CN109871812A - A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based - Google Patents
A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based Download PDFInfo
- Publication number
- CN109871812A CN109871812A CN201910135591.6A CN201910135591A CN109871812A CN 109871812 A CN109871812 A CN 109871812A CN 201910135591 A CN201910135591 A CN 201910135591A CN 109871812 A CN109871812 A CN 109871812A
- Authority
- CN
- China
- Prior art keywords
- vegetation
- remote sensing
- sensing image
- neural network
- area
- 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.)
- Pending
Links
Abstract
The invention discloses a kind of multi-temporal remote sensing image urban vegetation extracting methods neural network based, belong to vegetation extractive technique field, are based on multidate high-definition remote sensing image data, by carrying out image segmentation to image；Training sample is trained using BP neural network method, forms the neural network model for being directed to test block vegetation；Vegetation extraction is carried out to multi-temporal data using the model, is extracted using the voting rule fusion multidate based on weight as a result, to obtain the higher vegetation area of precision.Residential block vegetation is extracted using the present invention, extraction accuracy is promoted to 93.3% from 87.6%, has preferable practicability.The vegetation in other regions such as residential block, shopping centre, industrial area of the present invention suitable for real work is extracted, and may extend in the vegetation coverage research work in entire city.
Description
Technical field
The invention belongs to vegetation extractive technique fields, and in particular to a kind of multi-temporal remote sensing image city neural network based
City's vegetation extracting method.
Background technique
With the development of remote sensor technology, high spatial resolution remote sense image plays in city atural object automatically extracts
Increasingly important role.The atural objects such as road, vegetation, building required for how more effectively being extracted from remote sensing image,
It is an important process of current remote sensing fields.
The vegetation coverage measurement of residential block is an important and conventional job in current mapping operations, traditional people
Work mapping needs to survey and draw operating personnel and surveys on the spot, and the interior industry based on satellite remote-sensing image either aerial images is artificial
Interpretation, equally takes time and effort, but also has biggish operation subjectivity, not can guarantee unified objectively interior industry operation scale.
Currently, being directed to the extraction of the urban vegetation based on high-resolution remote sensing image, related scholar is studied, according to
The difference of pixel and object can be divided into the extracting method of extracting method and object-oriented based on pixel.Based on pixel method
It generallys use vegetation index and experience or adaptive threshold fuzziness extracts, general principles are that vegetation index is greater than corresponding shadow
The empirical value of picture, the pixel are just identified as vegetation area；And object-oriented method is then on the basis of pixel, by pixel
Composed object carries out feature judgement, can introduce other features such as texture, shape to increase recognition accuracy.And from resident
The extraction that vegetation is carried out in area is a typical case during city vegetation is extracted.However, due to the inclination angle of satellite remote sensing photography
Problem, the building in city would generally cover greenery patches on satellite image, while shade caused by building also can be very big
The stability of identification is influenced in degree, both of these problems bring bigger difficulty to the automatic identification of urban vegetation.
Summary of the invention
Goal of the invention: being directed to the deficiencies in the prior art, and the purpose of the present invention is to provide one kind to be based on nerve net
The multi-temporal remote sensing image urban vegetation extracting method of network.Residential block, shopping centre, industry of the present invention suitable for real work
The vegetation in other regions such as area is extracted, to extend in the vegetation coverage research work in entire city.
Technical solution: to solve the above-mentioned problems, the technical solution adopted in the present invention is as follows:
A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based, it is distant based on multidate high-resolution
Image data is felt, by carrying out image segmentation to image；Sample is trained using BP neural network method, is formed for real
Test the neural network model of area vegetation；Vegetation extraction is carried out to multi-temporal data using the model, using the ballot based on weight
Rule fusion multidate extracts as a result, to obtain the higher vegetation area of precision.Concrete operations are as follows:
(1) multi-temporal remote sensing image data cutting is multiple objects by the image segmentation stage；
(2) the image recognition stage identifies each object using vegetation identification model, if belonged in cut zone
The area of vegetation is greater than 50%, then determines the cut zone for vegetation area, otherwise, it is determined that being nonvegetated area domain；
(3) multi-temporal data fusing stage assigns recent remote sensing image biggish weight, after determining Remote Sensing Image Segmentation
Object whether be vegetation area voting rule using shown in following formula,
Wherein, yeariIndicate i-th of remote sensing image corresponding time, whether is the object in i-th of remote sensing image of expression
For vegetation area；Work as PFusionWhen greater than 0.5, the object is determined for vegetation area, otherwise, it is determined that being nonvegetated area domain.
Preferably, the image that the remote sensing image data is issued from Google Earth.
Preferably, the spatial resolution of the remote sensing image is 0.51 meter, and by pretreatment geometric correction and registration.
Preferably, described image dividing method is using K-means clustering method, the advantage of this method is that without setting
Partitioning parameters are set, there is preferable segmentation adaptivity, process is as follows:
1) image is converted, is transformed to the rectangle data format suitable for the operation of K-means method；
2) 6 pixels are randomly selected in the matrix after converting image as random center, clearly participate in calculating all
The center of image and this 6 pixel position consistencies；
3) characteristic distance calculating is carried out to the random center to each pixel；
4) each pixel is classified as to the random center of minimum distance；
5) center of each classification is readjusted；
6) iteration 3)~5) step, until the number of iterations is greater than the variation at 300 or the relatively last round of center in new center
Less than 0.05%, algorithm terminates rate.
Preferably, the step of acquisition vegetation identification model includes:
(1) sample area is determined, by manually visualizing the vegetation pixel chosen in sample area；
(2) sample area remote sensing image is scanned using the window of 64*64, according to statistics, determines that the window area is
No is vegetation area, i.e., if the pixel ratio for belonging to vegetation in window area is greater than 50%, determines window area to plant
By region and as the positive sample of machine learning, otherwise it is determined as that the window is nonvegetated area domain, and as the negative of machine learning
Sample；
(3) by 64*64 pixel of positive and negative sample window, neural network input layer is inputted, is calculated using BP neural network
Method carries out model training, obtains vegetation identification model.
Preferably, the step 3) characteristic distance uses the Euclidean distance of pixel value.
Preferably, the test block is city neighborhood, shopping centre or industrial area.
Preferably, accuracy rate calculating is carried out to the extraction result are as follows: use following formula
Wherein, SiIndicate whether the object i of Visual Outcomes is vegetation area, is then SiIt is 1, is otherwise 0；PiIt indicates more
Whether the object i after phase data fusion is vegetation area, is then PiIt is 1, is otherwise object sum for 0, n.
The utility model has the advantages that compared with prior art, the invention has the advantages that
(1) present invention by with manually visualize extract result compared with, multi-temporal remote sensing image city neural network based
City's vegetation extracting method has preferable accuracy rate, extracts compared with vegetation area with the remote sensing image of single phase is used, right
The extraction of residential block vegetation area, extraction accuracy are promoted to 93.3% from 87.6%, have preferable practicability.
(2) vegetation in other regions such as residential block, shopping centre, the industrial area of the present invention suitable for real work is extracted,
To extend in the vegetation coverage research work in entire city.
Detailed description of the invention
Fig. 1 is overall procedure schematic diagram；
Fig. 2 is the multi-temporal remote sensing image figure of test block, wherein figure a is the remote sensing image in test block on July 15th, 2018
Figure, figure b are the remote sensing image in test block on October 9th, 2017, and figure c is the remote sensing image in test block on July 7th, 2017,
Scheme the remote sensing image that d is test block on November 28th, 2016；
Fig. 3 is BP neural network structure chart, wherein x1, x2... xnFor input layer, y1, y2…ymFor output layer section
Point, wijFor the weight between input layer i and hiding node layer j, wjkTo hide between node layer j and output node layer k
Weight；
Fig. 4 a is the remote sensing image of sample area；
Fig. 4 b is the vegetation area figure of sample area；
Fig. 5 is vegetation identification model neural metwork training flow chart；
Fig. 6 is Remote Sensing Image Segmentation region recognition flow chart；
Fig. 7 is that test block remote sensing image vegetation in 2018 extracts result figure；
Fig. 8 a is that multi-temporal data fused vegetation in test block extracts result figure；
Fig. 8 b is that sample area vegetation extracts result figure.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, right combined with specific embodiments below
A specific embodiment of the invention is described in detail.
Embodiment 1
A kind of flow diagram such as Fig. 1 institute of multi-temporal remote sensing image urban vegetation extracting method neural network based
Show.As shown in Figure 1, vegetation extraction process is divided into three parts, and first part is the image segmentation stage, and its object is to will be more
Phase remote sensing image data cutting is multiple objects, and the extraction and identification of vegetation are carried out using object, is conducive to merge multiple features
Factor.The method of segmentation is using K-means clustering method, the advantage of this method is that no setting is required partitioning parameters, have
Preferable segmentation adaptivity；Second part is to be identified to the object after segmentation using BP neural network model, mainly
Utilize the vegetation identification model that neural network is constructed after expert along training sample；Part III is to multi-temporal remote sensing image data
It extracts result and carries out multi-temporal data fusion, export result after fused result is carried out accuracy test.
1, the image segmentation stage
A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based, by taking city neighborhood as an example.It is real
The area Yan Yang be Suzhou City newly create bamboo plantation cell, using come from July 15th, 2018, on October 9th, 2017, on July 7th, 2017 and
November in 2016 this four phases on the 28th remote sensing image data, as shown in Figure 2.The remote sensing image data of use from
The image of Google Earth publication, the spatial resolution of image are 0.51 meter.Currently, due to the image matter of Google Earth
Amount is more excellent, has been achieved for biggish achievement using the work that its image carries out scientific research.
K-means algorithm is Classic Clustering Algorithms, and the image partition method using the algorithm is normal image dividing method
One of.Compared with current another common multi-resolution segmentation method, K-means method is empirical with not needing to be arranged
Multi-resolution segmentation parameter and other relevant parameters.The principle of K-means partitioning algorithm is to utilize pixel to random setting
Target's center distance optimization process, with constantly iteration and adjustment cluster centre is calculated after distance, and finally by pixel
Different cluster centres is converged to, to realize the effect of image segmentation.
The process of K-means partitioning algorithm is as follows:
1) image is converted, is transformed to the rectangle data format suitable for the operation of K-means method；
2) 6 pixels are randomly selected in the matrix after converting image as random center, clearly participate in calculating all
The center of image and this 6 pixel position consistencies；
3) characteristic distance calculating is carried out to the random center to each pixel, the characteristic distance used is several for the Europe of pixel value
In distance；
4) each pixel is classified as to the random center of minimum distance；
5) center of each classification is readjusted；
6) 3~5 step of iteration, until the number of iterations is greater than the change rate at 300 or the relatively last round of center in new center
Less than 0.05%, algorithm terminates.
2, the image recognition stage
Neural network is the large-scale parallel and distributed process device being made of multiple simple process members, has and deposits
Storage Heuristics is simultaneously allowed to available characteristic.Neural network have non-linear, adaptability, fault-tolerance, analysis and design it is consistent
The features such as property.BP neural network is a kind of feedforward neural network of multilayer, structure as shown in figure 3, be divided into input layer, hidden layer,
Output layer, the back transfer of positive transmitting and error comprising information.Multiple features of input layer input sample data, output layer
The classification for exporting sample data contains several hidden layers between input layer and output layer.
Assuming that one three layers of BP neural network, as shown in figure 3, input layer number is n, output layer number of nodes is m, hidden
Hiding node layer number is h, and wherein n and m can determine that h can select suitably to be worth according to formula 1 by sample data.
Wherein, c is the positive integer between 1~10.
Assuming that input layer is x1, x2... xn, output node layer is y1, y2…ym, hiding node layer is s1, s2…sh,
Input layer xiWith hiding node layer sjBetween weight be wij, hide node layer sjWith output node layer ykBetween weight
For wjk, then transmitted according to the forward direction of information, can must hide node layer s respectively according to formula 2 and formula 3hWith output node layer
yk:
Wherein, ajFor the biasing of input layer to hidden layer, function f is activation primitive of the input layer to hidden layer, bkIt is hiding
Layer arrives the biasing of output layer, and function g is activation primitive of the hidden layer to output layer.In the present invention, sigmoid letter is used respectively
Number (formula 4) and activation primitive with softmax function (formula 5) as input layer to hidden layer and hidden layer to output layer:
Wherein, z1, z2...zhFor one group of sequence of values.
BP neural network makes to minimize the error by back transfer error.Assuming that error function is E, pass through formula 6, public affairs
Formula 7 and formula 8 update weight w, obtain the weight so that minimizing the error, and similarly, can obtain the biasing for minimizing error, thus
Obtain neural network model.Error function is using intersection entropy function (formula 9).
W=w+ η Δ w formula 8
Wherein, η is learning efficiency, η ∈ (0,1)
Wherein t is true value, and p is predicted value, and j is classification
The remote sensing image in 2018 (0.51 meter of spatial resolution) for choosing Suzhou City of Jiangsu Province Xiangcheng District pond-inputting community is made
For the sample of experiment, remote sensing influence have passed through the pretreatment such as geometric correction and registration, as shown in fig. 4 a.The training of vegetation identification model
Process is as shown in figure 5, preferred manually visualize the vegetation pixel chosen in pond-inputting community, as shown in Figure 4 b.Then 64*64 is used
Window pond-inputting community remote sensing image is scanned, according to statistics, determine whether the window area is vegetation area, i.e., if
The pixel ratio for belonging to vegetation in window area is greater than 50%, then determines window area for vegetation area, and as engineering
Otherwise the positive sample of habit is determined as that the window is nonvegetated area domain, and the negative sample as machine learning；Finally, by positive and negative sample
The 64*64 pixel of this window inputs neural network input layer, carries out model training using BP neural network algorithm, obtains
Vegetation identification model.
The process of Remote Sensing Image Segmentation region recognition uses the window of 64*64 as shown in fig. 6, for each cut zone
Cut zone is scanned, and whether belongs to vegetation using vegetation identification model identification window, if belonged in cut zone
The area of vegetation is greater than 50%, then determines the cut zone for vegetation area, otherwise, it is determined that being nonvegetated area domain.According to vegetation
It is as shown in Figure 7 that the new wound bamboo plantation vegetation in 2018 that identification model obtains extract result.
3, multi-temporal data fusing stage
When being extracted using single remote sensing image to vegetation, since the factors such as remote sensing image spectrum, shooting angle are made
At the more difficult amendment of error, as shown in the black box region in Fig. 7；Plant in the vegetation blocked by house and shaded area
It cannot preferably be identified, in addition, the part eaves region in the remote sensing image is missed since spectrum and vegetation are closer to
It is identified as vegetation area.
From the angle of social engineering, it is assumed that residential block vegetation is smaller in 3 year-end drawdown levels, and based on this, comprehensive
It closes in November, 2016, in July, 2017, in October, 2017 and in July, 2018, the remote sensing image of totally 4 phases, and using based on power
The voting rule of weight merges multi-temporal data, determines whether the object in remote sensing image is vegetation area.In view of recent
Remote sensing image extracts result to vegetation and is affected, and assigns recent remote sensing image biggish weight, determines in remote sensing image
The object whether be vegetation voting rule it is as shown in formula 10, wherein yeariIndicate i-th of remote sensing image corresponding time,
Indicate whether the object in i-th of remote sensing image is vegetation area.When the result calculated according to formula 10 is greater than 0.5, sentence
The fixed object is vegetation area, otherwise, it is determined that being nonvegetated area domain.
4, the analysis of result accuracy rate is extracted
The hardware environment that the present embodiment uses is respectively: operating system is 7 professional version of Windows, the exploitation language used
Speech is Python3.6, and CPU is Intel Core i7, inside saves as 16G, hard disk position PCIe SSD, video card Geoforce
GTX1060.It is as shown in Figure 8 a that result is extracted based on the fused new wound bamboo plantation vegetation of multi-temporal data；It manually visualizes and extracts result
As shown in Figure 8 b；Extraction result using only the remote sensing image in July, 2018 is as shown in Figure 7；It is new to create the distant of bamboo plantation in July, 2018
Image is felt as shown in figure a.
The formula for calculating extraction result preparation rate is as shown in formula 11, wherein SiWhether the object i in expression Visual Outcomes
It is then S for vegetation areaiIt is 1, is otherwise 0, PiIndicate whether the fused object i of multi-temporal data is vegetation area, is
Then PiIt is 1, is otherwise object sum for 0, n.
The method based on BP neural network is calculated according to formula 11, using only the extraction knot of remote sensing images in 2018
Fruit accuracy rate is 87.6%, the use of the fused accuracy rate for extracting result of multi-temporal data is 93.3%.The experimental results showed that
Multi-temporal remote sensing image residential block vegetation extracting method neural network based has certain reality for the discrimination for improving vegetation
Border effect.
Claims (9)
1. a kind of multi-temporal remote sensing image urban vegetation extracting method neural network based, which is characterized in that be based on multidate
High-definition remote sensing image data, by carrying out image segmentation to image；Sample is trained using BP neural network method,
Form the neural network model for being directed to test block vegetation；Vegetation extraction is carried out to multi-temporal data using the model, using being based on
The voting rule fusion multidate of weight extracts as a result, to obtain the higher vegetation area of precision.
2. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1, feature
It is, concrete operations are as follows:
(1) multi-temporal remote sensing image data are divided into multiple objects by the image segmentation stage；
(2) the image recognition stage is trained sample using BP neural network method, obtains vegetation identification model；Use this
Model identifies each object, if the area for belonging to vegetation in cut zone is greater than 50%, determines the cut zone
For vegetation area, otherwise, it is determined that being nonvegetated area domain；
(3) multi-temporal data fusing stage assigns recent remote sensing image biggish weight, pair after determining Remote Sensing Image Segmentation
As if the no voting rule for vegetation area uses following formula,
Wherein, yeariIndicate i-th of remote sensing image corresponding time, PiIndicate the object in i-th of remote sensing image whether be
Vegetation area；Work as PFusionWhen greater than 0.5, the object is determined for vegetation area, otherwise, it is determined that being nonvegetated area domain.
3. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
Sign is, the image that the remote sensing image data is issued from Google Earth.
4. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
Sign is that the spatial resolution of the remote sensing image is 0.51 meter, and by pretreatment, the pretreatment is geometric correction and matches
It is quasi-.
5. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
Sign is that for described image dividing method using K-means clustering method, process is as follows:
1) image is converted, is transformed to the rectangle data format suitable for the operation of K-means method；
2) 6 pixels are randomly selected in the matrix after converting image as random center, clearly participate in all images calculated
Center and this 6 pixel position consistencies；
3) characteristic distance calculating is carried out to the random center to each pixel；
4) each pixel is classified as to the random center of minimum distance；
5) center of each classification is readjusted；
6) iteration 3)~5) step, until the number of iterations is greater than 300 or the change rate at the relatively last round of center in new center is small
In 0.05%, algorithm terminates.
6. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
The step of sign is, the acquisition vegetation identification model include:
(1) sample area is determined, by manually visualizing the vegetation pixel chosen in sample area；
(2) training sample area remote sensing image is scanned using the window of 64*64, according to statistics, determines that the window area is
No is vegetation area, i.e., if the pixel ratio for belonging to vegetation in window area is greater than 50%, determines window area to plant
By region and as the positive sample of machine learning, otherwise it is determined as that the window is nonvegetated area domain, and as the negative of machine learning
Sample；
(3) by 64*64 pixel of positive and negative sample window, input neural network input layer, using BP neural network algorithm into
Row model training obtains vegetation identification model.
7. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 5, feature
It is, the step 3) characteristic distance uses the Euclidean distance of pixel value.
8. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
Sign is that the test block is city neighborhood, shopping centre or industrial area.
9. multi-temporal remote sensing image urban vegetation extracting method neural network based according to claim 1 or 2, special
Sign is, carries out accuracy rate calculating to the extraction result using following formula:
Wherein, SiIndicate whether the object i of Visual Outcomes is vegetation area, is then SiIt is 1, is otherwise 0；PiIndicate multidate number
Whether it is vegetation area according to fused object i, is then PiIt is 1, is otherwise object sum for 0, n.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910135591.6A CN109871812A (en) | 2019-02-22 | 2019-02-22 | A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910135591.6A CN109871812A (en) | 2019-02-22 | 2019-02-22 | A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109871812A true CN109871812A (en) | 2019-06-11 |
Family
ID=66919173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910135591.6A Pending CN109871812A (en) | 2019-02-22 | 2019-02-22 | A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871812A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582156A (en) * | 2020-05-07 | 2020-08-25 | 武汉大势智慧科技有限公司 | Oblique photography-based tall and big vegetation extraction method for urban three-dimensional model |
| CN114019296A (en) * | 2021-09-26 | 2022-02-08 | 广西电网有限责任公司电力科学研究院 | Distribution line ground fault identification method based on BP neural network |
| CN114490449A (en) * | 2022-04-18 | 2022-05-13 | 飞腾信息技术有限公司 | Memory access method and device and processor |
| CN114529721A (en) * | 2022-02-08 | 2022-05-24 | 山东浪潮科学研究院有限公司 | Urban remote sensing image vegetation coverage identification method based on deep learning |
| CN114627389A (en) * | 2022-03-23 | 2022-06-14 | 中国科学院空天信息创新研究院 | Raft culture area extraction method based on multi-temporal optical remote sensing image |
| CN114529721B (en) * | 2022-02-08 | 2024-05-10 | 山东浪潮科学研究院有限公司 | Urban remote sensing image vegetation coverage recognition method based on deep learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108288014A (en) * | 2017-01-09 | 2018-07-17 | 北京四维图新科技股份有限公司 | Intelligent road extracting method and device, extraction model construction method and hybrid navigation system |
| CN108399423A (en) * | 2018-02-01 | 2018-08-14 | 南京大学 | A kind of multidate-Combining Multiple Classifiers of classification of remote-sensing images |
-
2019
- 2019-02-22 CN CN201910135591.6A patent/CN109871812A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108288014A (en) * | 2017-01-09 | 2018-07-17 | 北京四维图新科技股份有限公司 | Intelligent road extracting method and device, extraction model construction method and hybrid navigation system |
| CN108399423A (en) * | 2018-02-01 | 2018-08-14 | 南京大学 | A kind of multidate-Combining Multiple Classifiers of classification of remote-sensing images |
Non-Patent Citations (3)
| Title |
|---|
| 张恩泽: "面向对象的遥感图像识别研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 申怀飞: "基于遥感多时相影像的南海北部湾地区地物分类与识别方法研究", 《中国博士学位论文全文数据库 基础科学辑》 * |
| 蒋锦刚等: "多时相遥感影像提取湖泊边界信息的融合算法", 《湖泊科学》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582156A (en) * | 2020-05-07 | 2020-08-25 | 武汉大势智慧科技有限公司 | Oblique photography-based tall and big vegetation extraction method for urban three-dimensional model |
| CN111582156B (en) * | 2020-05-07 | 2023-12-05 | 武汉大势智慧科技有限公司 | High and large vegetation extraction method based on oblique photography urban three-dimensional model |
| CN114019296A (en) * | 2021-09-26 | 2022-02-08 | 广西电网有限责任公司电力科学研究院 | Distribution line ground fault identification method based on BP neural network |
| CN114529721A (en) * | 2022-02-08 | 2022-05-24 | 山东浪潮科学研究院有限公司 | Urban remote sensing image vegetation coverage identification method based on deep learning |
| CN114529721B (en) * | 2022-02-08 | 2024-05-10 | 山东浪潮科学研究院有限公司 | Urban remote sensing image vegetation coverage recognition method based on deep learning |
| CN114627389A (en) * | 2022-03-23 | 2022-06-14 | 中国科学院空天信息创新研究院 | Raft culture area extraction method based on multi-temporal optical remote sensing image |
| CN114627389B (en) * | 2022-03-23 | 2023-01-31 | 中国科学院空天信息创新研究院 | Raft culture area extraction method based on multi-temporal optical remote sensing image |
| CN114490449A (en) * | 2022-04-18 | 2022-05-13 | 飞腾信息技术有限公司 | Memory access method and device and processor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111666918B (en) | Coastline change identification method based on multiple factors | |
| CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN109871812A (en) | A kind of multi-temporal remote sensing image urban vegetation extracting method neural network based | |
| CN101271469B (en) | Two-dimension image recognition based on three-dimensional model warehouse and object reconstruction method | |
| CN110458048A (en) | Take population distribution Spatio-temporal Evolution and the cognition of town pattern feature into account | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN100595782C (en) | Classification method for syncretizing optical spectrum information and multi-point simulation space information | |
| CN105787501B (en) | Power transmission line corridor region automatically selects the vegetation classification method of feature | |
| CN106845499A (en) | A kind of image object detection method semantic based on natural language | |
| CN110046572A (en) | A kind of identification of landmark object and detection method based on deep learning | |
| CN110263111A (en) | Land use based on priori knowledge/cover information Space monitoring method | |
| CN111598048A (en) | Urban village-in-village identification method integrating high-resolution remote sensing image and street view image | |
| CN109993803A (en) | The intellectual analysis and evaluation method of city tone | |
| CN111985389A (en) | Basin similarity discrimination method based on basin attribute distance | |
| CN110119740A (en) | A method of it is extracted from complicated urban road network and simplifies topological relation | |
| Wen et al. | Monitoring the green evolution of vernacular buildings based on deep learning and multi-temporal remote sensing images | |
| Zhou et al. | A large-scale mapping scheme for urban building from Gaofen-2 images using deep learning and hierarchical approach | |
| Xu et al. | A framework for land use scenes classification based on landscape photos | |
| CN114943902A (en) | Urban vegetation unmanned aerial vehicle remote sensing classification method based on multi-scale feature perception network | |
| CN110287798A (en) | Vector network pedestrian detection method based on characteristic module and context fusion | |
| CN112347926B (en) | High-resolution image city village detection method based on building morphology distribution | |
| Stark | Using deep convolutional neural networks for the identification of informal settlements to improve a sustainable development in urban environments | |
| CN114511787A (en) | Neural network-based remote sensing image ground feature information generation method and system | |
| Qu et al. | A modified self-adaptive method for mapping annual 30-m land use/land cover using Google Earth Engine: A case study of Yangtze River Delta | |
| Zhou et al. | Automatic responsive-generation of 3D urban morphology coupled with local climate zones using generative adversarial network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190611 |
|
| RJ01 | Rejection of invention patent application after publication |