CN110110586A - The method and device of remote sensing airport Airplane detection based on deep learning - Google Patents
The method and device of remote sensing airport Airplane detection based on deep learning Download PDFInfo
- Publication number
- CN110110586A CN110110586A CN201910204847.4A CN201910204847A CN110110586A CN 110110586 A CN110110586 A CN 110110586A CN 201910204847 A CN201910204847 A CN 201910204847A CN 110110586 A CN110110586 A CN 110110586A
- Authority
- CN
- China
- Prior art keywords
- airport
- candidate regions
- remote sensing
- feature
- doubtful
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The method and device of the embodiment of the invention provides a kind of remote sensing airport Airplane detection based on deep learning, comprising: obtain the remote sensing Airport Images for inputting big visual field；Acquired remote sensing Airport Images are split using maximum between-cluster variance, extract airport profile, determine that airport candidate regions are sliced；It is sliced according to identified airport candidate regions, extracts the feature of airport candidate regions, extracted feature is clustered with K mean cluster algorithm, obtained the feature combination on airport, doubtful airport candidate regions are filtered out using support vector machine classifier；Aircraft Targets extraction is carried out to the doubtful airport candidate regions filtered out using light network algorithm, calculate position of the Aircraft Targets in acquired remote sensing Airport Images, the location information of Aircraft Targets is obtained, the coordinate frame of overlapping is rejected by the method that non-maximum value inhibits.Combination of embodiment of the present invention machine learning and deep learning realize and quickly detection accurate to big visual field remote sensing airport aircraft.
Description
Technical field
The invention belongs to remote sensing images detection technique fields, fly more particularly to a kind of remote sensing airport based on deep learning
The method and device of machine testing.
Background technique
With the rapid development of artificial intelligence, many technologies are innovated to every profession and trade bring and change should not be underestimated.Aircraft
As one of modern very convenient and fast vehicles, increasingly important role is all played in military domain and civil field,
No matter military or civilian, the true dynamic for grasping aircraft has and its important role.In military domain, enemy is obtained in time
The aircraft in battlefield is most important for the triumph of war；It is supervised in civil field, such as aviation, real-time statistics aircraft berthing time,
It is particularly necessary to airport implementation intelligent management that aircraft passes in and out airport situation etc..
In the prior art, the algorithm towards the detection of big visual field remote sensing images Aircraft Targets mainly has traditional algorithm and depth
Learning algorithm.In traditional algorithm, such as the raw institute of air force engineering university research, Zhu Mingming etc., propose based on Fusion Features with it is soft
The remote sensing images Airplane detection of judgement, by Fusion Features, the methods of scale scaling reduces omission factor；For another example Hohai University counts
Calculation machine and Information Institute, Li Shijin etc., propose view-based access control model word selection Airplane detection method, by combine correlation and
Redundancy Analysis improves verification and measurement ratio.In deep learning algorithm, as Chinese Academy of Sciences's Changchun optical precision optical machinery is ground with physics
Study carefully institute, wear big acute hearing etc., checks primary detection algorithm (You Only Look Once V3, abbreviation with improved need
YOLOV3) carry out Airplane detection, by improve only need to check between the intensive link block of primary detection algorithm a feature transmitting come
Reach and accurately detects；For another example Shanghai Inst. of Technical Physics, Chinese Academy of Sciences, Zhao Dan are new etc., propose residual error network
(Residual Neural Network, abbreviation ResNet), it is special by the context for extracting different layers in full convolutional network structure
Reference breath reaches accurate detection precision.
However, big view field image background is complicated in practical applications, resolution ratio is big, and pixel is more, and data transfer rate is high.According to phase
The record of document is closed, existing traditional algorithm is mostly limited due to extracting feature capabilities, and verification and measurement ratio is not high.Existing typical depth
Learning algorithm is spent, such as need to only check that YOLOv3, residual error network ResNet, the more frame detector (Single of single-point are calculated in primary detection
Shot MultiBox Detector, abbreviation SSD) etc., other than network structure is complex, the algorithm pair of current depth study
Big visual field remote sensing images Airplane detection mainly traverses full figure by sliding window and is detected, for that the place of target can not occur
There are a large amount of redundancies, and detection efficiency is low, and empty scape can also be generated by traversing full figure detection at the same time, and accuracy rate will receive shadow
It rings.
Therefore, how innovatively a technical problem that needs to be urgently solved by technical personnel in the field at present is exactly:
A kind of method for proposing effectively remote sensing airport Airplane detection based on deep learning, overcomes defect of the existing technology, with
Meet the greater demand in practical application.
Summary of the invention
In view of the above problems, it proposes the embodiment of the present invention and overcomes the above problem or at least partly in order to provide one kind
The method and apparatus of the remote sensing airport Airplane detection based on deep learning to solve the above problems, combination machine of the embodiment of the present invention
Study and deep learning improve the speed and accuracy of the test of big visual field remote sensing airport aircraft.
To solve the above-mentioned problems, the side of the invention discloses a kind of remote sensing airport Airplane detection based on deep learning
Method, comprising:
Acquired remote sensing Airport Images are split using maximum between-cluster variance, airport profile is extracted, determines airport
Candidate regions slice；
It is sliced according to identified airport candidate regions, extracts the feature of airport candidate regions, with K mean cluster algorithm to being mentioned
The feature taken is clustered, and is obtained the feature combination on airport, is filtered out doubtful airport candidate regions using support vector machine classifier；
Aircraft Targets extraction is carried out to the doubtful airport candidate regions filtered out using light network algorithm, calculates aircraft mesh
It is marked on the position in acquired remote sensing Airport Images, obtains the location information of Aircraft Targets, the side inhibited by non-maximum value
Method rejects the coordinate frame of overlapping.
Optionally, the method is after the remote sensing Airport Images for obtaining the big visual field of input further include: to acquired remote sensing
Airport Images carry out the image preprocessing of enhancing image characteristic point, and described image pretreatment includes image sharpening, image filtering, mistake
Zero point detection and/or image edge processing.
Optionally, described when being split using maximum between-cluster variance to acquired remote sensing Airport Images, using maximum
Inter-class variance carries out multi-scale division from different threshold values.
Optionally, extraction airport profile, and determine airport candidate regions slice, it specifically includes: according to extracted airport
Profile determines the central point of homogeneity local, calculates four boundary's boundary points of homogeneity local, and it is candidate to form the first airport by boundary point
Area's slice, sets the area ratio of slice, after rejecting to the slice for forming the first airport candidate regions, forms the second airport
Second airport candidate regions are determined as final airport candidate regions and are sliced by the slice of candidate regions.
Optionally, described to be sliced according to determined airport candidate regions, extract the feature of airport candidate regions, extracted feature
It is the textural characteristics and/or color characteristic of airport candidate regions.
Optionally, the feature combination for obtaining airport, it is candidate to filter out doubtful airport using support vector machine classifier
Area specifically includes:
It is supported vector machine classifier training by the feature combination to airport, obtains airport training pattern；
With airport training pattern obtained to final airport candidate regions slice classification is determined as, extracts doubtful airport and wait
Constituency.
Optionally, described that the doubtful airport candidate regions progress Aircraft Targets filtered out are mentioned using light network algorithm
It takes, calculates position of the Aircraft Targets in acquired remote sensing Airport Images, obtain the location information of Aircraft Targets, further includes:
To the doubtful airport candidate regions filtered out, sliding window inspection is carried out using the trained more frame detector models of light network algorithm single-point
It surveys, determines Aircraft Targets in the location information of candidate regions.
Optionally, the more frame detector models progress sliding window detections of the light network algorithm single-point using training are also wrapped
It includes: converting, averages and/or return using channel into the picture before the more frame detector model trainings of light network algorithm single-point
The mode of one change data is handled；
Wherein, described extract to the doubtful airport candidate regions filtered out progress Aircraft Targets mainly uses dividing method,
Selective search method and/or region merging process are realized.
The device of the invention also discloses a kind of remote sensing airport Airplane detection based on deep learning, comprising:
Image collection module, for obtaining the remote sensing Airport Images for inputting big visual field；
Image segmentation module is split acquired remote sensing Airport Images using maximum between-cluster variance, extracts airport
Profile determines that airport candidate regions are sliced；
Feature Selection module, for being sliced according to determined airport candidate regions, the feature of extraction airport candidate regions is equal with K
Value clustering algorithm clusters extracted feature, obtains the feature combination on airport, is screened using support vector machine classifier
Doubtful airport candidate regions out；
Object extraction module carries out Aircraft Targets to the doubtful airport candidate regions filtered out using light network algorithm and mentions
Take, calculate position of the Aircraft Targets in acquired remote sensing Airport Images, obtain the location information of Aircraft Targets, by it is non-most
The method that big value inhibits rejects the coordinate frame of overlapping.
The embodiment of the present invention includes following advantages:
Combination of embodiment of the present invention machine learning and deep learning use machine learning method by dividing airport candidate regions
Quickly screening airport candidate regions select the light network of the higher deep learning of timeliness by the advantage of deep learning
Mobilenet_SSD, the semantic class feature of Aircraft Targets, determines aircraft in the candidate regions of airport in the candidate regions of rapidly extracting airport
And quickly detection accurate to big visual field remote sensing airport aircraft is realized in position.
Detailed description of the invention
The step of Fig. 1 is a kind of embodiment one of the method for remote sensing airport Airplane detection based on deep learning of the invention
Flow chart；
Fig. 2 is a kind of total stream of the embodiment two of the method for remote sensing airport Airplane detection based on deep learning of the invention
Journey schematic diagram；
Fig. 3 is a kind of the novel of the embodiment two of the method for remote sensing airport Airplane detection based on deep learning of the invention
The schematic diagram of light network Mobilenet_SSD testing process；And
Fig. 4 is a kind of structural schematic diagram of the device of remote sensing airport Airplane detection based on deep learning of the invention.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
The method of the invention proposes a kind of remote sensing airport Airplane detection based on deep learning.This method is directed to depth
It practises and proposes the screening of airport candidate regions, small-sized deep learning network inspection policies, especially for airport aircraft in remote sensing images
High robust detection.
Embodiment of the method one
Referring to Fig. 1, a kind of implementation of the method for remote sensing airport Airplane detection based on deep learning of the invention is shown
The step flow chart of example one, can specifically include following steps:
Step 101, the remote sensing Airport Images for inputting big visual field are obtained；
Step 102, acquired remote sensing Airport Images are split using maximum between-cluster variance, extract airport profile,
Determine that airport candidate regions are sliced；
For inputting the remote sensing Airport Images of big visual field, in order to efficiently quickly filter out big visual field remote sensing images airport candidate
Area carries out multi-scale division by maximum between-cluster variance (abbreviation Otsu), extracts airport profile, and determines that airport candidate regions are cut
Piece.Wherein, maximum between-cluster variance algorithm algorithm is simple, can be effectively to figure when the area of target and background is not much different
As being split.In the specific implementation, due to airport scale, shape difference.In order to filter out all airports, using maximum
Inter-class variance carries out multi-scale division to acquired remote sensing Airport Images from different threshold values, is effectively reduced and loses the general of airport
Rate.
It more specifically realizes, according to extracted airport profile, determines the central point of homogeneity local, calculate homogeneity local
Four boundary's boundary points form the first airport candidate regions slice by boundary point, since airport candidate domain is larger, set the area of slice
Ratio after rejecting to the slice for forming the first airport candidate regions, forms the slice of the second airport candidate regions, by the second machine
Field candidate regions are determined as final airport candidate regions slice.Airport candidate regions slice is filtered out, and records its location information.
Step 103, it is sliced according to identified airport candidate regions, extracts the feature of airport candidate regions, calculated with K mean cluster
Method clusters extracted feature, obtains the feature combination on airport, filters out doubtful machine using support vector machine classifier
Field candidate regions；
For the airport candidate regions slice of extraction determined by step 102, doubtful airport candidate regions are further screened.It is first
First, the feature of airport candidate regions is extracted.Then, using K mean cluster algorithm (k-means clustering algorithm,
Abbreviation k-means) feature of extraction is clustered, form the feature combination with airport.Finally use support vector machines
(Support Vector Machine, abbreviation SVM) classifier filters out doubtful airport candidate regions.In the specific implementation, extractor
The feature of candidate regions, can be the structure feature for extracting airport candidate regions, the structure feature include aircraft, hashigakari and/or
Runway.The features such as texture, the color in candidate regions are extracted, cluster to form the distinctive feature combination in airport by K-means.
More specifically, vector machine classifier training is supported by the feature combination to airport, obtains airport training
Model；With airport training pattern obtained to final airport candidate regions slice classification is determined as, it is candidate to extract doubtful airport
Area.
Step 104, Aircraft Targets extraction, meter are carried out to the doubtful airport candidate regions filtered out using light network algorithm
Position of the Aircraft Targets in acquired remote sensing Airport Images is calculated, the location information of Aircraft Targets is obtained, passes through non-maximum value
The method of inhibition rejects the coordinate frame of overlapping.
For doubtful airport candidate regions obtained in step 103, this step will select light network algorithm Mobilenet_
SSD carries out the extraction of Aircraft Targets for doubtful airport candidate regions.By using to the doubtful airport candidate regions filtered out
The more frame detector models of trained light network algorithm single-point carry out sliding window detection, determine that Aircraft Targets are believed in the position of candidate regions
Breath.In realization, it includes: to entrance that the more frame detector models of the light network algorithm single-point using training, which carry out sliding window detection,
Picture before the light more frame detector model trainings of network algorithm single-point uses channel to convert, average and/or normalization data
Mode handled；Candidate regions extract and mainly use dividing method, and selective search method and/or region merging technique obtain target and exist
Location information in image.
In practical applications, the method is after step 101 further include:
Step 100, acquired remote sensing Airport Images are carried out with the image preprocessing of enhancing image characteristic point, described image
Pretreatment includes image sharpening, image filtering, zero-crossing examination and/or image edge processing.
The remote sensing Airport Images background for inputting big visual field in practical applications is complicated, in order to more rapidly accurately realize remote sensing
The Airplane detection on airport usually can also pre-process it, such as sharpening, filtering, zero-crossing examination and/or image edge processing
Deng to enhance image characteristic point.
Embodiment of the method two
Referring to fig. 2, a kind of implementation of the method for remote sensing airport Airplane detection based on deep learning of the invention is shown
The main-process stream schematic diagram of example two, concrete processing procedure are as follows:
Step 201, big in order to efficiently quickly filter out for the remote sensing images inputted, specially remote sensing Airport Images
The airport candidate regions of visual field remote sensing Airport Images carry out multi-scale division by maximum between-cluster variance (Otsu) using traditional,
Airport profile is extracted, and determines airport candidate regions slice.Detailed step is as follows:
The big visual field remote sensing images background that step 11 inputs is complicated, pre-processes to it, such as sharpens, filtering, has increased
Strong image characteristic point.
Step 12, due to airport scale, shape difference.The organic field in order to filter out, using Otsu from different
Multiple dimensioned segmentation is carried out under threshold value, reduces the probability for losing airport.
Step 13, it according to the profile of step 12 cut zone, determines the central point of homogeneity local, calculates the four of homogeneity local
Boundary's boundary point.The area ratio of slice is set since airport candidate domain is larger according to the slice that boundary point forms airport candidate regions
Example rejects very big or minimum abnormal candidate regions slice.Airport candidate regions slice is filtered out, and records its location information.
Step 202, it is sliced for the airport candidate regions that step 201 is extracted, further screens doubtful airport candidate regions.It is first
First, the feature of airport candidate regions atural object is extracted.Then, the feature of extraction is clustered using k-means, being formed has airport
The feature of feature combines.Finally doubtful airport candidate regions are filtered out with traditional SVM classifier.Detailed step is as follows:
The airport candidate regions slice that step 21 is screened based on step 13, extracts the features such as texture, the color in candidate regions, leads to
K-means is crossed to cluster to form the distinctive feature combination in airport.
Determination of the step 22 based on the doubtful airport candidate regions of SVM: distinctive to airport candidate regions based on step 21 method
Feature combination carries out SVM training, obtains the training pattern on airport.For the airport candidate regions that step 13 is screened, with trained SVM
Category of model extracts the region on doubtful airport.Svm classifier specific algorithm is as follows:
The main thought of SVM algorithm: input is clustered into sample D={ x1,x2......xnNonlinear Mapping is to a higher-dimension
Feature space establishes hyperplane ω φ (x)-ρ=0 in this high-dimensional feature space.Wherein ω is hyperplane method vector ф
It (x) is mapping point of the sample in higher dimensional space.In order to find the optimal hyperlane farthest away from origin, need to maximizeIn order to
The robustness for improving algorithm introduces slack variable ξi.The optimization problem of SVM is converted into solution quadratic programming problem at this time:
Wherein υ ∈ (0,1], n υ indicate boundary supporting vector the upper bound, the lower bound of supporting vector, with standard support vector machines
In punishment parameter it is similar.Introducing glug draws day coefficient solution problem above to obtain
Wherein αiAnd βiIt is the number greater than 0, ω, ρ and ξ is optimized respectively
It is substituted into and inner product (xi, xj), with kernel function K (xi,xj) replace
K(xi, xj)=exp (- | | xi-xj||2/σ2)
Obtain dual formula:
By arbitrarySample can obtainTerminal decision function is
Wherein xiFor supporting vector, x is sample to be tested.By solving above
Journey is related to parameter υ and σ when it is found that training2, crosscheck can be rolled over by k realize to (υ, σ2) parameter optimization.Use K-means
Cluster k obtained classifier and according to judgement formulaTo each of training set
Sample is respectively processed, to differentiate doubtful airport candidate regions.
Step 203, for the candidate regions on doubtful airport obtained in step 202, this step will select light network
Mobilenet_SSD algorithm carries out the extraction of Aircraft Targets for doubtful airport candidate regions.And Aircraft Targets are calculated in former remote sensing
Position in image obtains the location information of Aircraft Targets, and practical is candidate regions coordinate frame, is inhibited eventually by non-maximum value
Method rejects the coordinate frame of overlapping.Detailed step is as follows:
Airport Airplane detection of the step 31 based on Mobilenet_SSD: doubtful airport candidate regions are filtered out by the first step
Slice.Then, sliding window detection is carried out to the Mobilenet_SSD model of airport candidate regions slice training, determines that target is being waited
The location information in constituency.Mobilenet_SSD testing process is as follows:
Referring to Fig. 3, a kind of implementation of the method for remote sensing airport Airplane detection based on deep learning of the invention is shown
The schematic diagram of the novel and portable network Mobilenet_SSD testing process of example two, mainly has picture pretreatment, candidate region mentions
It takes, feature extraction, feature classifiers, recurrence device correction and etc..Picture pretreatment is handled before training pattern, mainly
It is converted comprising channel, averages and normalization data etc. operates；Candidate region, which is extracted, mainly uses dividing method, selective search
Method, region merging technique etc. obtain the location information of target in the picture；The traditional convolutional neural networks of feature extraction, due to tradition
The neural network number of plies it is more, the complicated network structure, convolution decomposition method very good solution in Mobilenet convolution quantity is superfluous
The problems such as remaining.Feature classifiers use the SSD framework of deep learning network, classify on the characteristic pattern of different scale, final to return
Return coordinate information and classification information.It is as follows to decompose convolution operation:
Step 32, the coordinate frame determined based on step 31 is mapped back original according to the location information that step 11 records
Figure, obtains coordinate frame of the target in big figure.Due to can generate redundancy when detecting in candidate regions screening and step 31
Coordinate frame, this step will by non-maximum value inhibit reject overlapping coordinate frame, obtain final goal in big visual field remote sensing figure
Coordinate position as in.
It should be noted that for simple description, therefore, it is stated as a series of action groups for embodiment of the method
It closes, but those skilled in the art should understand that, embodiment of that present invention are not limited by the describe sequence of actions, because according to
According to the embodiment of the present invention, some steps may be performed in other sequences or simultaneously.Secondly, those skilled in the art also should
Know, the embodiments described in the specification are all preferred embodiments, and the related movement not necessarily present invention is implemented
Necessary to example.
Installation practice
Referring to Fig. 4, a kind of structure of the device of remote sensing airport Airplane detection based on deep learning of the invention is shown
Schematic diagram, comprising:
Image collection module 401, for obtaining the remote sensing Airport Images for inputting big visual field；
Image segmentation module 402 is split acquired remote sensing Airport Images using maximum between-cluster variance, extractor
Field profile determines that airport candidate regions are sliced；
Feature Selection module 403 extracts the feature of airport candidate regions, uses K for being sliced according to determined airport candidate regions
Means clustering algorithm clusters extracted feature, obtains the feature combination on airport, is sieved using support vector machine classifier
Select doubtful airport candidate regions；
Object extraction module 404 carries out aircraft mesh to the doubtful airport candidate regions filtered out using light network algorithm
Mark extracts, and calculates position of the Aircraft Targets in acquired remote sensing Airport Images, obtains the location information of Aircraft Targets, pass through
The method that non-maximum value inhibits rejects the coordinate frame of overlapping.
For device embodiment, since it is basically similar to the method embodiment, related so being described relatively simple
Place illustrates referring to the part of embodiment of the method.
All the embodiments in this specification are described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiments, the same or similar parts between the embodiments can be referred to each other.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to of the invention its
Its embodiment.The present invention is directed to cover any variations, uses, or adaptations of the invention, these modifications, purposes or
Person's adaptive change follows general principle of the invention and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be understood that the present invention is not limited to the precise structure already described above and shown in the accompanying drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present invention is limited only by the attached claims
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Above to a kind of method and apparatus of the remote sensing airport Airplane detection based on deep learning provided by the present invention, into
It has gone and has been discussed in detail, used herein a specific example illustrates the principle and implementation of the invention, the above implementation
The explanation of example is merely used to help understand method and its core concept of the invention；Meanwhile for the general technology people of this field
Member, according to the thought of the present invention, there will be changes in the specific implementation manner and application range, in conclusion this explanation
Book content should not be construed as limiting the invention.
Claims (9)
1. a kind of method of the remote sensing airport Airplane detection based on deep learning characterized by comprising
Obtain the remote sensing Airport Images for inputting big visual field；
Acquired remote sensing Airport Images are split using maximum between-cluster variance, extract airport profile, determine airport candidate
Area's slice；
It is sliced according to identified airport candidate regions, the feature of airport candidate regions is extracted, with K mean cluster algorithm to extracted
Feature is clustered, and is obtained the feature combination on airport, is filtered out doubtful airport candidate regions using support vector machine classifier；
Aircraft Targets extraction is carried out to the doubtful airport candidate regions filtered out using light network algorithm, Aircraft Targets is calculated and exists
Position in acquired remote sensing Airport Images, obtains the location information of Aircraft Targets, is picked by the method that non-maximum value inhibits
Except the coordinate frame of overlapping.
2. the method according to claim 1, wherein the method is obtaining the remote sensing airport figure for inputting big visual field
As after further include:
Acquired remote sensing Airport Images are carried out with the image preprocessing of enhancing image characteristic point, described image pretreatment includes figure
As sharpening, image filtering, zero-crossing examination and/or image edge processing.
3. method according to claim 1 or 2, which is characterized in that it is described using maximum between-cluster variance to acquired distant
When sense Airport Images are split, multi-scale division is carried out from different threshold values using maximum between-cluster variance.
4. according to the method described in claim 3, it is characterized in that, extraction airport profile, and determining that airport candidate regions are cut
Piece specifically includes:
According to extracted airport profile, the central point of homogeneity local is determined, calculate four boundary's boundary points of homogeneity local, pass through boundary
Point forms the first airport candidate regions slice, sets the area ratio of slice, carries out to the slice for forming the first airport candidate regions
After rejecting, the slice of the second airport candidate regions is formed, the second airport candidate regions are determined as final airport candidate regions and are sliced.
5. method according to claim 1 or 2, which is characterized in that it is described to be sliced according to determined airport candidate regions, it extracts
The feature of airport candidate regions, extracted feature are the textural characteristics and/or color characteristic of airport candidate regions.
6. according to the method described in claim 4, it is characterized in that, the feature for obtaining airport combines, using supporting vector
Machine classifier filters out doubtful airport candidate regions, specifically includes:
It is supported vector machine classifier training by the feature combination to airport, obtains airport training pattern；
With airport training pattern obtained to final airport candidate regions slice classification is determined as, it is candidate to extract doubtful airport
Area.
7. method according to claim 1 or 2, which is characterized in that it is described using light network algorithm to being filtered out
Doubtful airport candidate regions carry out Aircraft Targets extraction, calculate position of the Aircraft Targets in acquired remote sensing Airport Images, obtain
To the location information of Aircraft Targets, further includes:
To the doubtful airport candidate regions filtered out, slided using the trained more frame detector models of light network algorithm single-point
Window detection, determines Aircraft Targets in the location information of candidate regions.
8. the method according to the description of claim 7 is characterized in that the more frame inspections of the light network algorithm single-point using training
It surveys device model and carries out sliding window detection further include:
To enter the more frame detector model trainings of light network algorithm single-point before picture using channel conversion, average and/or
The mode of normalization data is handled；
Wherein, described that Aircraft Targets extraction is carried out mainly using dividing method, selection to the doubtful airport candidate regions filtered out
Property search method and/or region merging process realize.
9. a kind of device of the remote sensing airport Airplane detection based on deep learning characterized by comprising
Image collection module, for obtaining the remote sensing Airport Images for inputting big visual field；
Image segmentation module is split acquired remote sensing Airport Images using maximum between-cluster variance, extracts airport profile,
Determine that airport candidate regions are sliced；
Feature Selection module is extracted the feature of airport candidate regions, is gathered with K mean value for being sliced according to determined airport candidate regions
Class algorithm clusters extracted feature, obtains the feature combination on airport, is filtered out using support vector machine classifier doubtful
Like airport candidate regions；
Object extraction module carries out Aircraft Targets extraction to the doubtful airport candidate regions filtered out using light network algorithm,
Position of the Aircraft Targets in acquired remote sensing Airport Images is calculated, the location information of Aircraft Targets is obtained, passes through non-maximum
The method that value inhibits rejects the coordinate frame of overlapping.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910204847.4A CN110110586A (en) | 2019-03-18 | 2019-03-18 | The method and device of remote sensing airport Airplane detection based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910204847.4A CN110110586A (en) | 2019-03-18 | 2019-03-18 | The method and device of remote sensing airport Airplane detection based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110110586A true CN110110586A (en) | 2019-08-09 |
Family
ID=67484401
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910204847.4A Pending CN110110586A (en) | 2019-03-18 | 2019-03-18 | The method and device of remote sensing airport Airplane detection based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110110586A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340815A (en) * | 2020-03-09 | 2020-06-26 | 电子科技大学 | Adaptive image segmentation method based on Otsu method and K mean value method |
| CN111862006A (en) * | 2020-06-30 | 2020-10-30 | 北京北方智图信息技术有限公司 | Detection method and device for small aircraft |
| CN111881831A (en) * | 2020-07-28 | 2020-11-03 | 南京拟态智能技术研究院有限公司 | Multi-scale feature fusion target detection system based on deep learning |
| CN112213264A (en) * | 2020-09-22 | 2021-01-12 | 武汉工程大学 | Airport reference map preparation method for scene matching guidance |
| CN113688777A (en) * | 2021-09-07 | 2021-11-23 | 西北工业大学 | Airport pavement airplane real-time detection method based on embedded CPU |
| CN114638474A (en) * | 2022-02-18 | 2022-06-17 | 南京航空航天大学 | Airport group homogeneity measurement method based on multi-dimensional indexes |
| CN115034129A (en) * | 2022-05-17 | 2022-09-09 | 齐鲁工业大学 | NOx emission concentration soft measurement method for thermal power plant denitration device |
| CN118052997A (en) * | 2024-04-16 | 2024-05-17 | 北京航空航天大学 | Target confirmation method embedded with physical characteristics and common sense |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069460A (en) * | 2015-08-21 | 2015-11-18 | 航天长征火箭技术有限公司 | ISAR image ship target feature extraction method |
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
| CN108564602A (en) * | 2018-04-16 | 2018-09-21 | 北方工业大学 | Airplane detection method based on airport remote sensing image |
| CN108664939A (en) * | 2018-05-16 | 2018-10-16 | 东南大学 | A kind of remote sensing images aircraft recognition method based on HOG features and deep learning |
| CN108804988A (en) * | 2017-05-04 | 2018-11-13 | 上海荆虹电子科技有限公司 | A kind of remote sensing image scene classification method and device |
-
2019
- 2019-03-18 CN CN201910204847.4A patent/CN110110586A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069460A (en) * | 2015-08-21 | 2015-11-18 | 航天长征火箭技术有限公司 | ISAR image ship target feature extraction method |
| CN108804988A (en) * | 2017-05-04 | 2018-11-13 | 上海荆虹电子科技有限公司 | A kind of remote sensing image scene classification method and device |
| CN107423760A (en) * | 2017-07-21 | 2017-12-01 | 西安电子科技大学 | Based on pre-segmentation and the deep learning object detection method returned |
| CN108564602A (en) * | 2018-04-16 | 2018-09-21 | 北方工业大学 | Airplane detection method based on airport remote sensing image |
| CN108664939A (en) * | 2018-05-16 | 2018-10-16 | 东南大学 | A kind of remote sensing images aircraft recognition method based on HOG features and deep learning |
Non-Patent Citations (3)
| Title |
|---|
| SHUO ZHUANG ET AL.: "A Single Shot Framework with Multi-Scale Feature Fusion for Geospatial Object Detection", 《REMOTE SENSING》 * |
| ZHONG CHEN ET AL.: "End-to-End Airplane Detection Using Transfer Learning in Remote Sensing Images", 《REMOTE SENSING》 * |
| 余良凯: "基于深度学习的机场场面目标检测", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340815A (en) * | 2020-03-09 | 2020-06-26 | 电子科技大学 | Adaptive image segmentation method based on Otsu method and K mean value method |
| CN111862006A (en) * | 2020-06-30 | 2020-10-30 | 北京北方智图信息技术有限公司 | Detection method and device for small aircraft |
| CN111881831A (en) * | 2020-07-28 | 2020-11-03 | 南京拟态智能技术研究院有限公司 | Multi-scale feature fusion target detection system based on deep learning |
| CN112213264A (en) * | 2020-09-22 | 2021-01-12 | 武汉工程大学 | Airport reference map preparation method for scene matching guidance |
| CN112213264B (en) * | 2020-09-22 | 2024-04-05 | 武汉工程大学 | Airport reference map preparation method for scene matching guidance |
| CN113688777A (en) * | 2021-09-07 | 2021-11-23 | 西北工业大学 | Airport pavement airplane real-time detection method based on embedded CPU |
| CN113688777B (en) * | 2021-09-07 | 2024-09-06 | 西北工业大学 | Airport pavement plane real-time detection method based on embedded CPU |
| CN114638474A (en) * | 2022-02-18 | 2022-06-17 | 南京航空航天大学 | Airport group homogeneity measurement method based on multi-dimensional indexes |
| CN115034129A (en) * | 2022-05-17 | 2022-09-09 | 齐鲁工业大学 | NOx emission concentration soft measurement method for thermal power plant denitration device |
| CN115034129B (en) * | 2022-05-17 | 2024-08-20 | 齐鲁工业大学 | NOx emission concentration soft measurement method for thermal power plant denitration device |
| CN118052997A (en) * | 2024-04-16 | 2024-05-17 | 北京航空航天大学 | Target confirmation method embedded with physical characteristics and common sense |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110586A (en) | The method and device of remote sensing airport Airplane detection based on deep learning | |
| CN101403676B (en) | Insulator hydrophobicity rank amalgamation judging method based on D-S evidence theory | |
| CN104834942B (en) | Remote sensing image variation detection method and system based on mask classification | |
| CN112418117A (en) | Small target detection method based on unmanned aerial vehicle image | |
| CN109657610A (en) | A kind of land use change survey detection method of high-resolution multi-source Remote Sensing Images | |
| Zhang et al. | Unsupervised difference representation learning for detecting multiple types of changes in multitemporal remote sensing images | |
| CN109117836A (en) | Text detection localization method and device under a kind of natural scene based on focal loss function | |
| CN108154159B (en) | A kind of method for tracking target with automatic recovery ability based on Multistage Detector | |
| CN106355188A (en) | Image detection method and device | |
| CN109117883A (en) | SAR image sea ice classification method and system based on long memory network in short-term | |
| CN104866865B (en) | A kind of contact net balanced line fault detection method based on DHOG and discrete cosine transform | |
| CN108764018A (en) | A kind of multitask vehicle based on convolutional neural networks recognition methods and device again | |
| CN110298410A (en) | Weak target detection method and device in soft image based on deep learning | |
| CN102945374A (en) | Method for automatically detecting civil aircraft in high-resolution remote sensing image | |
| CN109993061A (en) | A kind of human face detection and tracing method, system and terminal device | |
| CN112308156A (en) | Two-stage image change detection method based on counterstudy | |
| CN115422962A (en) | Gesture and gesture recognition method and device based on millimeter wave radar and deep learning algorithm | |
| CN108154089A (en) | A kind of people counting method of head detection and density map based on dimension self-adaption | |
| CN114170487A (en) | Vision-based water surface oil stain detection method | |
| CN114663795A (en) | Target detection method for obtaining rear image of glass curtain wall by range gating imaging equipment | |
| CN106951924B (en) | Seismic coherence body image fault automatic identification method and system based on AdaBoost algorithm | |
| Maheen et al. | Machine learning algorithm for fire detection using color correlogram | |
| Shafay et al. | Programmable broad learning system to detect concealed and imbalanced baggage threats | |
| CN107403193A (en) | A kind of transponder image detecting method and device | |
| Valldor et al. | Firearm detection in social media images |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190809 |
|
| WD01 | Invention patent application deemed withdrawn after publication |