CN114022784A - Method and device for screening landmark control points - Google Patents
Method and device for screening landmark control points Download PDFInfo
- Publication number
- CN114022784A CN114022784A CN202111323543.3A CN202111323543A CN114022784A CN 114022784 A CN114022784 A CN 114022784A CN 202111323543 A CN202111323543 A CN 202111323543A CN 114022784 A CN114022784 A CN 114022784A
- Authority
- CN
- China
- Prior art keywords
- landmark
- control point
- image
- landmark control
- control points
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a method and a device for screening landmark control points, and belongs to the technical field of remote sensing data processing. The method comprises the steps of detecting an initial landmark control point in a high-positioning-precision reference image through a landmark detection network model, primarily selecting the initial landmark control point by taking whether the characteristics are obvious as a screening standard, and evaluating the quality of the landmark control point by constructing a landmark control point selection network model and realizing the selection of the landmark control point according to the characteristics of an image to be processed. The invention solves the problems that the control information is difficult to transmit and the control information transmission precision is low because the matching degree of the landmark control point and the image to be processed is not concerned in the prior art.
Description
Technical Field
The invention relates to the technical field of remote sensing data processing, in particular to a method and a device for screening landmark control points.
Background
In the field of remote sensing, the landmark control point refers to a ground control point which is established by acquiring a typical earth surface feature from ubiquitous geospatial information by an intelligent means to serve as a landmark, expanding or replacing a traditional control point by a landmark control point set and providing the landmark control point set for multi-source remote sensing platforms such as spaceborne platforms and airborne platforms (including unmanned aerial vehicles) for ground control, and has knowledge storage and intelligent service capability.
In the field of remote sensing mapping, uncontrolled positioning and controlled positioning are not contradictory, and the advantages and the values of the positioning are not replaceable. The high-precision ground control point has very important significance for improving the positioning precision of the earth observation image. At present, the ground control points are mostly obtained by manually laying or performing feature extraction on the reference image. The ground control points obtained by the manual layout mode are fixed in area, high in layout cost, required to be maintained regularly and controlled to measure on site, required to be manually selected on the image to be processed during application, and large in workload. The basic image feature extraction mode adopts artificially designed features, is limited by the quality of a reference image, needs to match the reference image with an image to be processed when in application, has higher matching difficulty due to differences of image sources, resolution, imaging environments and the like, and is not beneficial to generating and applying ground control points in a global range.
Disclosure of Invention
In view of the above, the present invention provides a method and an apparatus for screening landmark control points, and mainly aims to solve the problems in the prior art that the method for acquiring the landmark control points only focuses on the quality of the landmark image block itself, and the matching degree between the landmark control points and the to-be-processed image itself is not focused, so that the transmission of control information is difficult, and the adjustment accuracy is relatively low.
According to one aspect of the invention, a method for screening landmark control points is provided, which comprises the following steps: step 101: inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database; step 102: carrying out primary selection on the initial landmark control points in the landmark metadata base, and selecting the initial landmark control points with obvious characteristics and convenient control information transmission to form a primary selection landmark control point set; step 103: inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed.
As a further improvement of the invention, the landmark control point selection network model is obtained by training through the following method: acquiring a satellite image to be trained and a corresponding initially selected landmark control point; marking the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained according to the actual matching result of the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained; and taking the marked satellite image to be trained and the initially selected landmark control point as sample data, and training the landmark control point selection network model according to the sample data and a deep learning technology.
As a further improvement of the present invention, the landmark control point fine-selection network model specifically comprises: the multilayer convolutional neural network is used for extracting image characteristics of the satellite image and the initially selected landmark control points in the sample data; the attribute feature extraction network is used for extracting attribute features of the satellite images and the initially selected landmark control points in the sample data, and the attribute features comprise: resolution, imaging load, attitude and orbit state; and the full convolution network generates a characteristic vector according to the image characteristic and the attribute characteristic, inputs the characteristic vector into the full convolution network to obtain a response value, and the response value represents the matching degree of the satellite image and the initially selected landmark control point in the sample data.
As a further improvement of the present invention, the landmark metadata database is iteratively updated during the initial selection, specifically: for the same global mesh generation mesh, replacing the landmark metadata with the initially selected landmark control points which are acquired during initial selection and are better than the original landmark metadata in the mesh, and iteratively realizing optimization of a landmark metadata database.
As a further improvement of the invention, the preliminary selected indexes comprise: the landmark image resolution, the landmark image positioning precision, the landmark image state and the landmark distribution; and comprehensively scoring the indexes to evaluate the advantages and disadvantages of the initial landmark control points.
According to another aspect of the present invention, there is provided a landmark control point screening apparatus including: landmark metadata database module: the landmark detection network model is used for inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database; a primary selection module: the initial landmark control points in the landmark metadata base are initially selected, the initial landmark control points with obvious characteristics and convenient control information transmission are selected, and an initially selected landmark control point set is formed; a selection module: the network model is used for inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed.
As a further improvement of the invention, the landmark control point selection network model is obtained by training through the following method: acquiring a satellite image to be trained and a corresponding initially selected landmark control point; marking the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained according to the actual matching result of the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained; and taking the marked satellite image to be trained and the initially selected landmark control point as sample data, and training the landmark control point selection network model according to the sample data and a deep learning technology.
As a further improvement of the present invention, the landmark control point fine-selection network model specifically comprises: the multilayer convolutional neural network is used for extracting image characteristics of the satellite image and the initially selected landmark control points in the sample data; the attribute feature extraction network is used for extracting attribute features of the satellite images and the initially selected landmark control points in the sample data, and the attribute features comprise: resolution, imaging load, attitude and orbit state; and the full convolution network generates a characteristic vector according to the image characteristic and the attribute characteristic, inputs the characteristic vector into the full convolution network to obtain a response value, and the response value represents the matching degree of the satellite image and the initially selected landmark control point in the sample data.
By the technical scheme, the beneficial effects provided by the invention are as follows:
(1) the method comprises the steps of obtaining an image with high positioning precision as a reference image, detecting an initial landmark control point in the reference image through a landmark detection network model, primarily selecting the initial landmark control point by taking whether the characteristic is obvious as a screening standard, deleting the initial landmark control point which has low resolution, small image block and is shielded or damaged, leaving the initial landmark control point with obvious characteristic, conveniently controlling information transmission, forming a primarily selected landmark control point set, and ensuring the effectiveness of the primarily selected landmark control point in positioning.
(2) And storing the detected landmark image block, landmark attributes, position information, three-dimensional model information, projection information and the like into a database according to the format of a landmark metadata system to obtain initial landmark control points, continuously accumulating and storing a large number of initial landmark control points in the using process to form a landmark metadata database, and realizing the generation and application of a landmark control point set in the global scope by setting the landmark metadata system and storing the various information.
(3) The matching results of different images to be processed and corresponding landmark control points are evaluated by constructing a landmark control point selection network model, and a feedback mechanism is established in the application process of the landmark control points to realize further iterative updating of the landmark control points and realize selection of the landmark control points.
The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
fig. 1 is a flowchart illustrating a method for filtering landmark control points according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of a landmark control point refinement network model provided in an embodiment of the present specification.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
The selection of the proper landmark control point has very important significance for improving the positioning quality of the remote sensing image, the screening of abnormal landmarks is mainly considered in the conventional landmark control point screening mode, and the optimization is performed according to the principles of high spatial resolution, high positioning accuracy, landmark size and the like. However, due to differences in image sources, resolutions, imaging environments, and the like, even if the filtering is performed based on the quality of the landmark control points themselves, there is still a problem that the matching effect between the images to be processed and the landmark control points themselves is poor. In this case, in order to realize effective application of the landmark control points, it is necessary to improve the matching degree between the image to be processed and the selected landmark control points. Based on this, an embodiment of the present specification provides a landmark control point screening method, which adopts a landmark control point screening method based on reinforcement learning to solve the problem.
Fig. 1 is a schematic flowchart of a method for screening landmark control points according to an embodiment of the present disclosure, and as shown in fig. 1, the flowchart includes:
step 101: inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database;
step 102: carrying out primary selection on the initial landmark control points in the landmark metadata base, and selecting the initial landmark control points with obvious characteristics and convenient control information transmission to form a primary selection landmark control point set;
step 103: inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed.
In an embodiment of the present specification, an image with high positioning accuracy is first obtained as a reference image, then an initial landmark control point in the reference image is detected through a landmark detection network model, whether the characteristics are obvious is used as a screening standard to initially select the initial landmark control point, the initial landmark control point with low resolution, small image block and occlusion or damage is deleted, the initial landmark control point with obvious characteristics is left, which is convenient for controlling information transmission, and an initially selected landmark control point set is formed. And inputting the images to be processed into the landmark control point selection network model, and selecting the landmark image blocks with higher matching degree from the initially selected landmark image block set as the landmark control points.
In the step 102, the initial landmark control points are divided into two-dimensional initial landmark control points and three-dimensional initial landmark control points, and the two-dimensional initial landmark control points are landmark image blocks obtained by performing landmark detection on high-precision images; the three-dimensional landmark control points have two forms, one is the detection of an oblique image, and the image is the landmark image block; and the other method also needs to acquire the three-dimensional structure information of the landmark according to the detection result of the image to obtain a three-dimensional landmark block. The manner of generating the three-dimensional landmark control points is described in detail in "method and apparatus for generating and applying three-dimensional landmark control points", and is not described in detail in patent document No. 202010322946.5.
In this embodiment, a landmark detection network model is trained, and the landmark detection network model is obtained by training remote sensing image sample data based on a plurality of landmarks.
Optionally, the landmark detection network model may be obtained by training remote sensing image sample data of the created landmark based on a deep learning target detection and identification method. The target detection and identification method can be mainly divided into two categories, namely a target detection and identification method based on traditional image processing and machine learning algorithm and a target detection and identification method based on deep learning according to development. The target detection and identification method based on the traditional image processing and machine learning algorithm can be mainly expressed as follows: target feature extraction, target identification and target positioning. The target features used herein are often artificially designed features such as SIFT, HOG features, etc. And identifying the target through the target characteristics, and then positioning the target by combining a corresponding strategy.
The target detection and identification method based on deep learning can be mainly expressed as follows: extracting depth features of the image and identifying and positioning the target based on the depth neural network. The method can achieve good effect in target intelligent recognition and reduce manual intervention. Typical target detection and identification methods based on deep learning include fast RCNN, YOLO series, AttentionNet, ComerNet-Lite, and the like.
Optionally, in order to improve the recognition accuracy of the landmark recognition model, the sample data of the remote sensing image of the landmark includes at least one of the following:
the method comprises the steps that landmarks in a reference satellite image database are classified according to preset landmark types to obtain landmark sample data, wherein the reference satellite image database is obtained by observing high-precision space satellite images;
the landmark data set is created in a mode of manually or automatically generating labels;
and landmark sample data acquired through network open source data.
In the step 102, the detected landmark image blocks, landmark attributes, position information, three-dimensional model information, projection information and the like are stored in a database according to the format of a landmark metadata system to obtain initial landmark control points, and a large number of initial landmark control points are continuously accumulated and stored in the using process to form a landmark metadata database.
Specifically, the design of the landmark metadata system is the basis of landmark data storage, extension and application. By setting a landmark metadata system and storing the multiple information, the generation and the application of landmark control points in the global range are realized; preferably, the format of the landmark control point metadata hierarchy may be:
TABLE 1 Landmark control point metadata Format
After the landmark metadata base is generated, the step S103 is executed to perform preliminary screening on the initial landmark control points stored in the landmark metadata base, so as to obtain an initial landmark control point set.
The initial landmark control points obtained through the steps S101 and S102 are obtained through intelligent detection by a landmark detection network model, and the initial landmark control points obtained through detection may not necessarily be effective landmark control points, and the description is given by taking a road intersection as an example, and some of the initial landmark control points obtained through detection by the landmark detection network model are distributed too densely; some display on the image is too small due to resolution; some trees are shielded and damaged. Therefore, the initial landmark control points obtained by the initial detection need to be preliminarily screened.
Specifically, the landmark control points are mainly constructed to provide high-precision spatial reference for remote sensing image data with weak positioning precision, so that the preliminarily screened object is the landmark with obvious screening characteristics and convenient control information transmission. The primary screening principle mainly comprises the following steps: landmark image resolution, landmark image positioning accuracy, landmark image states (mainly including presence or absence of shielding, damage and the like), and landmark distribution; and comprehensively scoring each index to evaluate the quality of the initial landmark control point.
The distribution of the landmarks can be determined through global mesh division, and the size of the meshes and the number of the landmarks in each mesh are controlled to confirm. The technical solution of global mesh generation is disclosed in the patent application No. 201910963442.9 entitled "method and apparatus for constructing intelligent landmark control net", which is not repeated herein. Preferably, when the landmark metadata base is enriched in step 102, for the same grid, if a better initially selected landmark control point is obtained when the initial landmark control point is initially screened, the original initially selected landmark control point in the previous grid may be replaced, and the iteration is performed in this way, so as to realize the continuous update of the landmark control network.
The determination of the landmark image state (mainly including whether occlusion exists or not, damage and the like) can be realized by a change detection method, and the confirmation of the landmark state is realized by comparing with historical image data of the same region and integrating the image gray scale characteristics of the point. If the landmark is damaged, the landmark can be found according to the comparison with the historical image (change detection); when the landmark is shielded by the tree, the expressions of the shielded landmark and the unshielded landmark state gray feature and the like are inconsistent for the image of the same scene, and the filtering of the landmark control point is realized by comparing the gray features.
According to the step 103, the primary screening of the landmark control points is realized, and the effectiveness of the primary selection of the landmark control points in positioning is ensured. However, in the practical application process, the landmark control points have the characteristic of multiple sources, the accuracy of control information transmission depends on the image matching effect, and the landmark image blocks with the characteristics of high resolution, high positioning accuracy, no occlusion and the like do not necessarily realize the high-accuracy transmission of the control information, which has a direct relationship with the characteristics of the image to be processed and the image matching method. The characteristics of the images can be changed due to different imaging environments, and theoretically, the more similar the imaging environments, the better the image matching effect is; for the image matching method, different matching methods are adapted to different images. In order to realize effective application of the landmark control points, the selection of the landmark control points needs to be realized in an interactive mode, that is, a feedback mechanism is established in the application process of the landmark control points to realize further iterative update of the landmark control points, so as to realize the selection of the landmark control points.
In order to realize the above-mentioned landmark control point selection, in this embodiment, a landmark control point selection network model is trained, and the landmark control point selection network model is trained based on a large amount of satellite image sample data.
Fig. 2 is a schematic structural diagram of a landmark control point selection network model according to an embodiment of the present disclosure, where the goal of constructing the landmark control point selection network model is to realize the screening of the optimal landmark control points by evaluating the matching results between different images to be processed and corresponding landmark control points. As shown in fig. 2, inputting a satellite image in sample data and one or more initially selected landmark control points labeled thereto, and extracting image features of the satellite image and the initially selected landmark control points in the sample data through a multilayer convolutional neural network, respectively, and extracting attribute features of the satellite image and the initially selected landmark control points in the sample data, such as resolution, imaging load, attitude and orbit state, respectively, by an attribute feature extraction network; extracting feature vectors from the extracted image features and attribute features, inputting the feature vectors into a full convolution network, and giving a large response value q to the adapted to-be-processed image and the landmark control point image and a small response value q to the unadapted image as shown in fig. 2; and taking the satellite image in the sample data and the matching result true of the initial landmark control point corresponding to the satellite image as the most important reference for feedback, feeding back high return for the image to be processed with the excellent matching result true effect and the landmark control point image, and feeding back low return for the poor effect.
For images to be processed from different sources, because image characteristics (including imaging environment, gray scale characteristics, attitude and orbit states, etc.) are different, when image matching is performed to realize control information transfer, a landmark control point with similar characteristics needs to be adapted to the images, which is different from a conventional network.
Through the landmark control point selection network model and the feedback mechanism thereof, samples can be continuously accumulated in the practical application process of the landmark control point, and the training of the landmark control point selection network model is realized. In the actual use process, the landmark control points matched with the images to be processed can be optimized through the reinforcement learning network. In the process of earlier application, landmark control points with the same sources as possible can be selected according to experience to transmit control information for the image to be processed; as the amount of data accumulates, the landmark control point refinement network becomes increasingly intelligent. Can realize that: firstly, the landmark control points with poor matching effect or poor distribution can be removed, and the selection of the landmark control points is realized; secondly, the landmark control points can be automatically selected according to the images to be processed to transmit control information, and high precision can be ensured.
The training process can be represented by an expression Q (Input, action, true, w), where Input represents a satellite image and an initial landmark control point corresponding to the satellite image in Input sample data, action represents different image matching algorithms, true represents a matching result of the satellite image and an initial landmark control point labeled to the satellite image in the sample data, and w represents a parameter value set constituting the whole landmark control point selection network model. The feedback process is to compare the adaptation degree Q value obtained by the calculation of the matching result true value and the landmark control point selection network model, and realize the determination of the w value by a continuous iterative reverse calculation mode. The formula for the inverse calculation is as follows:
further, as an implementation of the method shown in the above embodiment, another embodiment of the present invention further provides a device for screening landmark control points. The embodiment of the apparatus corresponds to the embodiment of the method, and for convenience of reading, details in the embodiment of the apparatus are not repeated one by one, but it should be clear that the apparatus in the embodiment can correspondingly implement all the contents in the embodiment of the method. In the apparatus of this embodiment, there are the following modules:
firstly, the method comprises the following steps: landmark metadata database module: the landmark detection network model is used for inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database; the landmark metadata base module corresponds to step 101 in embodiment 1.
II, secondly: a primary selection module: the initial landmark control points in the landmark metadata base are initially selected, the initial landmark control points with obvious characteristics and convenient control information transmission are selected, and an initially selected landmark control point set is formed; the preliminary selection module corresponds to step 102 in embodiment 1.
Thirdly, the method comprises the following steps: a selection module: the network model is used for inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed. The culling module corresponds to step 103 in example 1.
In the above embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments of the apparatus.
It will be appreciated that the relevant features of the above methods and systems may be referred to one another. In addition, "first", "second", and the like in the above embodiments are for distinguishing the embodiments, and do not represent merits of the embodiments.
The algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose systems may also be used with the teachings herein. The required structure for constructing such a system will be apparent from the description above. Moreover, the present invention is not directed to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the invention as described herein, and any descriptions of specific languages are provided above to disclose the best mode of the invention.
In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the invention and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.
Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.
Claims (8)
1. A method for screening landmark control points is characterized by comprising the following steps:
step 101: inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database;
step 102: carrying out primary selection on the initial landmark control points in the landmark metadata base, and selecting the initial landmark control points with obvious characteristics and convenient control information transmission to form a primary selection landmark control point set;
step 103: inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed.
2. The method for screening landmark control points according to claim 1, wherein the landmark control point refinement network model is trained by:
acquiring a satellite image to be trained and a corresponding initially selected landmark control point;
marking the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained according to the actual matching result of the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained;
and taking the marked satellite image to be trained and the initially selected landmark control point as sample data, and training the landmark control point selection network model according to the sample data and a deep learning technology.
3. The method for screening landmark control points according to claim 2, wherein the landmark control point culling network model is specifically configured as follows:
the multilayer convolutional neural network is used for extracting image characteristics of the satellite image and the initially selected landmark control points in the sample data;
the attribute feature extraction network is used for extracting attribute features of the satellite images and the initially selected landmark control points in the sample data, and the attribute features comprise: resolution, imaging load, attitude and orbit state;
and the full convolution network generates a characteristic vector according to the image characteristic and the attribute characteristic, inputs the characteristic vector into the full convolution network to obtain a response value, and the response value represents the matching degree of the satellite image and the initially selected landmark control point in the sample data.
4. The method for screening landmark control points according to claim 1, wherein the landmark metadata base is iteratively updated during the initial selection, specifically: for the same global mesh generation mesh, replacing the landmark metadata with the initially selected landmark control points which are acquired during initial selection and are better than the original landmark metadata in the mesh, and iteratively realizing optimization of a landmark metadata database.
5. The method for screening landmark control points according to claim 4, wherein the initially selected index includes: the landmark image resolution, the landmark image positioning precision, the landmark image state and the landmark distribution; and comprehensively scoring the indexes to evaluate the advantages and disadvantages of the initial landmark control points.
6. A device for screening landmark control points, comprising:
landmark metadata database module: the landmark detection network model is used for inputting a reference image with high positioning precision into a pre-trained landmark detection network model; detecting an initial landmark control point set in the reference image with high positioning accuracy, and storing attribute information of landmark control points in the initial landmark control point set into a database according to a format of a landmark metadata system to form a landmark metadata database;
a primary selection module: the initial landmark control points in the landmark metadata base are initially selected, the initial landmark control points with obvious characteristics and convenient control information transmission are selected, and an initially selected landmark control point set is formed;
a selection module: the network model is used for inputting the image to be processed into a pre-trained landmark control point selection network model; and detecting the initially selected landmark control point with higher matching degree with the image to be processed from the initially selected landmark control point set to serve as the selected landmark control point of the image to be processed.
7. The apparatus for filtering landmark control points according to claim 6, wherein the landmark control point refinement network model is trained by:
acquiring a satellite image to be trained and a corresponding initially selected landmark control point;
marking the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained according to the actual matching result of the satellite image to be trained and the primary landmark control point corresponding to the satellite image to be trained;
and taking the marked satellite image to be trained and the initially selected landmark control point as sample data, and training the landmark control point selection network model according to the sample data and a deep learning technology.
8. The device for filtering landmark control points according to claim 7, wherein the network model for landmark control point refinement is specifically configured as follows:
the multilayer convolutional neural network is used for extracting image characteristics of the satellite image and the initially selected landmark control points in the sample data;
the attribute feature extraction network is used for extracting attribute features of the satellite images and the initially selected landmark control points in the sample data, and the attribute features comprise: resolution, imaging load, attitude and orbit state;
and the full convolution network generates a characteristic vector according to the image characteristic and the attribute characteristic, inputs the characteristic vector into the full convolution network to obtain a response value, and the response value represents the matching degree of the satellite image and the initially selected landmark control point in the sample data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111323543.3A CN114022784B (en) | 2021-11-09 | 2021-11-09 | Method and device for screening landmark control points |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111323543.3A CN114022784B (en) | 2021-11-09 | 2021-11-09 | Method and device for screening landmark control points |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114022784A true CN114022784A (en) | 2022-02-08 |
| CN114022784B CN114022784B (en) | 2022-04-19 |
Family
ID=80062734
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111323543.3A Active CN114022784B (en) | 2021-11-09 | 2021-11-09 | Method and device for screening landmark control points |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114022784B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117907242A (en) * | 2024-03-15 | 2024-04-19 | 贵州省第一测绘院(贵州省北斗导航位置服务中心) | Homeland mapping method, system and storage medium based on dynamic remote sensing technology |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101446642A (en) * | 2008-04-11 | 2009-06-03 | 国家卫星气象中心 | Automatic matching method for remote sensing satellite data ground control point based on knowledge learning |
| CN101672646A (en) * | 2009-10-09 | 2010-03-17 | 国家卫星气象中心 | Satellite-image automatic, geometric and precise correction method |
| CN103235810A (en) * | 2013-04-23 | 2013-08-07 | 国家测绘地理信息局卫星测绘应用中心 | Intelligent remote-sensing image control point data search method |
| US20150043772A1 (en) * | 2013-08-09 | 2015-02-12 | Toshiba Medical Systems Corporation | Method of, and apparatus for, landmark location |
| CN108090898A (en) * | 2017-12-20 | 2018-05-29 | 西北工业大学 | The satellite remote sensing images typical case's terrestrial reference detection method represented based on dictionary |
| CN109542932A (en) * | 2018-11-28 | 2019-03-29 | 中国科学院遥感与数字地球研究所 | The selected remotely-sensed data collection of Landsat-8 satellite customizes screening technique |
| CN110287831A (en) * | 2019-06-13 | 2019-09-27 | 中国人民解放军战略支援部队信息工程大学 | A kind of acquisition methods, device and the electronic equipment at the control point based on terrestrial reference |
| US20200134311A1 (en) * | 2018-10-26 | 2020-04-30 | Here Global B.V. | Method, apparatus, and system for determining a ground control point from image data using machine learning |
| CN111552751A (en) * | 2020-04-22 | 2020-08-18 | 中国人民解放军战略支援部队信息工程大学 | Three-dimensional landmark control point generation and application method, generation and application device |
| EP3885990A1 (en) * | 2020-03-26 | 2021-09-29 | Siemens Healthcare GmbH | Landmark detection |
-
2021
- 2021-11-09 CN CN202111323543.3A patent/CN114022784B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101446642A (en) * | 2008-04-11 | 2009-06-03 | 国家卫星气象中心 | Automatic matching method for remote sensing satellite data ground control point based on knowledge learning |
| CN101672646A (en) * | 2009-10-09 | 2010-03-17 | 国家卫星气象中心 | Satellite-image automatic, geometric and precise correction method |
| CN103235810A (en) * | 2013-04-23 | 2013-08-07 | 国家测绘地理信息局卫星测绘应用中心 | Intelligent remote-sensing image control point data search method |
| US20150043772A1 (en) * | 2013-08-09 | 2015-02-12 | Toshiba Medical Systems Corporation | Method of, and apparatus for, landmark location |
| CN108090898A (en) * | 2017-12-20 | 2018-05-29 | 西北工业大学 | The satellite remote sensing images typical case's terrestrial reference detection method represented based on dictionary |
| US20200134311A1 (en) * | 2018-10-26 | 2020-04-30 | Here Global B.V. | Method, apparatus, and system for determining a ground control point from image data using machine learning |
| CN109542932A (en) * | 2018-11-28 | 2019-03-29 | 中国科学院遥感与数字地球研究所 | The selected remotely-sensed data collection of Landsat-8 satellite customizes screening technique |
| CN110287831A (en) * | 2019-06-13 | 2019-09-27 | 中国人民解放军战略支援部队信息工程大学 | A kind of acquisition methods, device and the electronic equipment at the control point based on terrestrial reference |
| EP3885990A1 (en) * | 2020-03-26 | 2021-09-29 | Siemens Healthcare GmbH | Landmark detection |
| CN111552751A (en) * | 2020-04-22 | 2020-08-18 | 中国人民解放军战略支援部队信息工程大学 | Three-dimensional landmark control point generation and application method, generation and application device |
Non-Patent Citations (3)
| Title |
|---|
| LAI GUANGLING 等: "Research on Automatic Generation and Data Organization Method of Control Points", 《 2018 10TH IAPR WORKSHOP ON PATTERN RECOGNITION IN REMOTE SENSING (PRRS)》 * |
| 丁璐 等: "三轴稳定型静止轨道遥感卫星指向确定的地标匹配方法", 《测绘学报》 * |
| 赖广陵: "全球智慧地标控制网构建关键技术研究", 《中国博士学位论文全文数据库基础科学辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117907242A (en) * | 2024-03-15 | 2024-04-19 | 贵州省第一测绘院(贵州省北斗导航位置服务中心) | Homeland mapping method, system and storage medium based on dynamic remote sensing technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114022784B (en) | 2022-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222626B (en) | Unmanned scene point cloud target labeling method based on deep learning algorithm | |
| CN108871353B (en) | Road network map generation method, system, equipment and storage medium | |
| CN111527467A (en) | Method and apparatus for automatically defining computer-aided design files using machine learning, image analysis, and/or computer vision | |
| US20190303725A1 (en) | Neural network training system | |
| CN111523545B (en) | Article searching method combined with depth information | |
| Chen et al. | 3D photogrammetry point cloud segmentation using a model ensembling framework | |
| CN111179440A (en) | Three-dimensional object model retrieval method oriented to natural scene | |
| CN116524369B (en) | Remote sensing image segmentation model construction method and device and remote sensing image interpretation method | |
| CN114022784B (en) | Method and device for screening landmark control points | |
| CN116844068B (en) | Building mapping method, system, computer equipment and storage medium | |
| CN115147599A (en) | Object six-degree-of-freedom pose estimation method for multi-geometric feature learning of occlusion and truncation scenes | |
| CN104063421B (en) | Magnanimity traffic remotely-sensed data search method and device | |
| WO2004023394A1 (en) | Environmental reasoning using geometric data structure | |
| CN112241676A (en) | Method for automatically identifying terrain sundries | |
| Chen et al. | Improved fast r-cnn with fusion of optical and 3d data for robust palm tree detection in high resolution uav images | |
| Zhang et al. | Improved feature point extraction method of ORB-SLAM2 dense map | |
| CN117237623B (en) | Semantic segmentation method and system for remote sensing image of unmanned aerial vehicle | |
| CN115797795B (en) | Remote sensing image question-answer type retrieval system and method based on reinforcement learning | |
| CN116091706A (en) | Three-dimensional reconstruction method for multi-mode remote sensing image deep learning matching | |
| CN114998630A (en) | Ground-to-air image registration method from coarse to fine | |
| CN115565072A (en) | Road garbage recognition and positioning method and device, electronic equipment and medium | |
| CN114780758A (en) | Construction method of image control point library based on stereo mapping satellite image | |
| CN111435537B (en) | Model training method and device and pose optimization method and device based on mosaic | |
| CN113808186A (en) | Training data generation method and device and electronic equipment | |
| Brightman et al. | A brief overview of the current state, challenging issues and future directions of point cloud registration |
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 |