CN109242922A - A kind of landform synthetic method based on radial primary function network - Google Patents

A kind of landform synthetic method based on radial primary function network Download PDF

Info

Publication number
CN109242922A
CN109242922A CN201810942069.4A CN201810942069A CN109242922A CN 109242922 A CN109242922 A CN 109242922A CN 201810942069 A CN201810942069 A CN 201810942069A CN 109242922 A CN109242922 A CN 109242922A
Authority
CN
China
Prior art keywords
landform
height
point
network
training
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
Application number
CN201810942069.4A
Other languages
Chinese (zh)
Other versions
CN109242922B (en
Inventor
全红艳
周双双
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
East China Normal University
Original Assignee
East China Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by East China Normal University filed Critical East China Normal University
Priority to CN201810942069.4A priority Critical patent/CN109242922B/en
Publication of CN109242922A publication Critical patent/CN109242922A/en
Application granted granted Critical
Publication of CN109242922B publication Critical patent/CN109242922B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The landform synthetic method based on radial primary function network that the invention discloses a kind of, this method utilize the terrain data block of digital elevation model (Digital Elevation Mode, vehicle economy M), study and extraction using radial primary function network to features of terrain；In landform synthesis, according to user's cartographical sketching, customization landform corresponding with user's sketch can be synthesized using the feature learnt in advance in conjunction with input altitude data.This method has the characteristics that simple, effective, can synthesize specific landform according to user's cartographical sketching.

Description

A kind of landform synthetic method based on radial primary function network
Technical field
The present invention relates to Virtual Simulation fields, and in particular to a kind of landform synthesis side based on radial primary function network Method, using the terrain data block of digital elevation model (Digital Elevation Mode, vehicle economy M), using radial base letter Study and extraction of the number network to features of terrain；In landform synthesis, according to user's cartographical sketching, in conjunction with input altitude data, Using the feature learnt in advance, customization landform corresponding with user's sketch can be synthesized.This method has simple, effective special Point can synthesize specific landform according to user's cartographical sketching.
Background technique
Landform is the principal visual element in three-dimensional virtual scene, since its extensive use in real life is worth, So that landform synthetic technology becomes the research hotspot of computer vision field.Either prevent in natural calamity or in video display In game creation, realistic terrain can preferably improve the usage experience of user.Landform synthetic technology can substantially be divided at present Three classes: the modeling method of Kernel-based methods, the modeling method based on physical erosion and the modeling method based on user's sketch.In recent years Come, with the universal use of DEM terrain data, people start to combine real terrain data characteristics, study the modeling based on sketch Method, and controlled by sketch, realize the customization function of user.
Meanwhile in order to realize that the intelligence of landform is combined to, people is studied in machine learning in recent years and deep learning rapid development Member attempts manually intelligent method and understands the implicit features in real terrain data, and the ground based on convolutional neural networks occurs Shape synthetic method, and realize and contour line progress depth information prediction in mountain range is inputted to user, improve the reasonable of synthesis landform Property.But landform is synthesized using deep learning method, existing main problem is exactly that structure is complicated for network, network parameter instruction The problem of white silk is difficult to restrain, these are all current intelligent landform study on the synthesis.
Summary of the invention
The purpose of the present invention is in view of the deficiencies of the prior art, and for the practical problem in landform synthesis, one kind is proposed Landform synthetic method based on radial primary function network, this method are utilized according to user's cartographical sketching in conjunction with input altitude data The feature learnt in advance can synthesize customization landform corresponding with user's sketch.This method have the characteristics that it is simple, effective, Specific landform can be synthesized according to user's cartographical sketching.
Realizing the specific technical solution of the object of the invention is:
A kind of landform synthetic method based on radial primary function network, feature be this method comprising the following specific steps
Step 1: preparing altitude data block
The altitude data block D of WGS84 coordinate system is downloaded from the website SRTM http://srtm.csi.cgiar.org, D's Spatial resolution is 90m × 90m between 200m × 200m, and the number of pixel is N in D, and the height of arbitrary point A is denoted as HA；
Step 2: establish training dataset:
(1) firstly, calculating the entropy E of arbitrary point AA:
Wherein, pkIt is that (its height is H k-th point in 3 × 3 neighborhood T of arbitrary point Ak) height distribution, k=1,2 ..., 9, pkIt calculates are as follows:
Wherein,It is the height statistics of 9 neighborhood points in T, η is the height minima of 9 points in T；δ For constant, 0.0001, H is takent(t=1,2 ..., 9) indicates the height of t-th of neighborhood point in T；
(2) feature vector of A: V is establishedA=(EA,HA), further using the feature vector construction feature of each point in D to Quantity set S={ (Es,Hs)|1≤s≤N}；
(3) M feature vector C is randomly selected from Se=(Ee,He) (e=1,2 ..., M), as M cluster centre, 3≤ The feature vector in S is gathered using K-means method for M class: L M≤5f(f=1,2 ..., M)；
(4) respectively to all kinds of LfThe entropy of pixel particles carries out summation statistics in (f=1,2 ..., M), will have maximum statistics The class of entropy is denoted as Lm(1≤m≤M), by LmRegard salient region as；
(5) L is usedmEstablish data set: LmThe corresponding pixel particles collection of middle feature vector is denoted as G, using D8 algorithm, using 3 × The skeleton B of 3 window calculation G specifically using 3 × 3 neighborhoods around pixel, calculates 8 pixels around center and its neighborhood It is poor to make, using the maximum direction of difference as skeleton direction.The set of eigenvectors of particle is U={ (d in Gg,hg) | 1≤g≤Q }, dgIt is The shortest distance of Arbitrary Particles j to B, h in GgFor the height of j, Q indicates the number of data element in G；
Step 3: building radial primary function network carries out learning training
Establish the radial primary function network X of three layers of propagated forward comprising input layer, hidden layer and output layer: when training, Using supervised learning method, network inputs are dg, the monitoring data of output is hg, g=1,2 ..., Q；In hidden layer, radial base Function use Gaussian function, the training parameter of model include: the number of Gaussian kernel, hidden layer neuron center (basic function Center), the weight between variance and hidden layer and output layer；Trained loss function is defined as:
Wherein, PgIt is the pre-computed altitude of X network output as a result, hgAs the monitoring data of height, 1≤g≤Q；
Training process: the number K of Gaussian kernel (a) is initialized as 5；(b) training X, and loss letter is calculated using formula (3) Number F, if F is greater than threshold value 0.01, the number K=K+5 of Gaussian kernel, and goes to step (b) and is trained next time；Otherwise, such as Fruit F is less than or equal to threshold value 0.01, and training terminates；
Step 4: landform synthesis
User inputs bianry image V as cartographical sketching, and V draws the sketch of user using drawing software, and saves as two Be worth image, using V be used as skeleton, calculating V on any point to skeleton distance Jb(b=1,2 ..., x), x are indicated in synthesis landform The number of pixel；
By Jb(b=1,2 ..., x) is input in network X, and using in X network, trained parameter has been predicted, is obtained To the height value of each point of synthesis landform, the landform synthesized in this way.
The present invention has the characteristics that simple, practical, according to user's cartographical sketching, in conjunction with input altitude data, can synthesize Specific customized landform.
Detailed description of the invention
Fig. 1 is the result figure of salient region of the present invention detection；
Fig. 2 is the result figure of landform of the present invention synthesis.
Specific embodiment
Embodiment
The following further describes the present invention with reference to the drawings.
The present embodiment is implemented under 64 bit manipulation system of Windows10 in PC machine, and hardware configuration is processor CoreTMI5-7500 3.4GHz CPU, 8GB memory, software environment are Matlab 2015b, and programming uses Python Language, in conjunction with vision open source library OpenCV 2.4.4 and open source raster spatial data transformation warehouse GDAL.
Specific embodiment of the invention scheme is:
A kind of landform synthetic method based on radial primary function network, this method comprising the following specific steps
Step 1: preparing altitude data block
The altitude data block D of WGS84 coordinate system is downloaded from the website SRTM http://srtm.csi.cgiar.org, D's Spatial resolution is between 90m × 90m, and the number of pixel is N in D, and the height of arbitrary point A is denoted as HA；
Step 2: establish training dataset:
(1) firstly, calculating the entropy E of AA:
Wherein, pkIt is that (its height is H k-th point in 3 × 3 neighborhood T of Ak) height distribution, k=1,2 ..., 9, pkIt calculates Are as follows:
Wherein,It is the height statistics of 9 neighborhood points in T, η is the height minima of 9 points in T；δ For constant, 0.0001, H is takent(t=1,2 ..., 9) indicates the height of t-th of neighborhood point in T；
(2) feature vector of A: V is establishedA=(EA,HA), further using the feature vector construction feature of each point in D to Quantity set S={ (Es,Hs)|1≤s≤N}；
(3) 3 feature vector C are randomly selected from Se=(Ee,He) (e=1,2,3), as 3 cluster centres, utilization K-means method gathers for 3 classes the feature vector in S: Lf(f=1,2,3)；
(4) respectively to all kinds of LfThe entropy of pixel particles carries out summation statistics in (f=1,2,3), will have maximum statistical entropy Class be denoted as Lm(1≤m≤3), by LmRegard salient region as；
(5) L is usedmEstablish data set: LmThe corresponding pixel particles collection of middle feature vector is denoted as G, using D8 algorithm, using 3 × The set of eigenvectors of particle is U={ (d in the skeleton B, G of 3 window calculation Gg,hg) | 1≤g≤Q }, dgIt is that Arbitrary Particles j is arrived in G The shortest distance of B, hgFor the height of j, Q indicates the number of data element in G；
Step 3: building radial primary function network carries out learning training
Establish the radial primary function network X of three layers of propagated forward comprising input layer, hidden layer and output layer: when training, Using supervised learning method, network inputs are dg, the monitoring data of output is hg, g=1,2 ..., Q；In hidden layer, radial base Function use Gaussian function, the training parameter of model include: the number of Gaussian kernel, hidden layer neuron center (basic function Center), the weight between variance and hidden layer and output layer；Trained loss function is defined as:
Wherein, PgIt is the pre-computed altitude of X network output as a result, hgAs the monitoring data of height, 1≤g≤Q；
Training process: the number K of Gaussian kernel (a) is initialized as 5；(b) training X, and loss letter is calculated using formula (3) Number F, if F is greater than threshold value 0.01, the number K=K+5 of Gaussian kernel, and goes to step (b) and is trained next time；Otherwise, such as Fruit F is less than or equal to threshold value 0.01, and training terminates；
Step 4: landform synthesis
User inputs bianry image V as cartographical sketching, and V draws the sketch of user using drawing software, and saves as two Be worth image, using V be used as skeleton, calculating V on any point to skeleton distance Jb(b=1,2 ..., x), x are indicated in synthesis landform The number of pixel；
By Jb(b=1,2 ..., x) is input in network X, and using in X network, trained parameter has been predicted, is obtained To the height value of each point of synthesis landform, the landform synthesized in this way.
Fig. 1 is the result that salient region detection is carried out according to DEM terrain data.3 groups of experimental results are given in figure, often One one group of behavior result；In every row, left side the 1st is classified as the altitude data block of input, the conspicuousness area as the result is shown on right side Domain detection as a result, it can be seen from the figure that salient region corresponds to the detail areas of DEM terrain block, mentioned from salient region Feature is taken, data set is created；
Fig. 2 is to input the result that sketch carries out landform synthesis according to user.3 groups of experimental results, every a line are given in figure For one group of result；In every row, the 1st column result of left side is the altitude data block of input, and centre one is classified as the sketch of user's input； The 1st column result of right side is the terrain result of synthesis, from the result of synthesis, it is apparent that method proposed by the invention can be with The synthesis of realistic terrain is realized according to user's sketch, method is simple and convenient.

Claims (1)

1. a kind of landform synthetic method based on radial primary function network, which is characterized in that this method comprising the following specific steps
Step 1: preparing altitude data block
From the space of altitude data the block D, D of the website SRTM http://srtm.csi.cgiar.org downloading WGS84 coordinate system Resolution ratio is 90m × 90m between 200m × 200m, and the number of pixel is N in D, and the height of arbitrary point A is denoted as HA；
Step 2: establish training dataset:
(1) firstly, calculating the entropy E of arbitrary point AA:
Wherein, pkIt is k-th point in 3 × 3 neighborhood T of arbitrary point A of height distribution, the height of k point is Hk；K=1,2 ..., 9, pk It calculates are as follows:
Wherein,It is the height statistics of 9 neighborhood points in T, η is the height minima of 9 points in T；δ is normal Amount, takes 0.0001, Ht(t=1,2 ..., 9) indicates the height of t-th of neighborhood point in T；
(2) feature vector of A: V is establishedA=(EA,HA), further utilize the feature vector construction feature vector set of each point in D S={ (Es,Hs) | 1 ￡ s ￡ N }；
(3) M feature vector C is randomly selected from Se=(Ee,He) (e=1,2 ..., M), as M cluster centre, 3≤M≤ 5, using K-means method, the feature vector in S is gathered for M class: Lf(f=1,2 ..., M)；
(4) respectively to all kinds of LfThe entropy of pixel particles carries out summation statistics in (f=1,2 ..., M), will have the class of maximum statistical entropy It is denoted as Lm(1 ￡ m ￡ M), by LmRegard salient region as；
(5) L is usedmEstablish data set: LmThe corresponding pixel particles collection of middle feature vector is denoted as G, using D8 algorithm, using 3 × 3 windows The set of eigenvectors that mouth calculates particle in the skeleton B, G of G is U={ (dg,hg) | 1 ￡ g ￡ Q }, dgIt is Arbitrary Particles j to B in G The shortest distance, hgFor the height of j, Q indicates the number of data element in G；
Step 3: building radial primary function network carries out learning training
It establishes the radial primary function network X of three layers of propagated forward comprising input layer, hidden layer and output layer: when training, using Supervised learning method, network inputs are dg, the monitoring data of output is hg, g=1,2 ..., Q；In hidden layer, radial basis function Using Gaussian function, the training parameter of model includes: in the center i.e. basic function of the number of Gaussian kernel, hidden layer neuron Weight between the heart, variance and hidden layer and output layer；Trained loss function is defined as:
Wherein, PgIt is the pre-computed altitude of X network output as a result, hgAs the monitoring data of height, 1 ￡ g ￡ Q；
Training process: the number K of Gaussian kernel (a) is initialized as 5；(b) training X, and loss function F is calculated using formula (3), If F is greater than threshold value 0.01, the number K=K+5 of Gaussian kernel, and goes to step (b) and trained next time；Otherwise, if F is small In or equal to threshold value 0.01, trains and terminate；
Step 4: landform synthesis
User inputs bianry image V as cartographical sketching, and V draws the sketch of user using drawing software, and saves as binary map Picture calculates on V any point to the distance J of skeleton using V as skeletonb(b=1,2 ..., x), x indicate pixel in synthesis landform Number；
By Jb(b=1,2 ..., x) is input in network X, and using in X network, trained parameter has been predicted, is synthesized The height value of each point of landform, the landform synthesized.
CN201810942069.4A 2018-08-17 2018-08-17 Terrain synthesis method based on radial basis function network Active CN109242922B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810942069.4A CN109242922B (en) 2018-08-17 2018-08-17 Terrain synthesis method based on radial basis function network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810942069.4A CN109242922B (en) 2018-08-17 2018-08-17 Terrain synthesis method based on radial basis function network

Publications (2)

Publication Number Publication Date
CN109242922A true CN109242922A (en) 2019-01-18
CN109242922B CN109242922B (en) 2023-04-07

Family

ID=65071520

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810942069.4A Active CN109242922B (en) 2018-08-17 2018-08-17 Terrain synthesis method based on radial basis function network

Country Status (1)

Country Link
CN (1) CN109242922B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112632869A (en) * 2020-12-23 2021-04-09 中国空气动力研究与发展中心计算空气动力研究所 Grid frame-based non-structural boundary layer grid generation method

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140270489A1 (en) * 2013-03-12 2014-09-18 Microsoft Corporation Learned mid-level representation for contour and object detection
CN107122396A (en) * 2017-03-13 2017-09-01 西北大学 Three-dimensional model searching algorithm based on depth convolutional neural networks

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140270489A1 (en) * 2013-03-12 2014-09-18 Microsoft Corporation Learned mid-level representation for contour and object detection
CN107122396A (en) * 2017-03-13 2017-09-01 西北大学 Three-dimensional model searching algorithm based on depth convolutional neural networks

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵鹏;刘杨;刘慧婷;姚晟;: "基于深度卷积-递归神经网络的手绘草图识别方法" *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112632869A (en) * 2020-12-23 2021-04-09 中国空气动力研究与发展中心计算空气动力研究所 Grid frame-based non-structural boundary layer grid generation method

Also Published As

Publication number Publication date
CN109242922B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
EP3944200B1 (en) Facial image generation method and apparatus, device and storage medium
CN109196526B (en) Method and system for generating multi-modal digital images
CN113469356B (en) Improved VGG16 network pig identity recognition method based on transfer learning
Xie et al. Learning energy-based spatial-temporal generative convnets for dynamic patterns
CN107229904A (en) A kind of object detection and recognition method based on deep learning
CN110419049A (en) Room layout estimation method and technology
CN108537818A (en) Crowd's trajectory predictions method based on cluster pressure LSTM
CN111199175A (en) Training method and device for target detection network model
CN112669448B (en) Virtual data set development method, system and storage medium based on three-dimensional reconstruction technology
CN108151728A (en) A kind of half dense cognitive map creation method for binocular SLAM
CN110136075A (en) It is a kind of to recycle the remote sensing image defogging method for generating confrontation network based on edge sharpening
WO2019207577A1 (en) System and method for training a machine-learning model to identify real-world elements
Wang et al. Spatiotemporal subpixel mapping of time-series images
CN109658508B (en) Multi-scale detail fusion terrain synthesis method
CN109598771A (en) A kind of landform synthetic method of more geomorphic feature constraints
CN114463837A (en) Human behavior recognition method and system based on self-adaptive space-time convolution network
CN104021395B (en) Target tracing algorithm based on high-order partial least square method
Salem et al. Semantic image inpainting using self-learning encoder-decoder and adversarial loss
Zhou et al. Image2GIF: Generating cinemagraphs using recurrent deep q-networks
CN116994022A (en) Object detection method, model training method, device, electronic equipment and medium
Ahmed et al. Robust Object Recognition with Genetic Algorithm and Composite Saliency Map
CN113673369A (en) Remote sensing image scene planning method and device, electronic equipment and storage medium
CN109242922A (en) A kind of landform synthetic method based on radial primary function network
CN113628349B (en) AR navigation method, device and readable storage medium based on scene content adaptation
CN113205028A (en) Pedestrian detection method and system based on improved YOLOv3 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