CN115100540B - Automatic road extraction method for high-resolution remote sensing image - Google Patents
Automatic road extraction method for high-resolution remote sensing image Download PDFInfo
- Publication number
- CN115100540B CN115100540B CN202210778771.8A CN202210778771A CN115100540B CN 115100540 B CN115100540 B CN 115100540B CN 202210778771 A CN202210778771 A CN 202210778771A CN 115100540 B CN115100540 B CN 115100540B
- Authority
- CN
- China
- Prior art keywords
- road
- extraction
- remote sensing
- module
- sensing image
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 113
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000013135 deep learning Methods 0.000 claims abstract description 13
- 238000004891 communication Methods 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 13
- 230000004927 fusion Effects 0.000 claims description 11
- 230000009471 action Effects 0.000 claims description 9
- 230000008447 perception Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 239000000725 suspension Substances 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 10
- 238000012545 processing Methods 0.000 abstract description 3
- 238000013461 design Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/52—Scale-space analysis, e.g. wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of video image processing, in particular to a method for automatically extracting a road of a high-resolution remote sensing image. According to the invention, the model is trained efficiently by the multi-Agent technology, and the high-resolution remote sensing image road extraction with high efficiency and high accuracy can be realized by utilizing the strong feature learning capability of deep learning.
Description
Technical Field
The invention relates to the technical field of video image processing, in particular to a method for automatically extracting a high-resolution remote sensing image road.
Background
The remote sensing technology is a large-range earth observation technology of ultra-long distance perception, and can provide important data support for the development of economy and society and the implementation of important national strategies by acquiring remote sensing images of an interested region and extracting and analyzing ground feature information. With the continuous development of remote sensing technology in recent years, remote sensing image data has features of high spatial resolution, high spectral resolution and high time resolution, and the high-resolution remote sensing images bring more abundant ground feature information, but increase the difficulty of acquiring accurate needed information from the high-resolution remote sensing images. Therefore, how to efficiently and accurately extract the needed ground object information from the high-resolution remote sensing image has become an important research direction.
The road is a typical object target with complex topological information in the remote sensing image, plays an important role in multiple scenes such as emergency response, traffic navigation and urban planning, and can provide priori knowledge for the identification of objects such as buildings, vegetation, rivers and the like in the remote sensing image scene understanding task. However, extracting road distribution from remote sensing images is a challenging task due to the diversity and complexity of road background environments in remote sensing images. The traditional road extraction method which relies on the artificial experience to design the feature extractor can obtain better extraction effect on the remote sensing image with relatively simple scene, but is difficult to cope with the interference caused by the homospectrum foreign matters and the homospectrum phenomenon, and is difficult to generate high-quality road extraction results to be applied in the actual scene. Especially, as the resolution of the remote sensing image is continuously improved, the effect of the traditional road extraction method without deep learning is poor due to interference factors such as illumination influence, tree and building shielding and the like.
Along with the cross combination of artificial intelligence and the remote sensing field, the deep learning method is applied to the recognition and extraction of the ground object targets of the remote sensing images and achieves better results. The convolutional neural network can automatically learn the rule of mapping the original input to the appointed label, the learning capability gradually replaces the traditional mode of designing the characteristics by means of manual experience, and the convolutional neural network becomes a main stream algorithm model in the image processing fields of image classification, target detection, semantic segmentation and the like, and provides a new thought for high-resolution remote sensing image road extraction. The current deep learning algorithm surrounding the high-resolution remote sensing image road extraction can be mainly divided into a semantic segmentation method based on pixels and a topology tracking method based on graph iteration. The former is characterized by predicting the category pixel by pixel, the principle is easy to operate but the efficiency is lower, the latter is characterized by taking the topological shape of the road as a main body, iteratively searching and constructing a road network, and the network model is relatively complex in design but better in road extraction effect. If the two methods can be further combined to realize advantage complementation, new breakthrough can be made in the field of high-resolution remote sensing image road extraction, and the landing of the road in practical application is promoted.
Disclosure of Invention
In order to solve the problems in the prior art, the invention aims to provide a method for automatically extracting the road of the high-resolution remote sensing image, which aims to solve the problems of low training efficiency and poor road extraction result of the current road extraction model of the high-resolution remote sensing image.
A method for automatically extracting a high-resolution remote sensing image road comprises the following steps:
acquiring a data set consisting of high-resolution remote sensing images with accurate road class labels;
preprocessing a high-resolution remote sensing image in a data set;
training a road extraction model:
Firstly, carrying out regional initialization on the preprocessed high-resolution remote sensing image to generate a preset number of generation agents, allowing adjacent regions to coincide, and randomly initializing the generation agents in the responsible regions to generate a plurality of extraction agents;
Each extraction Agent executes a road map iterative generation algorithm model based on deep learning to extract roads; in the operation period of the whole road map iteration generation algorithm model, information sharing is kept among all the extraction agents, generation agents and extraction agents in the same area and among the generation agents in different areas, and full-image road extraction of the high-resolution remote sensing image is completed cooperatively;
After each extraction Agent is terminated, the stored road extraction results are transmitted into a shared result library, and when all the extraction agents are terminated, all the road extraction results in the shared result library are combined to obtain a complete road extraction result diagram;
Carrying out loss calculation and back propagation on the road extraction result graph and the corresponding real label graph, and updating parameters of the road graph iteration generation algorithm model; and obtaining a final road extraction model after repeated training for many times; and realizing automatic extraction of the road based on the final road extraction model.
The invention introduces a multi-Agent technology based on the high-resolution remote sensing image road iterative generation algorithm of the current full-image starting point; in the multi-Agent system, the generation agents are responsible for generating a large number of extraction agents in regions on the image, each extraction Agent is used as a starting point of a road map iteration generation algorithm model, and road extraction is carried out in the region in each operation period; meanwhile, information sharing is kept among the extraction agents within a certain range, and road extraction work of the high-resolution remote sensing image is efficiently completed.
Preferably, the extraction Agent comprises a first sensing module, a first decision module, a first recording module, an action module and a first communication module;
The first perception module cuts a remote sensing image with a preset size by taking the position of the extracted Agent as the center, and transmits the cut remote sensing image to the first decision module;
The first decision module comprises a road map iterative generation algorithm model based on deep learning, and the road map iterative generation algorithm model outputs the position of the next pixel point belonging to the road category according to the cut remote sensing image and transmits the position to the action module;
The action module is used for guiding the extraction Agent to move to the predicted position of the next pixel point belonging to the road category;
the first recording module adds the predicted next pixel point position belonging to the road category to the separately maintained road map and adds corresponding edges;
the first communication module is used for guiding information sharing between the first communication module and other extraction agents in the same area and the area generation agents, and storing the road extraction result into a sharing result library when the first communication module is terminated.
Preferably, the preprocessing of the data includes performing data enhancement operation on the high-resolution remote sensing image and the corresponding road real label graph according to a predetermined probability, and uniformly adjusting the size of the high-resolution remote sensing image after the data enhancement operation to 512×512.
Preferably, the data enhancement operation includes performing saturation change, horizontal overturn, vertical overturn and the like on the high-resolution remote sensing image.
Preferably, each generation Agent is responsible for creating and monitoring management of the extraction Agent in the area;
The generation Agent comprises a second sensing module, a second communication module, a second recording module and a second decision module;
The second perception module receives the image information of the area, and takes the image information in the area as priori knowledge of creation and extraction agents;
The second communication module is responsible for communication with the generation Agent of the adjacent area and all the extraction agents in the area to realize information sharing;
The second recording module is responsible for recording the active states of all the extraction agents in the area, and when all the extraction agents are terminated, generating an Agent to enter a suspension state, and waiting for a next creation command;
the second decision module is responsible for random generation of the extraction Agent in the area and collision avoidance.
Preferably, the specific step of obtaining the position of the next pixel point belonging to the road category by the road map iterative generation algorithm model is as follows:
Extracting an Agent, taking the position as the center, intercepting a remote sensing image with H multiplied by W multiplied by 3, sending the remote sensing image into a coding network, obtaining a characteristic image of H multiplied by W multiplied by C after downsampling of the coding network, respectively sending the characteristic image into a parallel multi-scale cavity convolution fusion module and a spatial attention module to obtain first road characteristic information and second road characteristic information, merging the first road characteristic information and the second road characteristic information, and sending the first road characteristic information and the second road characteristic information into a decoding network; and obtaining the position of the next pixel point belonging to the road category through the decoding network.
Preferably, the training of the path graph iterative generation algorithm model constructs a loss function by combining binary cross entropy loss and Dice loss, as follows:
Loss()＝BCE_Loss()+Soft_Dice_Loss()。
Preferably, the multi-scale cavity convolution fusion module acquires the road feature information in a cascade mode of a plurality of convolution modules with different cavity convolution rates.
Preferably, the spatial attention module acquires global attention information of the road in the input image through a single cross spatial attention module; the calculation in the cross space attention module is as follows:
H′＝∑AiΦi+H；
Wherein: a i denotes the extracted attention feature, Φ i denotes the feature information extracted from the input feature map H, and finally, the cross attention feature-enhanced feature map H ′ is output.
Because the high-resolution remote sensing image contains more pixels and fewer pixels belong to a road, the design of the neural network model of the road map iterative algorithm model not only adopts an encoding-decoding structure, but also adds a multi-scale cavity convolution fusion module and a cross space attention module, which are parallel, thereby improving the extraction capability of the network model on road information.
The beneficial effects of the invention include:
The invention introduces a multi-Agent technology based on the high-resolution remote sensing image road iterative generation algorithm of the current full-image starting point; in the multi-Agent system, the generation agents are responsible for generating a large number of extraction agents in regions on the image, each extraction Agent is used as a starting point of a road map iteration generation algorithm model, and road extraction is carried out in the region in each operation period; meanwhile, information sharing is kept among the extraction agents within a certain range, and road extraction work of the high-resolution remote sensing image is efficiently completed.
Because the high-resolution remote sensing image contains more pixels and fewer pixels belong to a road, the design of the neural network model of the road map iterative algorithm model not only adopts an encoding-decoding structure, but also adds a multi-scale cavity convolution fusion module and a cross space attention module, which are parallel, thereby improving the extraction capability of the network model on road information.
The method for extracting the road of the high-resolution remote sensing image constructed by the invention is based on a topological tracking method based on graph iteration, and global road semantic segmentation prediction results are added as supplementary information when each iteration is explored, and the two are combined to improve the accuracy of road extraction. In addition, the existing high-resolution remote sensing image road extraction method based on graph iteration starts iterative exploration from a single starting point, and traverses the whole graph for a long time, so that model training efficiency is low. The invention creatively combines the multi-Agent technology and the road extraction method based on road map iteration generation, not only realizes the efficient training of the model by the multi-Agent technology, but also realizes the high-efficiency and high-accuracy road extraction of the high-resolution remote sensing image by utilizing the strong feature learning capability of deep learning.
Drawings
Fig. 1 is a basic flow chart of a high-resolution remote sensing image road automatic extraction method based on a multi-Agent technology and deep learning in the invention.
FIG. 2 is a diagram of a multi-Agent system design constructed in accordance with the present invention.
FIG. 3 is a diagram of a road map iterative algorithm model structure constructed by the invention and based on deep learning.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present application more apparent, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. The components of the embodiments of the present application generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the application, as presented in the figures, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present application.
Embodiments of the present invention are described in further detail below with reference to FIG. 1:
A method for automatically extracting a high-resolution remote sensing image road comprises the following steps:
acquiring a data set consisting of high-resolution remote sensing images with accurate road class labels;
Preprocessing a high-resolution remote sensing image in a data set; the preprocessing of the data comprises the steps of carrying out data enhancement operation on the high-resolution remote sensing image and a corresponding road real label graph according to a preset probability, and uniformly adjusting the size of the high-resolution remote sensing image subjected to the data enhancement operation to 512 multiplied by 512; the data enhancement operation comprises saturation change, horizontal overturn, vertical overturn and the like of the high-resolution remote sensing image.
Training a road extraction model:
Firstly, carrying out regional initialization on the preprocessed high-resolution remote sensing image to generate a preset number of generation agents, allowing adjacent regions to coincide, and randomly initializing the generation agents in the responsible regions to generate a plurality of extraction agents;
Each extraction Agent executes a road map iterative generation algorithm model based on deep learning to extract roads; in the operation period of the whole road map iteration generation algorithm model, information sharing is kept among all the extraction agents, generation agents and extraction agents in the same area and among the generation agents in different areas, and full-image road extraction of the high-resolution remote sensing image is completed cooperatively; after each extraction Agent is terminated, the stored road extraction results are transmitted into a shared result library, and when all the extraction agents are terminated, all the road extraction results in the shared result library are combined to obtain a complete road extraction result diagram;
Carrying out loss calculation and back propagation on the road extraction result graph and the corresponding real label graph, and updating parameters of the road graph iteration generation algorithm model; and obtaining a final road extraction model after repeated training for many times; and realizing automatic extraction of the road based on the final road extraction model.
The invention introduces a multi-Agent technology based on the high-resolution remote sensing image road iterative generation algorithm of the current full-image starting point; in the multi-Agent system, the generation agents are responsible for generating a large number of extraction agents in regions on the image, each extraction Agent is used as a starting point of a road map iteration generation algorithm model, and road extraction is carried out in the region in each operation period; meanwhile, information sharing is kept among the extraction agents within a certain range, and road extraction work of the high-resolution remote sensing image is efficiently completed.
Referring to fig. 2, the extraction Agent includes a first sensing module, a first decision module, a first recording module, an action module, and a first communication module;
The first perception module cuts a remote sensing image with a preset size by taking the position of the extracted Agent as the center, and transmits the cut remote sensing image to the first decision module;
The first decision module comprises a road map iterative generation algorithm model based on deep learning, and the road map iterative generation algorithm model outputs the position of the next pixel point belonging to the road category according to the cut remote sensing image and transmits the position to the action module;
The action module is used for guiding the extraction Agent to move to the predicted position of the next pixel point belonging to the road category;
The first recording module adds the predicted next pixel point position belonging to the road category to the separately maintained road map and adds corresponding edges; considering that the initial position of the extraction Agent is not necessarily on the road when being generated, the initial position is not added into the road map;
the first communication module is used for guiding information sharing between the first communication module and other extraction agents in the same area and the area generation agents, and storing the road extraction result into a sharing result library when the first communication module is terminated.
Referring to fig. 2, each of the generating agents is responsible for creation and monitoring management of the extracting Agent in the area;
The generation Agent comprises a second sensing module, a second communication module, a second recording module and a second decision module;
The second perception module receives the image information of the area, and takes the image information in the area as priori knowledge of creation and extraction agents;
The second communication module is responsible for communication with the generation Agent of the adjacent area and all the extraction agents in the area to realize information sharing;
the second recording module is responsible for recording the active states of all the extraction agents in the area, and when all the extraction agents are terminated, an Agent entering suspension state is generated and a next creation command is waited;
the second decision module is responsible for random generation of the extracted agents in the area and collision avoidance;
the above process is repeated in each generation Agent run period until all extraction agents in the region are terminated.
Referring to fig. 3, the specific steps of obtaining the position of the next pixel point belonging to the road category by the road map iterative generation algorithm model are as follows:
Extracting an Agent, taking the position as the center, intercepting a remote sensing image with H multiplied by W multiplied by 3, sending the remote sensing image into a coding network, obtaining a characteristic image of H multiplied by W multiplied by C after downsampling of the coding network, respectively sending the characteristic image into a parallel multi-scale cavity convolution fusion module and a spatial attention module to obtain first road characteristic information and second road characteristic information, merging the first road characteristic information and the second road characteristic information, and sending the first road characteristic information and the second road characteristic information into a decoding network; and obtaining the position of the next pixel point belonging to the road category through the decoding network.
Considering the problem of sample unbalance between a road and a background in a training sample, the training of the road map iterative generation algorithm model constructs a loss function by combining binary cross entropy loss and Dice loss, as follows:
Loss()＝BCE_Loss()+Soft_Dice_Loss()。
The multi-scale cavity convolution fusion module acquires road feature information by adopting a cascade mode of a plurality of convolution modules with different cavity convolution rates.
The space attention module acquires global attention information of a road in an input image through two cross space attention modules; the calculation in the cross space attention module is as follows:
H′＝∑AiΦi+H；
Wherein: a i denotes the extracted attention feature, Φ i denotes the feature information extracted from the input feature map H, and finally, the cross attention feature-enhanced feature map H ′ is output.
The multi-scale cavity convolution fusion module and the spatial attention module are combined, so that the defect that only local information is focused in the existing method is overcome, and the multi-scale cavity convolution fusion module and the spatial attention module are combined to improve the extraction effect of road features.
Because the high-resolution remote sensing image contains more pixels and fewer pixels belong to a road, the design of the neural network model of the road map iterative algorithm model not only adopts an encoding-decoding structure, but also adds a multi-scale cavity convolution fusion module and a cross space attention module, which are parallel, thereby improving the extraction capability of the network model on road information.
The above examples merely illustrate specific embodiments of the application, which are described in more detail and are not to be construed as limiting the scope of the application. It should be noted that it is possible for a person skilled in the art to make several variants and modifications without departing from the technical idea of the application, which fall within the scope of protection of the application.
Claims (7)
1. The method for automatically extracting the high-resolution remote sensing image road is characterized by comprising the following steps of:
acquiring a data set consisting of high-resolution remote sensing images with accurate road class labels;
preprocessing a high-resolution remote sensing image in a data set;
training a road extraction model:
Firstly, carrying out regional initialization on the preprocessed high-resolution remote sensing image to generate a preset number of generation agents, allowing adjacent regions to coincide, and randomly initializing the generation agents in the responsible regions to generate a plurality of extraction agents;
Each extraction Agent executes a road map iterative generation algorithm model based on deep learning to extract roads; in the operation period of the whole road map iteration generation algorithm model, information sharing is kept among all the extraction agents, generation agents and extraction agents in the same area and among the generation agents in different areas, and full-image road extraction of the high-resolution remote sensing image is completed cooperatively;
After each extraction Agent is terminated, the stored road extraction results are transmitted into a shared result library, and when all the extraction agents are terminated, all the road extraction results in the shared result library are combined to obtain a complete road extraction result diagram;
carrying out loss calculation and back propagation on the road extraction result graph and the corresponding real label graph, and updating parameters of the road graph iteration generation algorithm model; and obtaining a final road extraction model after repeated training for many times; realizing automatic extraction of the road based on a final road extraction model;
The extraction Agent comprises a first perception module, a first decision module, a first recording module, an action module and a first communication module;
The first perception module cuts a remote sensing image with a preset size by taking the position of the extracted Agent as the center, and transmits the cut remote sensing image to the first decision module;
The first decision module comprises a road map iterative generation algorithm model based on deep learning, and the road map iterative generation algorithm model outputs the position of the next pixel point belonging to the road category according to the cut remote sensing image and transmits the position to the action module;
The action module is used for guiding the extraction Agent to move to the predicted position of the next pixel point belonging to the road category;
the first recording module adds the predicted next pixel point position belonging to the road category to the separately maintained road map and adds corresponding edges;
the first communication module is used for guiding information sharing between the first communication module and other extraction agents in the same area and the area generation agents, and storing a road extraction result into a sharing result library when the first communication module is terminated;
Each generation Agent is responsible for the creation and monitoring management of the extraction Agent in the area;
The generation Agent comprises a second sensing module, a second communication module, a second recording module and a second decision module;
The second perception module receives the image information of the area, and takes the image information in the area as priori knowledge of creation and extraction agents;
The second communication module is responsible for communication with the generation Agent of the adjacent area and all the extraction agents in the area to realize information sharing;
The second recording module is responsible for recording the active states of all the extraction agents in the area, and when all the extraction agents are terminated, generating an Agent to enter a suspension state, and waiting for a next creation command;
the second decision module is responsible for random generation of the extraction Agent in the area and collision avoidance.
2. The method for automatically extracting the road from the high-resolution remote sensing image according to claim 1, wherein the preprocessing of the data comprises the steps of performing data enhancement operation on the high-resolution remote sensing image and a corresponding road real label graph according to a preset probability, and uniformly adjusting the size of the high-resolution remote sensing image after the data enhancement operation to 512×512.
3. The method of claim 2, wherein the data enhancement operation includes saturation change, horizontal flipping, and vertical flipping of the high resolution remote sensing image.
4. The method for automatically extracting the road from the high-resolution remote sensing image according to claim 1, wherein the specific step of obtaining the position of the next pixel belonging to the road class by the road map iterative generation algorithm model is as follows:
Extracting an Agent, taking the position as the center, intercepting a remote sensing image with H multiplied by W multiplied by 3, sending the remote sensing image into a coding network, obtaining a characteristic image of H multiplied by W multiplied by C after downsampling of the coding network, respectively sending the characteristic image into a parallel multi-scale cavity convolution fusion module and a spatial attention module to obtain first road characteristic information and second road characteristic information, merging the first road characteristic information and the second road characteristic information, and sending the first road characteristic information and the second road characteristic information into a decoding network; and obtaining the position of the next pixel point belonging to the road category through the decoding network.
5. The method for automatically extracting a road from a high-resolution remote sensing image according to claim 1, wherein the training of the road map iterative generation algorithm model constructs a loss function by combining binary cross entropy loss and Dice loss, as follows:
Loss()＝BCE_Loss()+Soft_Dice_Loss()。
6. The method for automatically extracting the road from the high-resolution remote sensing image according to claim 4, wherein the multi-scale cavity convolution fusion module acquires the road feature information by adopting a mode of cascade connection of a plurality of convolution modules with different cavity convolution rates.
7. The method for automatically extracting a road from a high-resolution remote sensing image according to claim 4, wherein the spatial attention module obtains global attention information of the road in the input image through two cross spatial attention modules; the calculation in the cross space attention module is as follows:
H′＝∑AiΦi+H；
Wherein: a i denotes the extracted attention feature, Φ i denotes the feature information extracted from the input feature map H, and finally, the cross attention feature-enhanced feature map H ′ is output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210778771.8A CN115100540B (en) | 2022-06-30 | 2022-06-30 | Automatic road extraction method for high-resolution remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210778771.8A CN115100540B (en) | 2022-06-30 | 2022-06-30 | Automatic road extraction method for high-resolution remote sensing image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115100540A CN115100540A (en) | 2022-09-23 |
| CN115100540B true CN115100540B (en) | 2024-05-07 |
Family
ID=83294221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210778771.8A Active CN115100540B (en) | 2022-06-30 | 2022-06-30 | Automatic road extraction method for high-resolution remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115100540B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105007191A (en) * | 2015-07-31 | 2015-10-28 | 上海红神信息技术有限公司 | Perception, decision and execution separation method for pseudo-route switching system |
| CN107480726A (en) * | 2017-08-25 | 2017-12-15 | 电子科技大学 | A kind of Scene Semantics dividing method based on full convolution and shot and long term mnemon |
| CN107729922A (en) * | 2017-09-20 | 2018-02-23 | 千寻位置网络有限公司 | Remote sensing images method for extracting roads based on deep learning super-resolution technique |
| WO2018132607A2 (en) * | 2017-01-12 | 2018-07-19 | Mobileye Vision Technologies Ltd. | Navigation based on vehicle activity |
| CN110659342A (en) * | 2019-08-21 | 2020-01-07 | 北京吉威数源信息技术有限公司 | Rule base automatic drawing comprehensive processing method and system based on information extraction features |
| WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
| CN113591608A (en) * | 2021-07-12 | 2021-11-02 | 浙江大学 | High-resolution remote sensing image impervious surface extraction method based on deep learning |
| CN114581748A (en) * | 2022-05-06 | 2022-06-03 | 南京大学 | Multi-agent perception fusion system based on machine learning and implementation method thereof |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8103435B2 (en) * | 2007-07-27 | 2012-01-24 | George Mason Intellectual Properties, Inc. | Near real-time traffic routing |
-
2022
- 2022-06-30 CN CN202210778771.8A patent/CN115100540B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105007191A (en) * | 2015-07-31 | 2015-10-28 | 上海红神信息技术有限公司 | Perception, decision and execution separation method for pseudo-route switching system |
| WO2018132607A2 (en) * | 2017-01-12 | 2018-07-19 | Mobileye Vision Technologies Ltd. | Navigation based on vehicle activity |
| CN107480726A (en) * | 2017-08-25 | 2017-12-15 | 电子科技大学 | A kind of Scene Semantics dividing method based on full convolution and shot and long term mnemon |
| CN107729922A (en) * | 2017-09-20 | 2018-02-23 | 千寻位置网络有限公司 | Remote sensing images method for extracting roads based on deep learning super-resolution technique |
| CN110659342A (en) * | 2019-08-21 | 2020-01-07 | 北京吉威数源信息技术有限公司 | Rule base automatic drawing comprehensive processing method and system based on information extraction features |
| WO2021184891A1 (en) * | 2020-03-20 | 2021-09-23 | 中国科学院深圳先进技术研究院 | Remotely-sensed image-based terrain classification method, and system |
| CN113591608A (en) * | 2021-07-12 | 2021-11-02 | 浙江大学 | High-resolution remote sensing image impervious surface extraction method based on deep learning |
| CN114581748A (en) * | 2022-05-06 | 2022-06-03 | 南京大学 | Multi-agent perception fusion system based on machine learning and implementation method thereof |
Non-Patent Citations (4)
| Title |
|---|
| Road Extraction Methods in High-Resolution Remote Sensing Images: A Comprehensive Review;Renbao Lian等;《IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing》;20200911;全文 * |
| 基于深度学习的高分辨率遥感影像道路提取方法研究;王琪;《万方数据》;20221202;全文 * |
| 多特征约束的高分辨率光学遥感影像道路提取;戴激光;杜阳;方鑫鑫;王杨;苗志鹏;;遥感学报;20180925(05);全文 * |
| 高空间分辨率遥感影像的多智能体分割方法研究;赵贝;钟燕飞;张良培;;测绘学报;20130215(01);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115100540A (en) | 2022-09-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Veillette et al. | Sevir: A storm event imagery dataset for deep learning applications in radar and satellite meteorology | |
| Khalel et al. | Automatic pixelwise object labeling for aerial imagery using stacked u-nets | |
| CN111046768B (en) | Deep learning method for simultaneously extracting road pavement and center line of remote sensing image | |
| CN111814683A (en) | Robust visual SLAM method based on semantic prior and deep learning features | |
| CN113344932B (en) | Semi-supervised single-target video segmentation method | |
| CN113076871A (en) | Fish shoal automatic detection method based on target shielding compensation | |
| CN110879960B (en) | Method and computing device for generating image data set for convolutional neural network learning | |
| CN102799900A (en) | Target tracking method based on supporting online clustering in detection | |
| WO2022141718A1 (en) | Method and system for assisting point cloud-based object detection | |
| Chen et al. | NIGAN: A framework for mountain road extraction integrating remote sensing road-scene neighborhood probability enhancements and improved conditional generative adversarial network | |
| CN114549537A (en) | Unstructured environment point cloud semantic segmentation method based on cross-modal semantic enhancement | |
| CN114943876A (en) | Cloud and cloud shadow detection method and device for multi-level semantic fusion and storage medium | |
| Xiong et al. | Contrastive learning for automotive mmWave radar detection points based instance segmentation | |
| CN115860102B (en) | Pre-training method, device, equipment and medium for automatic driving perception model | |
| CN110378284B (en) | Road front view generation method and device, electronic equipment and storage medium | |
| Xu et al. | Semantic segmentation of sparsely annotated 3D point clouds by pseudo-labelling | |
| Cui et al. | A graph-based dual convolutional network for automatic road extraction from high resolution remote sensing images | |
| Xiu et al. | Dynamic-scale graph convolutional network for semantic segmentation of 3d point cloud | |
| CN115100540B (en) | Automatic road extraction method for high-resolution remote sensing image | |
| Lei et al. | Automatic identification of street trees with improved RandLA-Net and accurate calculation of shading area with density-based iterative α-shape | |
| Jiang et al. | RailSeg: Learning Local-Global Feature Aggregation with Contextual Information for Railway Point Cloud Semantic Segmentation | |
| Hasecke et al. | Fake it, mix it, segment it: Bridging the domain gap between lidar sensors | |
| CN116503602A (en) | Unstructured environment three-dimensional point cloud semantic segmentation method based on multi-level edge enhancement | |
| Zao et al. | Topology-Guided Road Graph Extraction From Remote Sensing Images | |
| Mao et al. | City object detection from airborne Lidar data with OpenStreetMap‐tagged superpixels |
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 |