand-new road connection model, so that specific optimization is not required to be performed on a specific model, and a universal model is used for optimizing all similar road prediction models. The new road connection model adds certain priori knowledge before model input to help the model to distinguish the connection problem of the road. In order to simulate the effect that a road is blocked and add prior knowledge, the real value marked is selected to be randomly blocked, then the randomly blocked real value is used as the fourth dimension of an input picture and is transmitted into a network together with the picture, with the help of the prior knowledge, the model can finally connect the disconnected roads when the road segmentation model predicts, and then the connected roads are complemented with higher precision through a post-processing algorithm. By the method, the problem of manual optimization to a certain degree is solved, and meanwhile, the situation that in remote sensing image prediction, a road cannot be predicted after being shielded by trees or buildings can be overcome.
In another embodiment, as shown in fig. 15, based on the same inventive concept as the foregoing embodiment, an embodiment of the present application further provides a satellite remote sensing image road recognition apparatus, including:
the first obtaining module is used for inputting the original remote sensing image of the target object into the trained road segmentation model so as to obtain a segmentation result remote sensing image;
the second obtaining module is used for splicing the three-channel matrix of the native remote sensing image and the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result;
the third obtaining module is used for inputting the splicing result remote sensing image into a trained road connection model so as to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information simulating that a road in the native remote sensing image is blocked to disconnect the road;
and the identification module is used for identifying the road of the target object based on the remote sensing image of the prediction result.
It should be noted that, in this embodiment, each module in the satellite remote sensing image road identification device corresponds to each step in the satellite remote sensing image road identification method in the foregoing embodiment one to one, and therefore, the specific implementation and achieved technical effect of this embodiment may refer to the implementation of the satellite remote sensing image road identification method, which is not described herein again.
Furthermore, in an embodiment, the present application also provides a computer device comprising a processor, a memory and a computer program stored in the memory, which when executed by the processor implements the method in the preceding embodiment.
Furthermore, in an embodiment, the present application further provides a computer storage medium having a computer program stored thereon, where the computer program is executed by a processor to implement the method in the foregoing embodiment.
In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories. The computer may be a variety of computing devices including intelligent terminals and servers.
In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
By way of example, executable instructions may, but need not, correspond to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).
By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of other like elements in a process, method, article, or system comprising the element.
The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (e.g., a rom/ram, a magnetic disk, an optical disk) and includes instructions for enabling a multimedia terminal (e.g., a mobile phone, a computer, a television receiver, or a network device) to perform the method according to the embodiments of the present application.
The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all the equivalent structures or equivalent processes that can be directly or indirectly applied to other related technical fields by using the contents of the specification and the drawings of the present application are also included in the scope of the present application.
Claims (9)
1. A satellite remote sensing image road identification method is characterized by comprising the following steps:
inputting the original remote sensing image of the target object into the trained road segmentation model to obtain a segmentation result remote sensing image;
splicing the three-channel matrix of the native remote sensing image with the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result;
inputting the splicing result remote sensing image into a trained road connection model to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information for simulating that a road in the original remote sensing image is blocked so as to disconnect the road;
identifying a road of the target object based on the prediction result remote sensing image;
wherein the identifying the road of the target object based on the prediction result remote sensing image comprises:
obtaining a backbone remote sensing image based on the prediction result remote sensing image; the backbone remote sensing image comprises a plurality of backbone data; the backbone data are used for representing data of a road in the prediction result remote sensing image;
extending target backbone data in the backbone remote sensing image to obtain a final remote sensing image;
and identifying the road of the target object based on the final remote sensing image.
2. The method for road recognition based on satellite remote sensing images as claimed in claim 1, wherein before the step of inputting the remote sensing images of the stitching result into the trained road connection model to obtain the remote sensing images of the prediction result, the method further comprises:
acquiring a training set remote sensing image;
loading prior information to the training set remote sensing image to generate a prior information matrix;
splicing the three-dimensional matrix of the training set remote sensing image with the prior information matrix to obtain a four-dimensional input matrix;
and training the first basic network structure based on the four-dimensional input matrix to obtain a road connection model.
3. The method for identifying the satellite remote sensing image road according to claim 2, wherein before the step of loading the prior information into the training set remote sensing image to generate the prior information matrix, the method further comprises:
preprocessing the training set remote sensing images to unify color features of the training set remote sensing images;
filtering pictures with background specific gravity greater than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a first remote sensing image;
performing data enhancement on the first remote sensing image to obtain a second remote sensing image;
the loading prior information to the training set remote sensing image to generate a prior information matrix comprises:
and loading prior information to the second remote sensing image to generate a prior information matrix.
4. The method for identifying the satellite remote sensing image road as claimed in claim 1, wherein before the step of inputting the native remote sensing image of the target object into the trained road segmentation model to obtain the segmentation result remote sensing image, the method further comprises:
preprocessing the training set remote sensing images to unify color features of the training set remote sensing images;
filtering pictures with background specific gravity greater than preset specific gravity in the training set remote sensing images with uniform color characteristics to obtain a third remote sensing image;
performing data enhancement on the third remote sensing image to obtain a fourth remote sensing image;
and training a second basic network structure based on the fourth remote sensing image to obtain a road segmentation model.
5. The method for identifying the satellite remote sensing image road according to claim 1, wherein the obtaining of the backbone remote sensing image based on the prediction result remote sensing image comprises:
carrying out binarization processing on the remote sensing image of the prediction result;
and extracting a plurality of backbone data of the remote sensing image of the prediction result after binarization processing to obtain a backbone remote sensing image.
6. The method for identifying the satellite remote sensing image road according to claim 1, wherein the extending the target backbone data in the backbone remote sensing image to obtain a final remote sensing image comprises:
performing convolution operation on the backbone remote sensing image to obtain a characteristic remote sensing image;
restoring the size of the characteristic remote sensing image to the size of the native remote sensing image;
obtaining a plurality of endpoints in the reduced feature remote sensing image; the endpoint is a point with a pixel value of 11 in the characteristic remote sensing image;
based on a first recursive function, obtaining the length of a line segment where an endpoint in the characteristic remote sensing image is located;
screening out line segments of which the lengths of the line segments of the endpoints in the characteristic remote sensing image are greater than a preset length;
obtaining the gradient of the end point of the line segment with the length larger than the preset length;
based on a second recursive function, emitting rays along the direction of the gradient to obtain a final remote sensing image; wherein the ray is used for generating a road in the occluded characteristic remote sensing image; the gradient direction is the direction of the road.
7. A satellite remote sensing image road recognition device, its characterized in that, the device includes:
the first obtaining module is used for inputting the original remote sensing image of the target object into the trained road segmentation model so as to obtain a segmentation result remote sensing image;
the second obtaining module is used for splicing the three-channel matrix of the native remote sensing image and the remote sensing image of the segmentation result to obtain a remote sensing image of the splicing result;
the third obtaining module is used for inputting the splicing result remote sensing image into a trained road connection model so as to obtain a prediction result remote sensing image; the road connection model is obtained based on prior information training; the prior information comprises information for simulating that a road in the original remote sensing image is blocked so as to disconnect the road;
the identification module is used for identifying the road of the target object based on the prediction result remote sensing image;
wherein the identifying the road of the target object based on the remote sensing image of the prediction result comprises:
obtaining a backbone remote sensing image based on the prediction result remote sensing image; the backbone remote sensing image comprises a plurality of backbone data; the backbone data are used for representing data of a road in the prediction result remote sensing image;
extending target backbone data in the backbone remote sensing image to obtain a final remote sensing image;
and identifying the road of the target object based on the final remote sensing image.
8. A computer device, characterized in that it comprises a memory in which a computer program is stored and a processor which executes said computer program implementing the method according to any one of claims 1-6.
9. A computer-readable storage medium, having a computer program stored thereon, which, when executed by a processor, performs the method of any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211672238.XA CN115641512B (en) | 2022-12-26 | 2022-12-26 | Satellite remote sensing image road identification method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211672238.XA CN115641512B (en) | 2022-12-26 | 2022-12-26 | Satellite remote sensing image road identification method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115641512A CN115641512A (en) | 2023-01-24 |
| CN115641512B true CN115641512B (en) | 2023-04-07 |
Family
ID=84949781
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211672238.XA Active CN115641512B (en) | 2022-12-26 | 2022-12-26 | Satellite remote sensing image road identification method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115641512B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117612026B (en) * | 2023-11-24 | 2024-06-07 | 北京邮电大学 | Path prediction method and system based on remote sensing image road extraction |
| CN117939188A (en) * | 2024-01-25 | 2024-04-26 | 南京九才论信息科技有限公司 | Remote sensing image data downloading processing system based on communication link |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113850825A (en) * | 2021-09-27 | 2021-12-28 | 太原理工大学 | Remote sensing image road segmentation method based on context information and multi-scale feature fusion |
| CN114359143A (en) * | 2021-12-01 | 2022-04-15 | 武汉华中天经通视科技有限公司 | Remote sensing image road extraction method |
| CN114359120A (en) * | 2022-03-21 | 2022-04-15 | 深圳市华付信息技术有限公司 | Remote sensing image processing method, device, equipment and storage medium |
| CN114359859A (en) * | 2021-12-03 | 2022-04-15 | 南京佑驾科技有限公司 | Method and device for processing target object with shielding and storage medium |
| CN114494072A (en) * | 2022-01-28 | 2022-05-13 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN114758231A (en) * | 2022-04-07 | 2022-07-15 | 中山大学 | Remote sensing image occlusion processing method and device based on supervised contrast learning |
| CN114926797A (en) * | 2022-05-18 | 2022-08-19 | 中国地质大学（武汉） | Transformer double-branch road extraction method and device based on edge constraint and feature adaptation |
Family Cites Families (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956542B (en) * | 2016-04-28 | 2019-12-10 | 武汉大学 | High-resolution remote sensing image road extraction method based on statistical matching of structural wire harnesses |
| CN109086668B (en) * | 2018-07-02 | 2021-05-14 | 电子科技大学 | Unmanned aerial vehicle remote sensing image road information extraction method based on multi-scale generation countermeasure network |
| CN109145713B (en) * | 2018-07-02 | 2021-09-28 | 南京师范大学 | Small target semantic segmentation method combined with target detection |
| CN109145718A (en) * | 2018-07-04 | 2019-01-04 | 国交空间信息技术（北京）有限公司 | The road network extracting method and device of remote sensing image based on topology ambiguity |
| CN109493320B (en) * | 2018-10-11 | 2022-06-17 | 苏州中科天启遥感科技有限公司 | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment |
| CN110222615A (en) * | 2019-05-28 | 2019-09-10 | 上海海事大学 | The target identification method that is blocked based on InceptionV3 network |
| CN110287932B (en) * | 2019-07-02 | 2021-04-13 | 中国科学院空天信息创新研究院 | Road blocking information extraction method based on deep learning image semantic segmentation |
| CN110298321B (en) * | 2019-07-02 | 2021-03-23 | 中国科学院空天信息创新研究院 | Road blocking information extraction method based on deep learning image classification |
| CN110334693A (en) * | 2019-07-17 | 2019-10-15 | 中国电子科技集团公司第五十四研究所 | A kind of Remote Sensing Target sample generating method towards deep learning |
| CN110705381A (en) * | 2019-09-09 | 2020-01-17 | 北京工业大学 | Remote sensing image road extraction method and device |
| CN111259906B (en) * | 2020-01-17 | 2023-04-07 | 陕西师范大学 | Method for generating remote sensing image target segmentation countermeasures under condition containing multilevel channel attention |
| CN113496182B (en) * | 2020-04-08 | 2024-05-21 | 北京京东叁佰陆拾度电子商务有限公司 | Road extraction method and device based on remote sensing image, storage medium and equipment |
| CN112102274A (en) * | 2020-09-09 | 2020-12-18 | 昆明理工大学 | Small target semantic segmentation sun dark stripe detection method combined with target detection |
| CN112070037B (en) * | 2020-09-11 | 2022-09-30 | 中国科学院空天信息创新研究院 | Road extraction method, device, medium and equipment based on remote sensing image |
| CN112183258A (en) * | 2020-09-16 | 2021-01-05 | 太原理工大学 | Remote sensing image road segmentation method based on context information and attention mechanism |
| CN112733777B (en) * | 2021-01-18 | 2023-12-05 | 华侨大学 | Road extraction method, device and equipment of remote sensing image and storage medium |
| CN114693924A (en) * | 2022-03-14 | 2022-07-01 | 南京航空航天大学 | Road scene semantic segmentation method based on multi-model fusion |
-
2022
- 2022-12-26 CN CN202211672238.XA patent/CN115641512B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113850825A (en) * | 2021-09-27 | 2021-12-28 | 太原理工大学 | Remote sensing image road segmentation method based on context information and multi-scale feature fusion |
| CN114359143A (en) * | 2021-12-01 | 2022-04-15 | 武汉华中天经通视科技有限公司 | Remote sensing image road extraction method |
| CN114359859A (en) * | 2021-12-03 | 2022-04-15 | 南京佑驾科技有限公司 | Method and device for processing target object with shielding and storage medium |
| CN114494072A (en) * | 2022-01-28 | 2022-05-13 | 北京百度网讯科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN114359120A (en) * | 2022-03-21 | 2022-04-15 | 深圳市华付信息技术有限公司 | Remote sensing image processing method, device, equipment and storage medium |
| CN114758231A (en) * | 2022-04-07 | 2022-07-15 | 中山大学 | Remote sensing image occlusion processing method and device based on supervised contrast learning |
| CN114926797A (en) * | 2022-05-18 | 2022-08-19 | 中国地质大学（武汉） | Transformer double-branch road extraction method and device based on edge constraint and feature adaptation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115641512A (en) | 2023-01-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115641512B (en) | Satellite remote sensing image road identification method, device, equipment and medium | |
| CN110705405B (en) | Target labeling method and device | |
| CN110910343A (en) | Method and device for detecting pavement cracks and computer equipment | |
| CN109116129B (en) | Terminal detection method, detection device, system and storage medium | |
| CN109472193A (en) | Method for detecting human face and device | |
| US10726535B2 (en) | Automatically generating image datasets for use in image recognition and detection | |
| CN113205507B (en) | Visual question answering method, system and server | |
| CN110363176B (en) | Image analysis method and device | |
| CN114926511A (en) | High-resolution remote sensing image change detection method based on self-supervision learning | |
| CN113420871B (en) | Image quality evaluation method, image quality evaluation device, storage medium, and electronic device | |
| CN116363538B (en) | Bridge detection method and system based on unmanned aerial vehicle | |
| CN111067522A (en) | Brain addiction structural map assessment method and device | |
| CN114359658A (en) | Training data generation method and device, terminal equipment and storage medium | |
| CN117710756B (en) | Target detection and model training method, device, equipment and medium | |
| CN113033386A (en) | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system | |
| CN112560621A (en) | Identification method, device, terminal and medium based on animal image | |
| CN112087661A (en) | Video collection generation method, device, equipment and storage medium | |
| CN116109581A (en) | Concrete road monitoring method and device, electronic equipment and storage medium | |
| CN114626982A (en) | Scene reconstruction method and device, storage medium and computer equipment | |
| CN111243058A (en) | Object simulation image generation method and computer-readable storage medium | |
| CN112131418A (en) | Target labeling method, target labeling device and computer-readable storage medium | |
| CN114648706B (en) | Forest tree species identification method, device and equipment based on satellite remote sensing image | |
| CN113886745B (en) | Page picture testing method and device and electronic equipment | |
| CN116188998B (en) | Method, device, equipment and storage medium for identifying defects of overhead transmission line | |
| JP2024518696A (en) | Subdividing and removing objects from media items |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |