CN109903304A - A kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization - Google Patents
A kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization Download PDFInfo
- Publication number
- CN109903304A CN109903304A CN201910136835.2A CN201910136835A CN109903304A CN 109903304 A CN109903304 A CN 109903304A CN 201910136835 A CN201910136835 A CN 201910136835A CN 109903304 A CN109903304 A CN 109903304A
- Authority
- CN
- China
- Prior art keywords
- building
- polygon
- regularization
- convolutional neural
- algorithm
- 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
Abstract
The invention discloses a kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization includes the following steps: according to existing image and building covering vector file building sample database；The full convolutional neural networks of Multiscale Fusion are constructed, and it is trained by sample database, remote sensing image is predicted using trained network model, obtains the segmentation result of remote sensing image surface buildings covering；Based on building semantic segmentation as a result, carrying out building edge initialization, and obtain initialization vector polygon；The polygon of mistake and side, the node of polygon mistake are rejected using coarse regulation algorithm；Regularization is carried out to vector polygon using regularization algorithm, obtains the building vector edge of rule.The full convolutional neural networks scale strong robustness of Multiscale Fusion in the present invention, regularization algorithm are adapted to the vector edge in a variety of situations, the artificial workload for drawing building edge of the reduction of high degree.
Description
Technical field
The present invention relates to a kind of deep learning method extracted for remote sensing image building and building polygon profiles
Regularization algorithm, can be used for remote sensing image building extract, building vector edge generate, building variation detection etc..
Background technique
Remote sensing image building automation is extracted to be had in the application such as urban planning, population estimate, cartography and update
There is particularly important meaning.Traditionally, the groundwork that building is extracted from aviation/space flight image concentrates on: empirically setting
An appropriate feature is counted to express " what is building ", and creates corresponding feature for the automatic identification of building and mentions
It takes.Common index includes pixel, spectrum, length, edge, shape, texture, shade, height, semanteme etc..And these indexs
But significantly variation can occur with season, illumination, atmospheric conditions, sensor mass, scale, building style and environment.
Therefore, the method for this design feature by rule of thumb can only usually handle specific data, and can not be truly realized automation.Depth
Convolutional neural networks in study show powerful performance in image retrieval, image classification, target detection.Convolutional Neural net
Network can learn the feature representation of a multilayer automatically, and original input picture is mapped as unitary or polynary label.It is this
The ability beyond of self-teaching feature and gradually instead of the method for traditional artificial experience design feature.Building extracts not only
It is classification and semantic segmentation problem or a target detection, example segmentation problem.The target and non-interesting that building extracts
Whether some pixel is building, is more concerned about quantity, position, the shape of building.Draftsman's all over the world is main, numerous
One of work of weight is exactly to sketch the contours building polar plot by hand on aviation/space flight image, takes this to produce all kinds of topographic maps and specially
Topic figure.Therefore, most important for the vector data extraction research of building.
Summary of the invention
In view of the deficiencies of the prior art, the present invention provides a kind of neural networks of scale strong robustness, can adapt to not
Remote sensing image building with scale extracts, and priori sex knowledge is added to the building edge that semantic segmentation obtains and carries out rule
Change processing, obtains the regular building polygon of high quality.
Realize the object of the invention the technical solution adopted is that, it is a kind of based on convolutional Neural metanetwork and polygon regularization
Contour of building automatic Extraction Algorithm, includes the following steps:
Step 1, sample database is constructed according to existing image and building covering vector file；
Step 2, the full convolutional neural networks of Multiscale Fusion are constructed, and it is trained by sample database, utilize training
Good network model predicts remote sensing image, obtains the segmentation result of remote sensing image surface buildings covering；
Step 3, building semantic segmentation is based on as a result, carrying out building edge initialization, and it is polygon to obtain initialization vector
Shape；
Step 4, the polygon of mistake and side, the node of polygon mistake are rejected using coarse regulation algorithm；
Step 5, regularization is carried out to vector polygon using regularization algorithm, obtains the building vector edge of rule.
Further, the full convolutional neural networks of Multiscale Fusion described in step 2 include coding (encoding stage),
Decode (decoding stage) and output 3 parts (output)；Wherein coded portion is by 5 convolutional layers
(Convolution Layer), 4 maximum pond layers (Max Pooling Layer) form；Decoded portion by 4 convolutional layers,
4 warp lamination (Deconvolution Layer) compositions；Output par, c is made of 4 son outputs and 1 main output.
Further, the convolution that preceding two layers of convolutional layer of the coded portion is continuously stacked by two groups corrects linear unit
(Rectified Linear Unit, ReLU) and batch normalization layer (Batch Normalization, a BN) composition；
The convolution that three-layer coil lamination is continuously stacked by three groups afterwards, correct linear unit (Rectified Linear Unit, ReLU) and
Two batch normalization layer (Batch Normalization, BN) compositions；First three convolutional layer of the decoded portion and last
The convolution (Convolution) that one convolutional layer is continuously stacked by 4 groups and 3 groups respectively corrects linear unit (Rectified
Linear Unit, ReLU) composition；Each decoded portion finally obtains a son output, and main output is every by series connection decoded portion
The characteristic pattern in 1 last channel, obtains the characteristic pattern in 4 channels on a scale, activates to the end by Sigmoid function
Prediction result.
Further, edge initialization is carried out using Douglas-Peucker algorithm in step 3, according to each connected domain
Number of pixels different limit difference threshold value D is set, obtain initialization vector polygon.
Further, the specific implementation of step 4 is as follows,
41) side of building should have certain length: rejecting polygon and be less than TdSide；
42) corners of building cannot be excessively sharp or excessively gentle: being rejected and be less than α and the angle greater than β；
43) building should have certain area: delete the polygon that area is less than S；
44) building node should not be excessive: the ratio for rejecting area and node number is less than TaWith of perimeter and node
Number ratio is less than TpPolygon.
Further, the specific implementation of step 5 is as follows,
51) it determines the principal direction of building: different side length threshold value W is set according to the area of building, and length is greater than W's
While being long side, remaining is short side, and W is successively reduced w by the building that can not find at least one long side smaller for area, until
Find at least one long side；Firstly, being added in principal direction list, using longest a line as initial principal direction by remaining institute
There is long side compared with principal direction list, if the angle of the long side and all principal directions is greater than δminAnd it is less than δmax, then by the length
The direction on side is added to principal direction queue；
52) edge of polygon is adjusted according to principal direction: being first adjusted long side, by long side and principal direction, main side
To vertical direction compare, obtain the smallest angle, if the angle be less than δmin, then by the long side using midpoint as rotation center,
Rotate to the parallel or vertical direction that minimum angle corresponds to principal direction；Then short side adjustment is carried out, by short side and principal direction, main side
To vertical direction compare, obtain the smallest angle, if the angle be less than Ts, then by the short side using midpoint as rotation center,
Rotate to the parallel or vertical direction that minimum angle corresponds to principal direction；
53) vertical line is rejected and be added to redundancy parallel lines: edge after adjust is parallel there may be continuous two sides
Different distance threshold d is arranged according to the floor area of building in line, if the distance between continuous parallel lines is greater than d, in two parallel lines
Between be added a vertical line, otherwise, reject the lesser side of length；
54) intersection point successively calculated between continuous two sides successively connects all nodes as the node of vector polygon
It connects, the vector polygon after being adjusted.
Further, the specific implementation of step 1 is as follows,
The corresponding building covering vector file of image is converted into two-value grating image first, raster pixel image value is
Raw video and corresponding grid label are cut into tile structure wherein 0 represents background 1 as building by the label of image pixel
Build sample database.
The present invention has the advantage that the full convolutional neural networks scale strong robustness of Multiscale Fusion, is suitable for different rulers
The remote sensing image building of degree extracts, sustainable continuous iteration optimization；Regularization algorithm is adapted to the vector in a variety of situations
Edge, the artificial workload for drawing building edge of the reduction of high degree.
Detailed description of the invention
Fig. 1 is sample database building flow chart of the invention.
Fig. 2 is neural network structure schematic diagram of the invention.
Fig. 3 is the embodiment of the present invention data area figure.
Fig. 4 is the embodiment of the present invention data building schematic diagram.
Fig. 5 is the embodiment of the present invention result schematic diagram.
Specific embodiment
Below with reference to the embodiments and with reference to the accompanying drawing the technical solutions of the present invention will be further described.
A kind of contour of building based on convolutional Neural metanetwork and polygon regularization provided by the invention automatically extracts
Algorithm, comprising: step 1, the corresponding building covering vector file of image is converted into two-value grating image, grating image first
Pixel value is the label of image pixel, wherein 0 represents background 1 as building, raw video and corresponding grid label are cut
Sample database is constructed at tile；Step 2, to the full convolutional neural networks of Multiscale Fusion (Multi-scale Aggregation
Full Convolutional Neural Network, MA-FCN) it is trained, learn the feature of building in remote sensing image；
After network model training, remote sensing image is predicted using network model is trained, obtains remote sensing image earth's surface building
The segmentation result of object covering；Step 3, based on building semantic segmentation as a result, being built using Douglas-Peucker algorithm
Object edge initialization；Step 4, the polygon of mistake and side, the node of polygon mistake are rejected using coarse regulation algorithm；Step 5,
Regularization is carried out to vector polygon using regularization algorithm, obtains the building vector edge of rule.
The full convolutional neural networks of above-mentioned Multiscale Fusion include coding (encoding stage), decoding (decoding
Stage) and output 3 parts (output).Coded portion is by 5 convolutional layers (Convolution Layer), 4 maximums
Pond layer (Max Pooling Layer) composition；Decoded portion is by 4 convolutional layers, 4 warp lamination (Deconvolution
Layer it) forms；Output par, c is made of 4 son outputs and 1 main output.
The convolution that preceding two layers of convolutional layer of above-mentioned coded portion is continuously stacked by two groups corrects linear unit (Rectified
Linear Unit, ReLU) and batch normalization layer (Batch Normalization, a BN) composition；Three-layer coil product afterwards
The convolution that layer is continuously stacked by three groups, corrects linear unit (Rectified Linear Unit, ReLU) and two batches are returned
One changes layer (Batch Normalization, BN) composition.Convolution kernel size is 3 × 3, and convolution step-length is 1.
The maximum pond layer step-length of above-mentioned coded portion is 2 × 2, after the layer of pond, exports the height and width of characteristic pattern
Degree becomes the half of input.
Each convolutional layer of above-mentioned decoded portion, input are corresponding with coded portion for the characteristic pattern obtained after deconvolution
The series connection of size characteristic figure.
First three convolutional layer and the last one convolutional layer of above-mentioned decoded portion are respectively by 4 groups and 3 groups of volumes continuously stacked
Product (Convolution), amendment linear unit (Rectified Linear Unit, ReLU) composition.Convolution kernel size is 3
× 3, convolution step-length is 1.Each decoded portion finally obtains a son output.The feature in 1 channel that convolutional layer finally obtains
Figure is activated to obtain the sub- prognostic chart of corresponding scale by Sigmoid function.
Main output par, c obtains 4 channels by the characteristic pattern in 1 channel last on the series connection each scale of decoded portion
Characteristic pattern, the prediction result for activating to the end by Sigmoid function.
In step 3, after obtaining building semantic segmentation result, it is initial that edge is carried out using Douglas-Peucker algorithm
Change, different limit difference threshold value D is set according to the number of pixels of each connected domain, obtains initialization vector polygon.
In step 4, initial polygon is adjusted using coarse regulation algorithm:
41) side of building should have certain length: rejecting polygon and be less than TdSide；
42) corners of building cannot be excessively sharp or excessively gentle: being rejected and be less than α and the angle greater than β；
43) building should have certain area: the polygon that area is less than S is deleted,；
44) building node should not be excessive: the ratio for rejecting area and node number is less than TaWith of perimeter and node
Number ratio is less than TpPolygon.
In step 5, fine control is carried out for the polygon after coarse regulation:
51) principal direction of building is determined.Different side length threshold value W is set according to the area of building, and length is greater than W's
While being long side, remaining is short side.W is successively reduced w by the building that can not find at least one long side smaller for area, until
Find at least one long side.Firstly, being added in principal direction list using longest a line as initial principal direction.By remaining institute
There is long side compared with principal direction list, if the angle of the long side and all principal directions is greater than δminAnd it is less than δmax, then by the length
The direction on side is added to principal direction queue.
52) edge of polygon is adjusted according to principal direction.Long side is adjusted first.By long side and principal direction, main side
To vertical direction compare, obtain the smallest angle.If the angle is less than δmin, then by the long side using midpoint as rotation center,
Rotate to the parallel or vertical direction that minimum angle corresponds to principal direction.Then short side adjustment is carried out.By short side and principal direction, main side
To vertical direction compare, obtain the smallest angle.If the angle is less than Ts, then by the short side using midpoint as rotation center,
Rotate to the parallel or vertical direction that minimum angle corresponds to principal direction.
53) vertical line is rejected and be added to redundancy parallel lines.Edge after having adjusted is parallel there may be continuous two sides
Line.Different distance threshold d is set according to the floor area of building, if the distance between continuous parallel lines is greater than d, in two parallel lines
Between be added a vertical line.Otherwise, the lesser side of length is rejected.
54) intersection point between continuous two sides, the node as vector polygon are successively calculated.All nodes are successively connected
It connects, the vector polygon after being adjusted.
Embodiment:
Referring to Fig. 1 and Fig. 2, the present invention utilizes the full convolutional neural networks (Multi-scale of Multiscale Fusion
Aggregation Full Convolutional Neural Network, MA-FCN) learn in high-resolution remote sensing image
Then building feature covers remote sensing image building and carries out Pixel-level prediction.In order to train neural network model, need first
Training sample is obtained, attached drawing 1 illustrates the process of building training sample database.First remote sensing image cut and be adopted again
Sample obtains the image capturing range that resolution ratio is suitable, has building covering data.Then building corresponding in image capturing range is sweared
Data rasterizing is measured, keeps it consistent with image resolution.Finally, in conjunction with factors such as computer performance, atural object sizes, by remote sensing shadow
Sample block as being divided into suitable size (such as 256 × 256 pixels or 512 × 512 pixels) with corresponding label data.Attached drawing 3
For sample database data entire scope, wherein dotted box picture is used to train, and image is used to test in solid box.Attached drawing 4 is part
Building example.
After obtaining training data, training is iterated to neural network, until model is optimal.It, will after the completion of model training
Remote sensing image to be sorted be cut into training sample image blocks of the same size, using trained model to image blocks carry out
Building extracts, and remote sensing image building segmentation result can be obtained.Finally the prediction result of all image blocks is spliced,
Obtain the building segmentation figure of complete image.
After obtaining building semantic segmentation figure, suitable parameter is arranged according to image, utilizes Douglas-Peucker algorithm
Carry out the initialization of building edge vectors.Parameter is set to the vector polygon coarse regulation after initialization.Suitable parameter is set
Fine control is carried out to the polygon after coarse regulation, obtains the vector result of regularization to the end.In test data, setting α=π/
6, β=π/18, S=20, Ta=1, Tp=1.5, Td=0.5, w=0.2, δmin=π/12, δmax=5 π/12, Ts=π/4, D are set
It is set to:Wherein N is number of pixels；W and d are as follows:Wherein S is
The floor area of building.
Attached drawing 5 is last result example, and first is classified as raw video and building edge example, and second, which is classified as image, leads to
The segmentation result that the full convolutional neural networks of Multiscale Fusion obtain is crossed, third is classified as to be obtained by Douglas-Peucker algorithm
Preliminary examination edge, the 4th is classified as the regularization edge by obtaining after regularization algorithm.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (7)
1. a kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization, feature exist
In including the following steps:
Step 1, sample database is constructed according to existing image and building covering vector file；
Step 2, the full convolutional neural networks of Multiscale Fusion are constructed, and it is trained by sample database, utilization is trained
Network model predicts remote sensing image, obtains the segmentation result of remote sensing image surface buildings covering；
Step 3, building semantic segmentation is based on as a result, carrying out building edge initialization, and obtain initialization vector polygon；
Step 4, the polygon of mistake and side, the node of polygon mistake are rejected using coarse regulation algorithm；
Step 5, regularization is carried out to vector polygon using regularization algorithm, obtains the building vector edge of rule.
2. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as described in claim 1 mentions automatically
Take algorithm, it is characterised in that: the full convolutional neural networks of Multiscale Fusion described in step 2 include coding (encoding
Stage), (decoding stage) and output 3 parts (output) are decoded；Wherein coded portion is by 5 convolutional layers
(Convolution Layer), 4 maximum pond layers (Max Pooling Layer) form；Decoded portion by 4 convolutional layers,
4 warp lamination (Deconvolution Layer) compositions；Output par, c is made of 4 son outputs and 1 main output.
3. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as claimed in claim 2 mentions automatically
Take algorithm, it is characterised in that: the convolution that preceding two layers of convolutional layer of the coded portion is continuously stacked by two groups corrects linear unit
(Rectified Linear Unit, ReLU) and batch normalization layer (Batch Normalization, a BN) composition；
The convolution that three-layer coil lamination is continuously stacked by three groups afterwards, correct linear unit (Rectified Linear Unit, ReLU) and
Two batch normalization layer (Batch Normalization, BN) compositions；First three convolutional layer of the decoded portion and last
The convolution (Convolution) that one convolutional layer is continuously stacked by 4 groups and 3 groups respectively corrects linear unit (Rectified
Linear Unit, ReLU) composition；Each decoded portion finally obtains a son output, and main output is every by series connection decoded portion
The characteristic pattern in 1 last channel, obtains the characteristic pattern in 4 channels on a scale, activates to the end by Sigmoid function
Prediction result.
4. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as described in claim 1 mentions automatically
Take algorithm, it is characterised in that: edge initialization is carried out using Douglas-Peucker algorithm in step 3, according to each connected domain
Number of pixels different limit difference threshold value D is set, obtain initialization vector polygon.
5. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as described in claim 1 mentions automatically
Taking algorithm, it is characterised in that: the specific implementation of step 4 is as follows,
41) side of building should have certain length: rejecting polygon and be less than TdSide；
42) corners of building cannot be excessively sharp or excessively gentle: being rejected and be less than α and the angle greater than β；
43) building should have certain area: delete the polygon that area is less than S；
44) building node should not be excessive: the ratio for rejecting area and node number is less than TaWith the number ratio of perimeter and node
Less than TpPolygon.
6. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as described in claim 1 mentions automatically
Taking algorithm, it is characterised in that: the specific implementation of step 5 is as follows,
51) it determines the principal direction of building: different side length threshold value W is set according to the area of building, side of the length greater than W is
Long side, remaining is short side, and W is successively reduced w by the building that can not find at least one long side smaller for area, until finding
At least one long side；Firstly, being added in principal direction list, using longest a line as initial principal direction by remaining all length
Side is compared with principal direction list, if the angle of the long side and all principal directions is greater than δminAnd it is less than δmax, then by the long side
Direction is added to principal direction queue；
52) edge of polygon is adjusted according to principal direction: being first adjusted long side, by long side and principal direction, principal direction
Vertical direction compares, and obtains the smallest angle, if the angle is less than δmin, then by the long side using midpoint as rotation center, rotation
The parallel or vertical direction of principal direction is corresponded to minimum angle；Then short side adjustment is carried out, by short side and principal direction, principal direction
Vertical direction compares, and obtains the smallest angle, if the angle is less than Ts, then by the short side using midpoint as rotation center, rotation
The parallel or vertical direction of principal direction is corresponded to minimum angle；
53) vertical line is rejected and be added to redundancy parallel lines: edge after adjust is parallel lines there may be continuous two sides, according to
Different distance threshold d is set according to the floor area of building, if the distance between continuous parallel lines is greater than d, is added between two parallel lines
Enter a vertical line, otherwise, rejects the lesser side of length；
54) all nodes are sequentially connected by the intersection point successively calculated between continuous two sides as the node of vector polygon,
Vector polygon after being adjusted.
7. a kind of contour of building based on convolutional Neural metanetwork and polygon regularization as described in claim 1 mentions automatically
Taking algorithm, it is characterised in that: the specific implementation of step 1 is as follows,
The corresponding building covering vector file of image is converted into two-value grating image first, raster pixel image value is image
Raw video and corresponding grid label are cut into tile building sample wherein 0 represents background 1 as building by the label of pixel
This library.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910136835.2A CN109903304B (en) | 2019-02-25 | 2019-02-25 | Automatic building contour extraction algorithm based on convolutional neural network and polygon regularization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910136835.2A CN109903304B (en) | 2019-02-25 | 2019-02-25 | Automatic building contour extraction algorithm based on convolutional neural network and polygon regularization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903304A true CN109903304A (en) | 2019-06-18 |
| CN109903304B CN109903304B (en) | 2021-03-16 |
Family
ID=66945394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910136835.2A Active CN109903304B (en) | 2019-02-25 | 2019-02-25 | Automatic building contour extraction algorithm based on convolutional neural network and polygon regularization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903304B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909602A (en) * | 2019-10-21 | 2020-03-24 | 广联达科技股份有限公司 | Two-dimensional vector diagram sub-domain identification method and device |
| CN111080587A (en) * | 2019-12-04 | 2020-04-28 | 中山大学 | Resampling image detection method based on convolutional neural network |
| CN111127538A (en) * | 2019-12-17 | 2020-05-08 | 武汉大学 | Multi-view image three-dimensional reconstruction method based on convolution cyclic coding-decoding structure |
| CN111489387A (en) * | 2020-04-09 | 2020-08-04 | 湖南盛鼎科技发展有限责任公司 | Remote sensing image building area calculation method |
| CN111652251A (en) * | 2020-06-09 | 2020-09-11 | 星际空间（天津）科技发展有限公司 | Method and device for building remote sensing image building feature extraction model and storage medium |
| CN111652892A (en) * | 2020-05-02 | 2020-09-11 | 王磊 | Remote sensing image building vector extraction and optimization method based on deep learning |
| CN111652250A (en) * | 2020-06-09 | 2020-09-11 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on polygon and storage medium |
| CN112288647A (en) * | 2020-10-13 | 2021-01-29 | 武汉大学 | Remote sensing image cloud and shadow restoration method based on gating convolution |
| CN112348836A (en) * | 2020-11-06 | 2021-02-09 | 二十一世纪空间技术应用股份有限公司 | Method and device for automatically extracting building outline |
| CN112700464A (en) * | 2021-01-15 | 2021-04-23 | 腾讯科技（深圳）有限公司 | Map information processing method and device, electronic equipment and storage medium |
| CN112949407A (en) * | 2021-02-02 | 2021-06-11 | 武汉大学 | Remote sensing image building vectorization method based on deep learning and point set optimization |
| CN113160258A (en) * | 2021-03-31 | 2021-07-23 | 武汉汉达瑞科技有限公司 | Method, system, server and storage medium for extracting building vector polygon |
| CN113190639A (en) * | 2021-05-13 | 2021-07-30 | 重庆市勘测院 | Comprehensive drawing method for residential area |
| CN113920147A (en) * | 2021-12-14 | 2022-01-11 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on deep learning |
| CN114387329A (en) * | 2022-01-16 | 2022-04-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN114485694A (en) * | 2020-11-13 | 2022-05-13 | 元平台公司 | System and method for automatically detecting building coverage area |
| CN114863288A (en) * | 2022-07-05 | 2022-08-05 | 航天宏图信息技术股份有限公司 | Building contour extraction and regularization method and device |
| WO2022227486A1 (en) * | 2021-04-27 | 2022-11-03 | 西安微电子技术研究所 | Artificial intelligence-based remote sensing field ridge boundary detection method, system, computer device, and storage medium |
| CN115797592A (en) * | 2023-02-08 | 2023-03-14 | 航天宏图信息技术股份有限公司 | Method and device for automatically generating building block based on oblique photography three-dimensional model |
| CN116343070A (en) * | 2023-05-22 | 2023-06-27 | 武汉天地鹰测绘科技有限公司 | Intelligent interpretation method for aerial survey image ground object elements |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726255A (en) * | 2008-10-24 | 2010-06-09 | 中国科学院光电研究院 | Method for extracting interesting buildings from three-dimensional laser point cloud data |
| CN102521884A (en) * | 2011-12-16 | 2012-06-27 | 南京大学 | Three-dimensional roof reconstruction method based on LiDAR data and ortho images |
| CN102938066A (en) * | 2012-12-07 | 2013-02-20 | 南京大学 | Method for reconstructing outer outline polygon of building based on multivariate data |
| CN105608691A (en) * | 2015-12-17 | 2016-05-25 | 武汉大学 | High-resolution SAR image individual building extraction method |
| CN105608458A (en) * | 2015-10-20 | 2016-05-25 | 武汉大学 | High-resolution remote sensing image building extraction method |
| CN106127857A (en) * | 2016-06-24 | 2016-11-16 | 南京林业大学 | Synthetic data drives the on-board LiDAR data modeling method with model-driven |
| CN106464320A (en) * | 2014-05-28 | 2017-02-22 | 高通股份有限公司 | Method and apparatus for orientation based beam selection |
| CN106548479A (en) * | 2016-12-06 | 2017-03-29 | 武汉大学 | A kind of multi-level laser point cloud building border rule method |
| CN106650749A (en) * | 2016-11-16 | 2017-05-10 | 桂林理工大学 | Method for plotting right-angled building in high-resolution optical image |
| CN108154209A (en) * | 2016-12-02 | 2018-06-12 | 杭州海康机器人技术有限公司 | A kind of Quick Response Code generation, recognition methods and device for localization for Mobile Robot |
-
2019
- 2019-02-25 CN CN201910136835.2A patent/CN109903304B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726255A (en) * | 2008-10-24 | 2010-06-09 | 中国科学院光电研究院 | Method for extracting interesting buildings from three-dimensional laser point cloud data |
| CN102521884A (en) * | 2011-12-16 | 2012-06-27 | 南京大学 | Three-dimensional roof reconstruction method based on LiDAR data and ortho images |
| CN102938066A (en) * | 2012-12-07 | 2013-02-20 | 南京大学 | Method for reconstructing outer outline polygon of building based on multivariate data |
| CN106464320A (en) * | 2014-05-28 | 2017-02-22 | 高通股份有限公司 | Method and apparatus for orientation based beam selection |
| CN105608458A (en) * | 2015-10-20 | 2016-05-25 | 武汉大学 | High-resolution remote sensing image building extraction method |
| CN105608691A (en) * | 2015-12-17 | 2016-05-25 | 武汉大学 | High-resolution SAR image individual building extraction method |
| CN106127857A (en) * | 2016-06-24 | 2016-11-16 | 南京林业大学 | Synthetic data drives the on-board LiDAR data modeling method with model-driven |
| CN106650749A (en) * | 2016-11-16 | 2017-05-10 | 桂林理工大学 | Method for plotting right-angled building in high-resolution optical image |
| CN108154209A (en) * | 2016-12-02 | 2018-06-12 | 杭州海康机器人技术有限公司 | A kind of Quick Response Code generation, recognition methods and device for localization for Mobile Robot |
| CN106548479A (en) * | 2016-12-06 | 2017-03-29 | 武汉大学 | A kind of multi-level laser point cloud building border rule method |
Non-Patent Citations (4)
| Title |
|---|
| SHUNPING JI等: "A scale robust convolutional neural network for automatic building extraction from aerial and satellite imagery", 《INTERNATIONAL JOURNAL OF REMOTE SENSING》 * |
| SHUNPING JI等: "Fully Convolutional Networks for Multisource Building Extraction From an Open Aerial and Satellite Imagery Data Set", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| 朱琴等: "利用LiDAR和航空影像的屋顶边缘提取及优化", 《地理空间信息》 * |
| 杨洋等: "基于LIDAR数据的建筑物自动化重建", 《测绘科学技术学报》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909602A (en) * | 2019-10-21 | 2020-03-24 | 广联达科技股份有限公司 | Two-dimensional vector diagram sub-domain identification method and device |
| CN111080587A (en) * | 2019-12-04 | 2020-04-28 | 中山大学 | Resampling image detection method based on convolutional neural network |
| CN111080587B (en) * | 2019-12-04 | 2023-05-26 | 中山大学 | Resampling image detection method based on convolutional neural network |
| CN111127538A (en) * | 2019-12-17 | 2020-05-08 | 武汉大学 | Multi-view image three-dimensional reconstruction method based on convolution cyclic coding-decoding structure |
| CN111127538B (en) * | 2019-12-17 | 2022-06-07 | 武汉大学 | Multi-view image three-dimensional reconstruction method based on convolution cyclic coding-decoding structure |
| CN111489387B (en) * | 2020-04-09 | 2023-10-20 | 湖南盛鼎科技发展有限责任公司 | Remote sensing image building area calculation method |
| CN111489387A (en) * | 2020-04-09 | 2020-08-04 | 湖南盛鼎科技发展有限责任公司 | Remote sensing image building area calculation method |
| CN111652892A (en) * | 2020-05-02 | 2020-09-11 | 王磊 | Remote sensing image building vector extraction and optimization method based on deep learning |
| CN111652250A (en) * | 2020-06-09 | 2020-09-11 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on polygon and storage medium |
| CN111652251A (en) * | 2020-06-09 | 2020-09-11 | 星际空间（天津）科技发展有限公司 | Method and device for building remote sensing image building feature extraction model and storage medium |
| CN111652251B (en) * | 2020-06-09 | 2023-06-27 | 星际空间（天津）科技发展有限公司 | Remote sensing image building feature extraction model construction method, device and storage medium |
| CN111652250B (en) * | 2020-06-09 | 2023-05-26 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on polygons and storage medium |
| CN112288647A (en) * | 2020-10-13 | 2021-01-29 | 武汉大学 | Remote sensing image cloud and shadow restoration method based on gating convolution |
| CN112348836A (en) * | 2020-11-06 | 2021-02-09 | 二十一世纪空间技术应用股份有限公司 | Method and device for automatically extracting building outline |
| CN112348836B (en) * | 2020-11-06 | 2024-03-12 | 二十一世纪空间技术应用股份有限公司 | Method and device for automatically extracting building outline |
| CN114485694A (en) * | 2020-11-13 | 2022-05-13 | 元平台公司 | System and method for automatically detecting building coverage area |
| CN112700464A (en) * | 2021-01-15 | 2021-04-23 | 腾讯科技（深圳）有限公司 | Map information processing method and device, electronic equipment and storage medium |
| CN112949407A (en) * | 2021-02-02 | 2021-06-11 | 武汉大学 | Remote sensing image building vectorization method based on deep learning and point set optimization |
| CN112949407B (en) * | 2021-02-02 | 2022-06-14 | 武汉大学 | Remote sensing image building vectorization method based on deep learning and point set optimization |
| CN113160258A (en) * | 2021-03-31 | 2021-07-23 | 武汉汉达瑞科技有限公司 | Method, system, server and storage medium for extracting building vector polygon |
| WO2022227486A1 (en) * | 2021-04-27 | 2022-11-03 | 西安微电子技术研究所 | Artificial intelligence-based remote sensing field ridge boundary detection method, system, computer device, and storage medium |
| CN113190639A (en) * | 2021-05-13 | 2021-07-30 | 重庆市勘测院 | Comprehensive drawing method for residential area |
| CN113190639B (en) * | 2021-05-13 | 2022-12-13 | 重庆市勘测院 | Comprehensive drawing method for residential area |
| CN113920147B (en) * | 2021-12-14 | 2022-03-01 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on deep learning |
| CN113920147A (en) * | 2021-12-14 | 2022-01-11 | 星际空间（天津）科技发展有限公司 | Remote sensing image building extraction method and device based on deep learning |
| CN114387329A (en) * | 2022-01-16 | 2022-04-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN114387329B (en) * | 2022-01-16 | 2024-03-22 | 四川轻化工大学 | Building contour progressive regularization method based on high-resolution remote sensing image |
| CN114863288A (en) * | 2022-07-05 | 2022-08-05 | 航天宏图信息技术股份有限公司 | Building contour extraction and regularization method and device |
| CN115797592A (en) * | 2023-02-08 | 2023-03-14 | 航天宏图信息技术股份有限公司 | Method and device for automatically generating building block based on oblique photography three-dimensional model |
| CN115797592B (en) * | 2023-02-08 | 2023-04-14 | 航天宏图信息技术股份有限公司 | Method and device for automatically generating building block based on oblique photography three-dimensional model |
| CN116343070A (en) * | 2023-05-22 | 2023-06-27 | 武汉天地鹰测绘科技有限公司 | Intelligent interpretation method for aerial survey image ground object elements |
| CN116343070B (en) * | 2023-05-22 | 2023-10-13 | 武汉天地鹰测绘科技有限公司 | Intelligent interpretation method for aerial survey image ground object elements |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903304B (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109903304A (en) | A kind of contour of building automatic Extraction Algorithm based on convolutional Neural metanetwork and polygon regularization | |
| CN110443143B (en) | Multi-branch convolutional neural network fused remote sensing image scene classification method | |
| CN105894045B (en) | A kind of model recognizing method of the depth network model based on spatial pyramid pond | |
| CN107609525B (en) | Remote sensing image target detection method for constructing convolutional neural network based on pruning strategy | |
| CN108038853B (en) | Ceramic tile surface defect identification method based on convolutional neural network and active learning | |
| CN109977918B (en) | Target detection positioning optimization method based on unsupervised domain adaptation | |
| CN110414377B (en) | Remote sensing image scene classification method based on scale attention network | |
| CN108154158B (en) | Building image segmentation method for augmented reality application | |
| CN102902956B (en) | A kind of ground visible cloud image identifying processing method | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN104778242A (en) | Hand-drawn sketch image retrieval method and system on basis of image dynamic partitioning | |
| CN106096655B (en) | A kind of remote sensing image airplane detection method based on convolutional neural networks | |
| CN105046272B (en) | A kind of image classification method based on succinct non-supervisory formula convolutional network | |
| CN110223310B (en) | Line structure light center line and box edge detection method based on deep learning | |
| CN110543906B (en) | Automatic skin recognition method based on Mask R-CNN model | |
| CN108629368B (en) | Multi-modal foundation cloud classification method based on joint depth fusion | |
| CN111127360B (en) | Gray image transfer learning method based on automatic encoder | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN110517270B (en) | Indoor scene semantic segmentation method based on super-pixel depth network | |
| CN109360179A (en) | A kind of image interfusion method, device and readable storage medium storing program for executing | |
| CN114510772A (en) | Rapid generation method of building surface contour based on oblique scanning data | |
| CN105913451B (en) | A kind of natural image superpixel segmentation method based on graph model | |
| CN111161267A (en) | Segmentation method of three-dimensional point cloud model | |
| Kliangsuwan et al. | FFT features and hierarchical classification algorithms for cloud images | |
| CN107292268A (en) | The SAR image semantic segmentation method of quick ridge ripple deconvolution Structure learning model |
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 |