CN106372618A - Road extraction method and system based on SVM and genetic algorithm - Google Patents
Road extraction method and system based on SVM and genetic algorithm Download PDFInfo
- Publication number
- CN106372618A CN106372618A CN201610834791.7A CN201610834791A CN106372618A CN 106372618 A CN106372618 A CN 106372618A CN 201610834791 A CN201610834791 A CN 201610834791A CN 106372618 A CN106372618 A CN 106372618A
- Authority
- CN
- China
- Prior art keywords
- picture
- road
- training
- module
- svm
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a road extraction method and system based on an SVM and a genetic algorithm and belongs to the technical field of road extraction in a picture. The method comprises the following steps: finding one proper threshold for each picture through the genetic algorithm; segmenting roads and non-roads in the pictures by use of a threshold segmentation method; determining whether the roads are identified, if so, outputting processed pictures, and ending the process; according to training types, i.e., the roads and the non-roads, training data; performing SVM classification on the pictures; and outputting the processed pictures. The invention further provides a system for realizing the method. The method and system have the following advantages: by use of the road extraction method combining the SVM with the genetic algorithm, the method improves picture types of road extraction and also improves the road identification accuracy at the same time; and the problem of restrictions by use of a machine learning algorithm training set is solved, and the road extraction speed is also improved.
Description
Technical field
The present invention relates to the road extraction technology in picture, more particularly, to a kind of carried based on the road of svm and genetic algorithm
Take method and system.
Background technology
The road extraction of traditional method includes the following:
(1) method based on mathematical morphology
Basic morphological operations are corrosion and expand, two-value morphological dilation can be converted into the logical operationss of set with corrosion,
Algorithm is simple, is suitable to parallel processing, and is easy to hardware realization, is suitable to carry out image segmentation to bianry image, refinement, extracts bone
Frame, edge extracting, shape analysis.
Idiographic flow is: the first step is the median filter of the two dimension with 5*5, then passes through adaptive Threshold segmentation
Picture is converted into two-value picture by method;Process some noises remaining with opening and closing operations;Finally obtained with Refinement operation
The picture of road network.
But this method for specific picture recognition effect very well, but for frame of video, recognition effect
It is not fine.Although there are many similar picture in frame of video, once for crossing over big scene, this method is just
Be not suitable for.
(2) method of Hough transformation
Hough transformation is to identify one of basic skills of geometry in image procossing from image, and application is very extensive,
There are a lot of innovatory algorithm.
The advantage of Hough transformation is detection of straight lines, its principle: straight line is the set of series of discrete point in the picture,
By the discrete polar coordinate formula of a straight line, the discrete point geometric equality that can give expression to straight line is as follows:
X*cos θ+y*sin θ=r
Wherein angle theta refers to the angle between r and x-axis, and r is to rectilinear geometry vertical dimension.Any in Points on Straight Line,
X, y can express, wherein r, and theta is constant.
But in the image processing field realized, pixel coordinate p (x, y) of image is known, and r, theta is then intended to
The variable found.If each (r, theta) value can be drawn according to pixel point coordinates p (x, y) value, then just from image flute
Karr origin coordinate system transform is to polar coordinate hough space system, this Hough transformation being referred to as straight line from the change putting curve.
Conversion is limited value interval decile or cumulative grid by quantifying Hough parameter space.When Hough transformation algorithm starts, often
Individual pixel coordinate point p (x, y) is switched to above the curve point of (r, theta), is added to corresponding grid data point, when one
When crest occurs, illustrate with the presence of straight line.
(3) method of p non-parametric segmentation
When target and the histogram distribution of background have necessarily overlapping, the trough between two crests is very inconspicuous.If adopting
With the overall situation, poor effect.If being known a priori by ratio p that target accounts for whole image, p parametric method can be adopted.P parametric method is concrete
Step is as follows: assume to be known a priori by target to account for the ratio of whole image is p, and target is partially dark, and background is partially bright:
A). histogram distribution p (t) of calculating image, t=0,1 ... ..255;
B). calculate threshold value t so as to meetMinimum.
Here the principle of p non-parametric segmentation: select the road area that some are different, then calculate the average ash of selected portion
Angle value.If image pixel intensities are less than this value, this pixel value is set to 0, and otherwise this pixel value just belongs to the pixel of road.
P value segmentation is found certain threshold value to extract road.When using the method, have selected several different road areas, and count
Calculate the average gray value of selected part.If the intensity of a pixel is less than this value, pixel is arranged to zero, otherwise for representing
One pixel of road.P value method processed after image after, need further perfect, using image processing function.Open behaviour
Make symbol and can remove little noise as a good method.Afterwards, it is possible to use morphological operation extracts road network.
For the extraction of road information, due to not single, the discordance of video scene, with above-mentioned pure digi-tal image procossing
Method certainly can not be general, pure digi-tal image procossing is directed to specific scene effectively, but is aimed at universal field
For scape, the effect of extraction is not so notable.
Content of the invention
For solving the problems of the prior art, the present invention provides a kind of method for extracting roads based on svm and genetic algorithm,
Additionally provide a kind of system realizing said method.
The present invention is comprised the steps: based on the method for extracting roads of svm and genetic algorithm
S1: each picture is found a suitable threshold values by genetic algorithm;
S2: split road and the non-rice habitats in picture with the method for threshold segmentation;
S3: judge whether to identify road, if it is, execution step s6, if not, execution step s4;
S4: according to training classification: road and non-rice habitats, training data;
S5: picture is carried out svm classification;
S6: export the picture handled well.
The present invention is further improved, and in step s3, if identifying road, before execution step s6, also includes sentencing
Disconnected step a: judge whether the road identifying has barrier, if it is, according to training classification: road and barrier, training data,
Will identify that the picture of road carries out svm classification again, then execution step s6, if not, direct execution step s6.
The present invention is further improved, in step s1, the frame of video of the video that described picture shoots for unmanned plane.
The present invention is further improved, and after the execution of step s6, also includes Video Composition step b: output is handled well
Picture i.e. frame of video synthetic video, terminate.
The present invention is further improved, and in step s1, described genetic algorithm comprises the steps:
S11: random initializtion population, arranges evolutionary generation n;
S12: calculate individual fitness;
S13: select the individuality of pairing by fitness;
S14: intersect and produce new individuality, then new individuality is added to colony;
S15: setting probability and condition, if meeting condition, individuality makes a variation according to this probability, if being unsatisfactory for condition,
Then execution step s12;
S16: when colony's maximum adaptation degree continuous multi-generation is constant or has evolved n generation, terminate evolving.
The present invention is further improved, and the fitness function of the object function of described genetic algorithm and calculating fitness is by linear
Degree clWith adjacency prDetermined, the linearity and adjacency are defined as follows:
cl=1/ [(tana)2+(tanb)2+(tanc)2], (alpha+beta r) >=t1
pr=l2/(2πr2)≥t2
Wherein, l is the length in two straight-line segment compared with short straight line section, and r is the intersegmental beeline of two straight lines, a and b is respectively
Two straightways and the angle of r, c is the angle between two straightways, a and β is constant undetermined in experiment, and t1, t2 are specific
Thresholding, determines in test.
The present invention is further improved, and in step s5, what described svm classified comprises the steps:
S51: start, input picture to be processed;
S52: picture segmentation is become some fritters;
S53: fritter is individually placed to different files according to training classification;
S54: the training classification extracted will be needed as positive example, other training classification fritters are as counter-example;
S55: the support vector machine with kernel function for the training, obtain svm model;
S56: the positive example labelling output in picture terminates.
Present invention also offers a kind of system realizing said method, including threshold values computing module: for by each picture
One suitable threshold values is found by genetic algorithm;Picture segmentation module: split the road in picture with the method for threshold segmentation
Road and non-rice habitats;First judge module: it is used for judging whether to identify road, if it is, execution output module, if not, holding
Row training data module;Training data module: for according to training classification: road and non-rice habitats, training data;Svm classification mould
Block: for picture is carried out svm classification;Output module: for exporting the picture handled well,.
The present invention is further improved, after the first judge module execution, if identifying road, in execution output module
Before, also include the second judge module: whether the road for judging to identify has obstacle, if it is, according to training classification: road
And barrier, training data, then will identify that the picture of road carries out svm classification, then execute output module, if not, directly
Connect execution output module.
The present invention is further improved, in threshold values computing module, the video of the video that described picture shoots for unmanned plane
Frame, after output module execution, also includes Video Composition module, and picture i.e. frame of video for handling output well synthesize
Video.
Compared with prior art, the invention has the beneficial effects as follows: the road extraction being combined using svm and genetic algorithm
Method, the method increases the picture categories of road extraction, improves the accuracy of road Identification simultaneously;Both solved and used machine
The confinement problems of device learning algorithm training set, improve the speed extracting road again;Present invention incorporates image processing techniquess
And machine learning techniques, can effective district shunting road and non-rice habitats, road and barrier, improve speed and the standard of road Identification
Really rate.
Brief description
Fig. 1 is the inventive method flow chart;
Fig. 2 is svm sorting technique flow chart;
Fig. 3 is the picture example of training;
Fig. 4 file through svm classification category of roads for the fritter on Fig. 3;
Fig. 5 file through svm classification non-rice habitats classification for the fritter on Fig. 3;
Fig. 6 is Fig. 3 through the sorted design sketch of svm;
Fig. 7 is Fig. 3 through the sorted design sketch of the present invention.
Specific embodiment
With reference to the accompanying drawings and examples the present invention is described in further details.
As shown in figure 1, the present invention's is comprised the steps: based on the method for extracting roads of svm and genetic algorithm
S1: each picture is found a suitable threshold values by genetic algorithm;
S2: split road and the non-rice habitats in picture with the method for threshold segmentation;
S3: in the case of after segmentation, then represent, without barriers such as buildings, the road success that Threshold segmentation goes out,
Directly export it is therefore desirable to judge as successful frame, judge whether to identify road, if it is, execution step s6, if not,
Execution step s4;
S4: according to training classification: road and non-rice habitats, training data;
S5: obtain svm model and then split, picture is carried out svm classification, wherein, svm (support vector
Machine) it is support vector machine;
S6: export the picture handled well.
Wherein, in step s1, the image data of this example derives from the frame of video of the video that unmanned plane shoots.By unmanned
The video that machine shoots, first by video sub-frame processing, is divided into frame of video one by one, then carries out road to each frame of video and carry
Take.After the execution of step s6, also include Video Composition step b: the picture that handle well output i.e. frame of video synthetic video,
Then terminate.
In order to provide the degree of accuracy in road extraction, in step s3, if identifying road, before execution step s6,
Also include judging step a: judge whether the road identifying has barrier, if it is, according to training classification: road and barrier,
Training data, then obtain svm model and carry out further Optimized Segmentation by training barrier and road, then execution step
S6, if not, direct execution step s6.
In step s1, described genetic algorithm comprises the steps: this example
S11: random initializtion population, arranges evolutionary generation n;
S12: calculate individual fitness;
S13: select the individuality of pairing by fitness;
S14: intersect and produce new individuality, then new individuality is added to colony;
S15: setting probability and condition, if meeting condition, individuality makes a variation according to this probability, if being unsatisfactory for condition,
Then execution step s12;
S16: when colony's maximum adaptation degree continuous multi-generation is constant or has evolved n generation, terminate evolving.
Genetic algorithm is that one kind is directed to original optimization problem, finds the technology of accurate or approximate solution.One
As in the case of, genetic algorithm is didactic, be can find the overall situation solution.
The specific algorithm of this example is as follows:
In algorithm, the fitness function of object function and calculating fitness is determined by linearity cl and adjacency pr, linearly
Degree and adjacency are defined as follows:
cl=1/ [(tana)2+(tanb)2+(tanc)2], (alpha+beta r) >=t1
pr=l2/(2πr2)≥t2
Wherein, l is the length in two straight-line segment compared with short straight line section, and r is the intersegmental beeline of two straight lines, a and b is respectively
Two straightways and the angle of r, c is the angle between two straightways, a and β is constant undetermined in experiment, and t1, t2 are specific
Thresholding, determines in test.
In every generation colony, the half individuality with larger fitness is chosen to remain, two from these individualities
Two pairings carry out bright spot crossing operation, produce two new individualities, in colony, total individual number is constant.For preventing algorithm from entering local
Optimal solution, carries out mutation operator with the probability of m% to the gene position in each individuality, produces new individuality.If colony is maximum suitable
In response continuous 15 generations, are constant or evolved n generation, then terminate evolving.The individuality wherein obtaining maximum adaptation degree is globally optimal solution,
If this optimal solution is more than certain thresholding t2, the multiple highway fragments being connected with Current Highway seed can be recognized as.If obtaining
Optimum highway fragment in, consider, highway fragment farther out is from current public affairs by apart from Current Highway fragment by as far as near order
Road seed beeline is d, if there is optimum highway fragment in this distance, fragment length the longest is l therebetween, then according to following
They can be linked up by criterion:
Criterion one: d < t3
Criterion two: l/d < t4
Wherein, t3 and t4 is also one of experimentation thresholding, is determined according to the value in experiment.Accurate meeting
Then one, or in the case of being unsatisfactory for criterion two, the highway being considered fragment can be coupled together, as new highway fragment kind
Son.
In the specific implementation process of the example method, we first pass through genetic algorithm and find the suitable threshold value of in figure one,
By Threshold segmentation, rough to road and background is separated, simultaneously for some and road pixel value very close to region, that is, hinder
Hinder a little, we, are then peeled off coming in a part of sample by svm.
Further, since Threshold segmentation there is also this certain limitation, although genetic algorithm find be one overall
Excellent threshold value, but when the overall pixel value of picture very close to when, Threshold segmentation there is also limitation, therefore this pixel value phase
Near picture is also required to be trained by svm, then extracts.
In this example, svm is proposed in nineteen ninety-five by cortes and vapnik.The method can overcome lacking of multilayer neural network
Point.The main thought of svm: linear can divide situation, fairly simple;When linearly inseparable, by using non-thread
The sample of low-dimensional input space linearly inseparable is converted into high-dimensional feature space by property mapping algorithm makes its linear separability, so that
High-dimensional feature space carries out linear analysiss using linear algorithm to the nonlinear characteristic of sample and is possibly realized.
In order to obtain globally optimal solution, problem is transformed into quadratic programming (qp) by this algorithm.Sample is mapped to
Higher-dimension, then using kernel function.
Concept with core is suggested, such as linear kernel function, Polynomial kernel function, RBF (rbf) core and
Sigmoid kernel function.Because linear kernel is considered as a kind of special circumstances of rbf kernel function it may be said that there being three kinds of inhomogeneities
Other kernel, is expressed as follows respectively:
Polynomial kernel function: k (x, y)=<φ (x), φ (y)>=(1+<x, y>)d
Rbf:
Sigmoid kernel function: k (x, y)=tanh [ν (<x, y>)+c]
X and y is categorized good model vector, and kernel is counted as a kind of broad sense inner product.In rbf kernel function, radially
The corresponding supporting vector in each center of function and its weight are to be fixed by suitable algorithm.In sigmoid kernel function
Core comprises the multilayer perceptron of a hidden layer with hiding node layer to be determined by algorithm.This example uses Polynomial kernel function.
For the ease of the training of svm, for the selection of training set, first picture segmentation is become a lot of fritters, then only
The fritter having road is taken out as positive example, and the fritter on complete Bu Shi road is as negative example;Test set is exactly this pictures and is similar to
Picture.Go training pattern with various fritter pictures, finally can extract the road in video.
It should there be the model of a pre-training before starting svm.It should there be 2 groups of samples in training, one is used for
Road, another is used for non-rice habitats.As shown in figure 3, the example image of display, artificial selection goes out which is which Shi Fei road of road
Road, in this example, dotted box portion is the segmentation picture of road, and solid box part is off-highroad segmentation picture, and Fig. 4 is under preservation
The path file coming, such as 5 is the off-highroad file preserving.Then, using support vector machine pre-training model, using many
Xiang Shihe, setting degree is 1.Afterwards, predict the new input data of class, according to the model of pre-training, and the latter is arranged to 0,
So non-rice habitats region can be deleted.
As shown in Fig. 2 this example is in step s5, described svm classification specifically include following steps:
S51: start, input picture to be processed;
S52: picture segmentation is become the fritter that some pixel values are 16*16dpi;
S53: be put into a file by only having the fritter of road to branch away in fritter, off-highroad fritter branches away and is placed on
Another file;
S54: using the fritter of road as positive example, off-highroad fritter is as counter-example;
S55: the support vector machine with kernel function for the training, obtain svm model;
S56: the positive example labelling output in picture terminates.
From result, individually the result of road extraction is carried out as shown in fig. 6, most piece of region using svm classification method
It is converted into black and still some blocks and speckle noise is present in image, but some non-rice habitats regions are also taken as roadway area
Domain is extracted out.The road extraction result being combined using svm and genetic algorithm is as shown in fig. 7, road must be extracted by accurate
Out.
By contrast as can be seen that the method for the present invention improves the picture categories of road extraction, improve road simultaneously
The accuracy of identification;The present invention had both been solved simple Picture Valve Value Division and had been asked using the limitation of machine learning algorithm training set
Topic, improves the speed extracting road again;In combination with image processing techniquess and machine learning techniques, being capable of effective district shunting road
With non-rice habitats, road and barrier, improve speed and the accuracy rate of road Identification.
Additionally, present invention also offers a kind of system realizing said method, including threshold values computing module: for by each
Picture finds a suitable threshold values by genetic algorithm;Picture segmentation module: split in picture with the method for threshold segmentation
Road and non-rice habitats;First judge module: it is used for judging whether to identify road, if it is, execution output module, if
No, execute training data module;Training data module: for according to training classification: road and non-rice habitats, training data;Svm divides
Generic module: for picture is carried out svm classification;Output module: for exporting the picture handled well,.
This example, after the first judge module executes, if identifying road, before execution output module, also includes second
Judge module: whether the road for judging to identify has obstacle, if it is, according to training classification: road and barrier, train number
According to, then will identify that the picture of road carries out svm classification, then execute output module, if not, directly executing output module.
Preferably, in threshold values computing module, the frame of video of the video that described picture shoots for unmanned plane, in output for this example
After module execution, also include Video Composition module, for the picture handled well i.e. frame of video synthetic video will be exported.
The specific embodiment of the above is the better embodiment of the present invention, not limits the concrete of the present invention with this
Practical range, the scope of the present invention includes being not limited to this specific embodiment, and all equivalence changes made according to the present invention are equal
Within the scope of the present invention.
Claims (10)
1. a kind of method for extracting roads based on svm and genetic algorithm is it is characterised in that comprise the steps:
S1: each picture is found a suitable threshold values by genetic algorithm;
S2: split road and the non-rice habitats in picture with the method for threshold segmentation;
S3: judge whether to identify road, if it is, execution step s6, if not, execution step s4;
S4: according to training classification: road and non-rice habitats, training data;
S5: picture is carried out svm classification;
S6: export the picture handled well.
2. method for extracting roads according to claim 1 it is characterised in that: in step s3, if identifying road,
Before execution step s6, also include judging step a: judge whether the road identifying has barrier, if it is, according to training classification:
Road and barrier, training data, then will identify that the picture of road carries out svm classification, then execution step s6, if not,
Directly execution step s6.
3. method for extracting roads according to claim 1 it is characterised in that: in step s1, described picture be unmanned plane
The frame of video of the video shooting.
4. method for extracting roads according to claim 3 it is characterised in that: step s6 execution after, also include video close
Become step: the picture that handle well output i.e. frame of video synthetic video, terminate.
5. method for extracting roads according to claim 1 it is characterised in that: in step s1, described genetic algorithm includes
Following steps:
S11: random initializtion population, arranges evolutionary generation n;
S12: calculate individual fitness;
S13: select the individuality of pairing by fitness;
S14: intersect and produce new individuality, then new individuality is added to colony;
S15: setting probability and condition, if meeting condition, individuality, according to the variation of this probability, if being unsatisfactory for condition, is held
Row step s12;
S16: when colony's maximum adaptation degree continuous multi-generation is constant or has evolved n generation, terminate evolving.
6. method for extracting roads according to claim 5 it is characterised in that: the object function of described genetic algorithm and calculating
The fitness function of fitness is by linearity clWith adjacency prDetermined, the linearity and adjacency are defined as follows:
cl=1/ [(tana)2+(tanb)2+(tanc)2], (alpha+beta r) >=t1
pr=l2/(2πr2)≥t2
Wherein, l is the length in two straight-line segment compared with short straight line section, and r is the intersegmental beeline of two straight lines, and it is straight that a and b is respectively two
Line segment and the angle of r, c is the angle between two straightways, a and β is constant undetermined in experiment, and t1, t2 are specific thresholding,
Determine in test.
7. method for extracting roads according to claim 1 it is characterised in that: in step s5, the inclusion of described svm classification
Following steps:
S51: start, input picture to be processed;
S52: picture segmentation is become some fritters;
S53: fritter is individually placed to different files according to training classification;
S54: the training classification extracted will be needed as positive example, other training classification fritters are as counter-example;
S55: the support vector machine with kernel function for the training, obtain svm model;
S56: the positive example labelling output in picture terminates.
8. a kind of system realizing method for extracting roads described in any one of claim 1-7 is it is characterised in that include:
Threshold values computing module: for each picture is found a suitable threshold values by genetic algorithm;
Picture segmentation module: split road and the non-rice habitats in picture with the method for threshold segmentation;
First judge module: it is used for judging whether to identify road, if it is, execution output module, if not, execution training number
According to module;
Training data module: for according to training classification: road and non-rice habitats, training data;
Svm sort module: for picture is carried out svm classification;
Output module: for exporting the picture handled well.
9. system according to claim 8 it is characterised in that: first judge module execution after, if identifying road,
Before execution output module, also include the second judge module: whether the road for judging to identify has obstacle, if it is, root
According to training classification: road and barrier, training data, then will identify that the picture of road carries out svm classification, then executes output
Module, if not, directly execute output module.
10. system according to claim 8 it is characterised in that: in threshold values computing module, described picture be unmanned plane clap
The frame of video of the video taken the photograph, after output module execution, also includes Video Composition module, for the picture handling output well
It is exactly frame of video synthetic video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610834791.7A CN106372618A (en) | 2016-09-20 | 2016-09-20 | Road extraction method and system based on SVM and genetic algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610834791.7A CN106372618A (en) | 2016-09-20 | 2016-09-20 | Road extraction method and system based on SVM and genetic algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106372618A true CN106372618A (en) | 2017-02-01 |
Family
ID=57897221
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610834791.7A Pending CN106372618A (en) | 2016-09-20 | 2016-09-20 | Road extraction method and system based on SVM and genetic algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106372618A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106971170A (en) * | 2017-04-07 | 2017-07-21 | 西北工业大学 | A kind of method for carrying out target identification using one-dimensional range profile based on genetic algorithm |
CN108345875A (en) * | 2018-04-08 | 2018-07-31 | 北京初速度科技有限公司 | Wheeled region detection model training method, detection method and device |
CN108513641A (en) * | 2017-05-08 | 2018-09-07 | 深圳市大疆创新科技有限公司 | Unmanned plane filming control method, unmanned plane image pickup method, control terminal, unmanned aerial vehicle (UAV) control device and unmanned plane |
CN110809767A (en) * | 2017-07-06 | 2020-02-18 | 华为技术有限公司 | Advanced driver assistance system and method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073848A (en) * | 2010-12-31 | 2011-05-25 | 深圳市永达电子股份有限公司 | Intelligent optimization-based road recognition system and method |
CN103914698A (en) * | 2014-03-27 | 2014-07-09 | 北京科技大学 | Method for recognizing and classifying road barriers based on video |
-
2016
- 2016-09-20 CN CN201610834791.7A patent/CN106372618A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102073848A (en) * | 2010-12-31 | 2011-05-25 | 深圳市永达电子股份有限公司 | Intelligent optimization-based road recognition system and method |
CN103914698A (en) * | 2014-03-27 | 2014-07-09 | 北京科技大学 | Method for recognizing and classifying road barriers based on video |
Non-Patent Citations (7)
Title |
---|
BYOUNG-KI JEON ET AL.: "Road detection in spaceborne SAR images using a genetic algorithm", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
R. HUBER ET AL.: "Road extraction from high-resolution airborne SAR using operator fusion", 《IGARSS 2001》 * |
杜明芳 等: "自主移动机器人自适应室外道路检测", 《中国图象图形学报》 * |
汪夕明: "遥感影像道路提取方法研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
沈照庆 等: "基于支持向量机的高光谱遥感影像道路提取", 《长安大学学报》 * |
胥亚: "中高分辨率遥感影像道路提取技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
邰晓雷: "遥感图像道路信息提取研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106971170A (en) * | 2017-04-07 | 2017-07-21 | 西北工业大学 | A kind of method for carrying out target identification using one-dimensional range profile based on genetic algorithm |
CN108513641A (en) * | 2017-05-08 | 2018-09-07 | 深圳市大疆创新科技有限公司 | Unmanned plane filming control method, unmanned plane image pickup method, control terminal, unmanned aerial vehicle (UAV) control device and unmanned plane |
US11290692B2 (en) | 2017-05-08 | 2022-03-29 | SZ DJI Technology Co., Ltd. | Unmanned aerial vehicle imaging control method, unmanned aerial vehicle imaging method, control terminal, unmanned aerial vehicle control device, and unmanned aerial vehicle |
US11722647B2 (en) | 2017-05-08 | 2023-08-08 | SZ DJI Technology Co., Ltd. | Unmanned aerial vehicle imaging control method, unmanned aerial vehicle imaging method, control terminal, unmanned aerial vehicle control device, and unmanned aerial vehicle |
CN110809767A (en) * | 2017-07-06 | 2020-02-18 | 华为技术有限公司 | Advanced driver assistance system and method |
CN110809767B (en) * | 2017-07-06 | 2022-09-09 | 华为技术有限公司 | Advanced driver assistance system and method |
CN108345875A (en) * | 2018-04-08 | 2018-07-31 | 北京初速度科技有限公司 | Wheeled region detection model training method, detection method and device |
CN108345875B (en) * | 2018-04-08 | 2020-08-18 | 北京初速度科技有限公司 | Driving region detection model training method, detection method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hridayami et al. | Fish species recognition using VGG16 deep convolutional neural network | |
CN108830188B (en) | Vehicle detection method based on deep learning | |
CN109902806A (en) | Method is determined based on the noise image object boundary frame of convolutional neural networks | |
Sun et al. | Automatic multi-fault recognition in TFDS based on convolutional neural network | |
CN106096602A (en) | Chinese license plate recognition method based on convolutional neural network | |
CN112598713A (en) | Offshore submarine fish detection and tracking statistical method based on deep learning | |
CN105718889A (en) | Human face identity recognition method based on GB(2D)2PCANet depth convolution model | |
CN101893704A (en) | Rough set-based radar radiation source signal identification method | |
CN106372618A (en) | Road extraction method and system based on SVM and genetic algorithm | |
CN105046259B (en) | Coronal mass ejection detection method based on multi-feature fusion | |
CN106296709B (en) | A kind of cell image segmentation method based on population and fuzzy means clustering | |
CN107679453A (en) | Weather radar electromagnetic interference echo recognition methods based on SVMs | |
CN104156945A (en) | Method for segmenting gray scale image based on multi-objective particle swarm optimization algorithm | |
Dudi et al. | Optimized threshold-based convolutional neural network for plant leaf classification: a challenge towards untrained data | |
Liu et al. | A image segmentation algorithm based on differential evolution particle swarm optimization fuzzy c-means clustering | |
CN106023159A (en) | Disease spot image segmentation method and system for greenhouse vegetable leaf | |
CN116721414A (en) | Medical image cell segmentation and tracking method | |
Zeng et al. | Steel sheet defect detection based on deep learning method | |
CN108073940A (en) | A kind of method of 3D object instance object detections in unstructured moving grids | |
CN104573701B (en) | A kind of automatic testing method of Tassel of Corn | |
CN113537173B (en) | Face image authenticity identification method based on face patch mapping | |
CN113657414B (en) | Object identification method | |
Mouna et al. | A vehicle detection approach using deep learning network | |
CN108665479A (en) | Infrared object tracking method based on compression domain Analysis On Multi-scale Features TLD | |
Sikand et al. | Using Classifier with Gated Recurrent Unit-Sigmoid Perceptron, Order to Get the Right Bird Species Detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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: 20170201 |
|
RJ01 | Rejection of invention patent application after publication |