CN106557740A - The recognition methods of oil depot target in a kind of remote sensing images - Google Patents

The recognition methods of oil depot target in a kind of remote sensing images Download PDF

Info

Publication number
CN106557740A
CN106557740A CN201610910833.0A CN201610910833A CN106557740A CN 106557740 A CN106557740 A CN 106557740A CN 201610910833 A CN201610910833 A CN 201610910833A CN 106557740 A CN106557740 A CN 106557740A
Authority
CN
China
Prior art keywords
area
interest
target
oil depot
feature
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
CN201610910833.0A
Other languages
Chinese (zh)
Other versions
CN106557740B (en
Inventor
孙向东
朱军
杨卫东
赵革
邹腊梅
翟展
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.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
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 Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN201610910833.0A priority Critical patent/CN106557740B/en
Publication of CN106557740A publication Critical patent/CN106557740A/en
Application granted granted Critical
Publication of CN106557740B publication Critical patent/CN106557740B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00624Recognising scenes, i.e. recognition of a whole field of perception; recognising scene-specific objects
    • G06K9/0063Recognising patterns in remote scenes, e.g. aerial images, vegetation versus urban areas
    • G06K9/00637Recognising patterns in remote scenes, e.g. aerial images, vegetation versus urban areas of urban or other man made structures

Abstract

The invention discloses in a kind of remote sensing images oil depot target recognition methods, calculate the phase spectrum conspicuousness of whole scene first, according to phase spectrum conspicuousness extract scene in the be possible to area-of-interest comprising target;In feature extraction, using the partial structurtes feature of the calculating area-of-interest of local regression nuclear model pointwise, and the Feature Descriptor that can describe object construction is generated;In the target detection stage, similarity measurement is made with cosine similarity, calculate the similarity of area-of-interest and oil depot sample image, and the characteristic in the positive and negative sample separating capacity using Feature Descriptor and similitude face builds the decision networks with adaptive ability, the PRELIMINARY RESULTS of target detection is obtained by the decision networks, unnecessary PRELIMINARY RESULTS is removed by non-maxima suppression algorithm, final target detection result is obtained;In this general remote sensing images proposed by the present invention, oil depot mesh object detection method is good for the target identification effect of multiple dimensioned, various visual angles.

Description

The recognition methods of oil depot target in a kind of remote sensing images
Technical field
The invention belongs to Remote Sensing Target technology of identification field, more particularly, to oil depot mesh in a kind of remote sensing images Target recognition methods.
Background technology
High-resolution remote sensing image provides abundant detailed information so that the identification of all kinds of specific targets becomes can Energy;But noise jamming, season weather, shade, intensity of illumination, the factor such as block and can cause the structure of target internal details, line Reason information produces fluctuation, brings difficulty to the identification of high-definition picture.
The method of the remote sensing images oil depot target detection of prior art includes:Based on the object detection method of deep learning, Based on the target identification detection method of priori, the object detection method based on model;Target detection based on deep learning Method is rich for sample information and sample quantity has at a relatively high dependence, and in most cases for remote sensing The identification of image oil depot target can only provide simple data image source;It is to utilize based on the target identification method of priori The average of the priori of target such as aircraft, variance, the priori features such as bending moment are not used as according to the position for judging target, needs Clarification of objective is accurately expressed, and is needed the decision-making technique with adaptive ability, in priori knowledge representation not In the case that enough accurate or decision-making techniques are not enough improved, the target detection degree of accuracy is relatively low;It is by big based on the method for model Target signature is extracted in the experiment of amount, is marked the model parameter of target to generate and is assumed and target property is predicted, in actual fortune The model that background or target are measured with is matched with the characteristic of prediction again, is reached certain similarity, that is, is considered mesh Mark;This object detection method based on model has very high requirement for the accuracy and serious forgiveness of modeling, and this is for distant The identification of the oil depot target under sense image complex scene is with very big difficulty.
The content of the invention
For the disadvantages described above or Improvement requirement of prior art, the invention provides oil depot target in a kind of remote sensing images Recognition methods, its object is to overcome the technology of identification under existing remote sensing images complex scene to sample size, priori features, builds The mould degree of accuracy and the dependence of serious forgiveness, there is provided the oil depot target identification method being suitable under remote sensing images complex scene.
For achieving the above object, according to one aspect of the present invention, there is provided the knowledge of oil depot target in a kind of remote sensing images Other method, comprises the steps:
(1) by remote sensing images are carried out with phase spectrum significant characteristics calculating, significant characteristics region segmentation extracting sense Interest region;
(2) multiple dimensioned region of interest area image is obtained by sampling is carried out to above-mentioned area-of-interest;And to oil depot target Template and above-mentioned multiple dimensioned region of interest area image carry out feature extraction and obtain multiple dimensioned Feature Descriptor;
(3) target search is carried out to multiple dimensioned Feature Descriptor and obtains multiple dimensioned similar face, and to multiple dimensioned similar face Carry out fusion and generate area-of-interest similar face;
Estimated according to positive and negative samples of the area-of-interest similar face with Feature Descriptor, built with adaptive ability Decision networks;
(4) Preliminary detection is carried out to the oil depot target in remote sensing images according to above-mentioned decision networks;And adopt non-maximum Restrainable algorithms identify oil depot target from the result of Preliminary detection;
Wherein, non-maxima suppression algorithm refers to, the similitude of the target area to detecting from area-of-interest Value carries out sequence from big to small, chooses maximum target region, and exclusion overlaps region of the area more than threshold value, and by repeatedly In generation, is both less than the algorithm of threshold value until all of target area overlapping area.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (1) is including following sub-step:
(1.1) for given remote sensing images carry out dimension normalization, normalized image is turned by Fourier's change Frequency domain is changed to, the phase place spectrum signature of frequency domain figure picture is obtained;And the phase place spectrum signature is carried out inversefouriertransform spy shown Work property characteristic pattern;
(1.2) the significant characteristics figure is split using maximum stability region feature extracting method, obtains many Individual area-of-interest.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (2) is including following sub-step:
(2.1) each area-of-interest is carried out upwards, down sample, obtain multiple dimensioned region of interest area image;
(2.2) template to oil depot target, multiple dimensioned region of interest area image carry out feature extraction acquisition template characteristic and retouch State the Feature Descriptor of sub and multiple dimensioned region of interest area image.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (2.2) is including following sub-step:
(2.2.1) local regression core feature W of the region of interest area image under each yardstick of calculating of individual elementj
Wherein:
K () is Gaussian function, and x refers to the center pixel coordinate of region of interest area image,Refer to interested Two-dimensional space coordinate in area image;MatrixH represents smoothing factor, Matrix ClIt is in the integrated of coordinate p × p The covariance matrix of the matrix of each pixel gradient composition in window;
N refers to piece unit number, and correspondence area-of-interest is divided into n p × p pieces unit, and j refers to the numbering of piece unit, and p is referred to The size of integrated window;
(2.2.2) the local regression core feature of each pixel is connected, obtains the initial configuration feature of region of interest area image DescriptionWherein, I refers to unit matrix, and R refers to real number;
(2.2.3) template and multiple dimensioned region of interest area image to oil depot target carries out feature extraction, obtains template special Levy the sub- W of descriptionQWith area-of-interest image feature descriptor WT
Wherein,Wherein, nTRefer to target image Piece unit number;
(2.2.4) sub- W is described to above-mentioned template characteristicQDimension-reduction treatment is carried out, the oil depot formwork structure after dimensionality reduction is obtained Feature
Wherein, PCA dimensionality reductions matrixWherein, d refers to the intrinsic dimensionality after dimensionality reduction;
(2.2.5) by PCA dimensionality reduction matrix AsQTo the area-of-interest image feature descriptor W under each yardstickTCarry out it is main into Divide and extract,
Obtain the Feature Descriptor of each yardstick region of interest area image
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (3) is including following sub-step:
(3.1) target is searched for oil depot To Template size and fixed step size on multiple dimensioned feature, it is right to obtain The multiple dimensioned similar face answered;
The similar face of each yardstick is mapped under archeus the similar face for obtaining archeus, and the similar face to archeus enters Row fusion obtains area-of-interest similar face;
(3.2) several positive and negative samples are intercepted in given remote sensing images, and obtains the feature of these positive and negative samples Description;
(3.3) characteristic face by the Feature Descriptor of above-mentioned positive and negative samples with oil depot template carries out similarity measure calculating, The positive and negative samples for obtaining Feature Descriptor are estimated;The similarity Nogata with area-of-interest similar face is estimated with reference to positive and negative samples Figure statistical property, builds the decision networks with adaptive ability.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (3.3) is including following sub-step:
(3.3.1) the local regression core feature of each positive and negative samples of remote sensing images is calculated, is entered with the characteristic face of oil depot template Row correlation calculations, obtain the frequency normalized curve of positive and negative samples coefficient correlation;Will be the frequency normalization of positive sample bent The abscissa of the intersection point of the frequency normalized curve of line and negative sample is used as the first decision-making value τ0
(3.3.2) for the area-of-interest similar face that fusion is obtained, with phase coefficient correlation K big in the similar face Relation number as threshold tau to be selected, by τ1=max (τ, τ0) as the second decision-making value;
(3.3.3) by the first decision-making value τ0With the second decision-making value τ1Constitute the decision networks.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (4) is including following sub-step:
(4.1) similar face screening is carried out by the decision networks, when there is coefficient correlation in similar face more than described the One threshold tau0Point, then there is target in judging the similar face;When the coefficient correlation of area-of-interest is more than second threshold Value τ1, then judge the area-of-interest as target area;
(4.2) oil depot target is identified from the target area using non-maxima suppression algorithm.
Preferably, in above-mentioned remote sensing images oil depot target recognition methods, its step (4.2) is including following sub-step:
(4.2.1) all of target area is sorted from high to low by similarity, determines the maximum target of similarity Region;
(4.2.2) obtain the overlapping area of the maximum target area of the similarity and all target areas;
(4.2.3) remove target area of the overlapping area more than area threshold;
(4.2.4) repeat step (4.2.1)~(4.2.3), until the overlapping area of all target areas is both less than default Area threshold;The common factor in two overlay target regions is caused to be less than 0.3 with the ratio of union by preset area threshold value.
In general, by the contemplated above technical scheme of the present invention compared with prior art, can obtain down and show Beneficial effect:
(1) present invention provide remote sensing images in oil depot target recognition methods, in the processing procedure of step (3) consider The architectural feature of oil depot target, by the use of Partial controll core as the feature interpretation of image, due to the local controlled of the target of oil depot Core feature processed has the separating capacity of positive negative sample well, therefore this feature can be good at distinguishing target and background Come;
(2) present invention provide remote sensing images in oil depot target recognition methods, by extract area-of-interest to remote sensing Image is pre-processed, and reduces the hunting zone of oil depot target, reduces complex background for the interference of information, is greatly dropped Low amount of calculation, improves process real-time;
(3) present invention provide remote sensing images in oil depot target recognition methods, as area-of-interest can extract standard Structured features description true and that there is fine sample separating capacity, therefore during solving oil depot target detection, scene In complexity, multiple views, multiple dimensioned, same scene, many of the targeted species technology that caused Detection accuracy is low, false alarm rate is high is asked Topic, through checking, in remote sensing images proposed by the invention, the recognition methods of oil depot target is under complex scene, more for recognizing Yardstick, the target of various visual angles have good recognition effect.
Description of the drawings
Fig. 1 is the schematic flow sheet of the recognition methods of oil depot target in remote sensing images provided in an embodiment of the present invention;
Fig. 2 is the Comparative result schematic diagram of similar face computational methods in base station provided in an embodiment of the present invention;Wherein, (a) it is Eigenmatrix f is obtained by similarity ρQAnd fTCorrelation surface;(b) be in embodiment area-of-interest fusion obtain it is similar Face;
Result schematic diagrams of the Fig. 3 for the remote sensing images oil depot target identification of the embodiment of the present invention.
Specific embodiment
In order that the objects, technical solutions and advantages of the present invention become more apparent, it is below in conjunction with drawings and Examples, right The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, and It is not used in the restriction present invention.As long as additionally, technical characteristic involved in invention described below each embodiment Do not constitute conflict each other can just be mutually combined.
The recognition methods of oil depot target in remote sensing images provided by the present invention, extracts interested first in remote sensing images Region, calculates phase place spectrum signature, and according to phase place spectrum signature, carries out salient region extraction, obtains the seat of area-of-interest The information such as mark, size;Then according to area-of-interest has been obtained, feature is carried out based on local regression core to area-of-interest and is retouched State;Finally carry out feature extraction to oil depot sample image and target region of interest respectively, calculate all target areas and oil depot The similarity of sample image, and the decision networks with adaptive ability is built, non-pole is carried out to the result of decision networks finally Big value suppresses, and obtains object detection results;
Its idiographic flow as schematically shown in Figure 1, comprises the steps:
(1) remote sensing images are carried out with the laggard line phase spectrum significant characteristics of dimension normalization to calculate and row salient region Extract;
And significant characteristics region is split using maximum stability region feature extracting method, obtain several little Region of interest area image;
(2) above-mentioned region of interest area image is carried out upwards, down sample, obtain the area-of-interest under multiple yardsticks Image;
(3) local regression core (the Local Steering of the region of interest area image under each yardstick of calculating pixel-by-pixel Kernel, LSK) feature, characteristic face of the region of interest area image under each yardstick is obtained, a series of multiple dimensioned features are obtained Face;
(4) target is carried out with the step-length of fixed target frame size, fixation on a series of above-mentioned multiple dimensioned characteristic faces Search, obtains corresponding multiple dimensioned similar face;
The similar face of each yardstick is mapped under archeus, and area-of-interest similar face is obtained by fusion;
(5) several positive and negative samples are intercepted in remote sensing images, calculates the Feature Descriptor LSKs of these positive and negative samples; Wherein, size is in the same size with oil depot To Template;
And the object module by positive and negative samples with oil depot carries out similarity measure calculating, obtains the positive and negative of Feature Descriptor Sample is estimated;The decision-making mode with adaptive ability is built with reference to the similarity histogram statistical property of area-of-interest similar face Network;
(6) preliminary detection is carried out according to decision networks to the oil depot target in scene, obtains preliminary target detection knot Really, oil depot target, how many oil depot target included whether;
(7) oil depot target is identified according to the Preliminary detection result using non-maxima suppression algorithm.
In the remote sensing images provided by embodiment, the recognition methods of oil depot target, specific as follows:
(1) Image semantic classification, extracts area-of-interest;Including following sub-step:
(1.1) for given remote sensing images, dimension normalization is carried out first, the significant characteristics of phase spectrum are then carried out Calculate, obtain significant characteristics figure;
(1.2) on the basis of remote sensing images significant characteristics figure, will using maximum stability region feature extracting method Significant characteristics figure is divided into little area-of-interest;
Maximum stability region feature extracting method refers to, carries out binary conversion treatment to gray level image using multiple threshold values; In all bianry images for obtaining, some of image connected region is varied less, and even without change, then the region is The maximum stable extremal region for getting, as area-of-interest;Wherein, threshold value according to image intensity value from 0 to 255 according to It is secondary to be incremented by;
(2) LSK Multi resolution feature extractions;
In this step, area-of-interest is carried out change of scale to meet the multiple dimensioned adaptability of oil depot target detection; Carry out in range scale first, by former region of interest area image I0Upwards, sampled downwards, obtained Ik=λ I0, λ is yardstick The factor of change;
Then template and multiple dimensioned region of interest area image to oil depot target carries out feature extraction and obtains Feature Descriptor; Specifically include following sub-step:
(2.1) architectural feature description is calculated:
(2.1.1) LSK features W per a piece of unit are calculated one by onej,
Wherein:
Wherein:K () is Gaussian function, and x represents center pixel coordinate,Two-dimensional space in expression image Coordinate;MatrixH represents smoothing factor, Matrix ClIt is each pixel gradient group in the integrated window of coordinate p × p Into matrix covariance matrix;
(2.1.2) the LSK features series connection of each piece unit will be obtained, will obtain the initial configuration Feature Descriptor of the width image LSKs:
(2.1.3) describing son according to architectural feature carries out spy to the template and multiple dimensioned region of interest area image of oil depot target Extraction is levied, preliminary template characteristic is obtained and is described sub- WQWith preliminary area-of-interest image feature descriptor WT;The two with row to Amount is expressed as follows:
(2.2) architectural feature principal component analysis;
(2.2.1) sub- W is described to template characteristicQDimension-reduction treatment is carried out, the characteristic vector of d dimensions, obtains PCA matrixes before retainingAnd the architectural feature of the oil depot template after dimensionality reduction
With above-mentioned PCA dimensionality reductions matrix AQTo the W under various yardsticksTCarry out Principle component extraction,
Obtain the Feature Descriptor of each scalogram picture
In this step, more notable, essence feature is retained using the method for above-mentioned PCA dimensionality reductions, and reduces answering for calculating Miscellaneous degree.
(3) oil depot target detection;Specifically include following sub-step:
(3.1) similar face is calculated;
By formulaSimilarity measure is carried out, wherein, ρ is cosine similarity, for the structure of higher-dimension The computing formula of feature cosine similarity matrix is as follows:
Wherein
All units in image calculateObtain:
WhereinWithFor l-th characteristic vectorJ-th element;As shown in Fig. 2 (a), it is by phase Like the eigenmatrix f that property value ρ is obtainedQAnd fTCorrelation surface.
(3.2) multiscale target search, including following sub-step:
(3.2.1) with architectural feature F of oil depot templateQIn the Feature Descriptor F of the image of each yardstickTUpper search oil depot mesh Mark, the similar face RM being calculated under each yardstickk
(3.2.2) each similar face is mapped under archeus, obtains the similar face of archeus, the similar face to archeus Carry out merging and obtain final similar face RM0
As shown in Fig. 2 (b), being area-of-interest fusion is obtained in embodiment similar face, as shown in the drawing, wherein existing Possible three oil depot targets;
(3.3) build adaptive decision networks;Build adaptive decision-making network it is critical only that determination first threshold τ0 With Second Threshold τ1;First threshold τ0For judging to whether there is target in scene:It is more than when there is coefficient correlation in similar face τ0, then it is assumed that there is target;Second Threshold τ1For judging to there are how many targets in scene:When the phase relation of area-of-interest Number is more than τ1, then it is assumed that the area-of-interest is oil depot target area;
The step includes following sub-step:
(3.3.1) LSKs of each positive and negative samples is calculated respectively, carries out correlation point with the characteristic face of oil depot template Analysis, and the frequency normalized curve for obtaining positive and negative samples coefficient correlation is counted, to select positive sample normalized curve and negative sample The abscissa of normalized curve intersection point is used as the first decision-making value τ0
(3.3.2) the similar face RM after statistics fusion0In similarity, similar face is made coefficient correlation histogram system Meter, takes coefficient correlation K big coefficient correlation in similar face as threshold tau to be selected, with τ1=max (τ, τ0) as the second decision-making Threshold value;In embodiment, COEFFICIENT K is 98%;
(3.3.3) by first threshold τ0With Second Threshold τ1Similar face screening is carried out, when there is coefficient correlation in similar face More than τ0Point, then there is target in judging the similar face;When the coefficient correlation of the point (suspected target) in area-of-interest it is big In first threshold τ1, then judge that the similar face is oil depot target area, wherein there is target;
(3.4) non-maxima suppression is processed:
The similar face of the presence target to screening through above-mentioned decision networks carries out non-maximum restraining process, obtains oil Storehouse object detection results;Specifically include following sub-step:
(3.4.1) all of target area is sorted from high to low by its similarity, determines similarity maximum Target area;
(3.4.2) calculate the overlapping area of the maximum target area of similarity and all target areas;
(3.4.3) remove target area of the overlapping area more than area threshold;In embodiment, area threshold takes 0.4;
(3.4.4) repeat the above steps (3.4.1)~(3.4.3), until the overlapping area of all target areas is respectively less than Area threshold;
In embodiment, the preliminary aim testing result obtained by decision networks is carried out by non-maxima suppression algorithm Screening, obtains a series of oil depot object detection results of scenes as shown in Figure 3, and oil depot target is marked with white box;From the figure In as can be seen that oil depot object detection results Detection accuracy rate is high, false alarm rate is almost nil.
As it will be easily appreciated by one skilled in the art that the foregoing is only presently preferred embodiments of the present invention, not to The present invention, all any modification, equivalent and improvement made within the spirit and principles in the present invention etc. are limited, all should be included Within protection scope of the present invention.

Claims (8)

1. in a kind of remote sensing images oil depot target recognition methods, it is characterised in that comprise the steps:
(1) it is interested to extract by remote sensing images are carried out with phase spectrum significant characteristics calculating, significant characteristics region segmentation Region;
(2) multiple dimensioned region of interest area image is obtained by sampling is carried out to the area-of-interest;And the mould to oil depot target Plate and the multiple dimensioned region of interest area image carry out feature extraction and obtain multiple dimensioned Feature Descriptor;
(3) target search) is carried out to the Feature Descriptor and obtains multiple dimensioned similar face, and multiple dimensioned similar face is merged Generate area-of-interest similar face;
Estimated according to positive and negative samples of the area-of-interest similar face with the Feature Descriptor, build with it is adaptive should be able to The decision networks of power;
(4) Preliminary detection is carried out to the oil depot target in remote sensing images according to the decision networks;Calculated using non-maxima suppression Method identifies oil depot target from the result of Preliminary detection.
2. recognition methods as claimed in claim 1, it is characterised in that the step (1) is including following sub-step:
(1.1) for given remote sensing images carry out dimension normalization, normalized image is transformed into by Fourier's change Frequency domain, obtains the phase place spectrum signature of frequency domain figure picture;And the special acquisition conspicuousness of inversefouriertransform is carried out to the phase place spectrum signature Characteristic pattern;
(1.2) the significant characteristics figure is split using maximum stability region feature extracting method, obtains multiple senses Interest region.
3. recognition methods as claimed in claim 1 or 2, it is characterised in that the step (2) is including following sub-step:
(2.1) each area-of-interest is carried out upwards, down sample, obtain multiple dimensioned region of interest area image;
(2.2) template to oil depot target, the multiple dimensioned region of interest area image carry out feature extraction acquisition template characteristic and retouch State the Feature Descriptor of sub and multiple dimensioned region of interest area image.
4. recognition methods as claimed in claim 3, it is characterised in that the step (2.2) is including following sub-step:
(2.2.1) local regression core feature W of the region of interest area image under each yardstick of calculating of individual elementj
W j ( x l - x ) = K j ( x l - x ; H l ) Σ l = 1 p 2 K j ( x l - x ; H l ) , j = 1 , ... , n l = 1 , ... , p 2 ;
Wherein:
K () is Gaussian function, and x refers to the center pixel coordinate of region of interest area image,Refer to area-of-interest Two-dimensional space coordinate in image;MatrixH represents smoothing factor;Matrix ClIt is every in the integrated window of p × p The covariance matrix of the matrix of individual pixel gradient composition;N refers to piece unit number, and j refers to the numbering of piece unit;
(2.2.2) the local regression core feature of each pixel is connected, obtains the initial configuration feature interpretation of region of interest area image SonWherein, I refers to unit matrix, and R refers to real number;
(2.2.3) template and multiple dimensioned region of interest area image to oil depot target carries out feature extraction, obtains template characteristic and retouches State sub- WQWith area-of-interest image feature descriptor WT
Wherein,Wherein, nTRefer to the piece of target image First number;
(2.2.4) sub- W is described to the template characteristicQDimension-reduction treatment is carried out, the oil depot formwork structure feature after dimensionality reduction is obtained
Wherein, PCA dimensionality reductions matrixD is the intrinsic dimensionality after dimensionality reduction;
(2.2.5) by the PCA dimensionality reductions matrix AQTo the area-of-interest image feature descriptor W under each yardstickTCarry out it is main into Divide and extract,
Obtain the Feature Descriptor of each yardstick region of interest area image
5. recognition methods as claimed in claim 1, it is characterised in that the step (3) is including following sub-step:
(3.1) target is searched for oil depot To Template size and fixed step size on multiple dimensioned feature, is obtained corresponding Multiple dimensioned similar face;The similar face of each yardstick is mapped under archeus the similar face for obtaining archeus, and the phase to archeus Fusion is carried out like face and obtains area-of-interest similar face;
(3.2) several positive and negative samples are intercepted in given remote sensing images and obtains the feature interpretation of these positive and negative samples Son;
(3.3) characteristic face by the Feature Descriptor of the positive and negative samples with oil depot template carries out similarity measure calculating, obtains The positive and negative samples of Feature Descriptor are estimated;The similarity Nogata with area-of-interest similar face is estimated with reference to the positive and negative samples Figure statistical property, builds the decision networks with adaptive ability.
6. recognition methods as claimed in claim 5, it is characterised in that the step (3.3) is including following sub-step:
(3.3.1) the local regression core feature of each positive and negative samples of remote sensing images is calculated, and phase is carried out with the characteristic face of oil depot template Closing property is calculated, and obtains the frequency normalized curve of positive and negative samples coefficient correlation;By the frequency normalized curve of positive sample with The abscissa of the intersection point of the frequency normalized curve of negative sample is used as the first decision-making value τ0
(3.3.2) for the area-of-interest similar face that fusion is obtained, with phase relation coefficient correlation K big in the similar face Count as threshold tau to be selected, by τ1=max (τ, τ0) as the second decision-making value;
(3.3.3) by the first decision-making value τ0With the second decision-making value τ1Constitute the decision networks.
7. recognition methods as claimed in claim 6, it is characterised in that the step (4) is including following sub-step:
(4.1) similar face screening being carried out by the decision networks, is more than first threshold when there is coefficient correlation in similar face Value τ0Point, then there is target in judging the similar face;When the coefficient correlation of area-of-interest is more than the Second Threshold τ1, Then judge the area-of-interest as target area;
(4.2) oil depot target is identified from the target area using non-maxima suppression algorithm.
8. recognition methods as claimed in claim 7, it is characterised in that the step (4.2) is including following sub-step:
(4.2.1) all of target area is sorted from high to low by similarity, determines the maximum target area of similarity;
(4.2.2) obtain the overlapping area of the maximum target area of the similarity and all target areas;
(4.2.3) remove target area of the overlapping area more than area threshold;
(4.2.4) repeat step (4.2.1)~(4.2.3), until the overlapping area of all target areas is both less than area threshold.
CN201610910833.0A 2016-10-19 2016-10-19 The recognition methods of oil depot target in a kind of remote sensing images Active CN106557740B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610910833.0A CN106557740B (en) 2016-10-19 2016-10-19 The recognition methods of oil depot target in a kind of remote sensing images

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610910833.0A CN106557740B (en) 2016-10-19 2016-10-19 The recognition methods of oil depot target in a kind of remote sensing images

Publications (2)

Publication Number Publication Date
CN106557740A true CN106557740A (en) 2017-04-05
CN106557740B CN106557740B (en) 2019-05-03

Family

ID=58443331

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610910833.0A Active CN106557740B (en) 2016-10-19 2016-10-19 The recognition methods of oil depot target in a kind of remote sensing images

Country Status (1)

Country Link
CN (1) CN106557740B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108734158A (en) * 2017-04-14 2018-11-02 成都唐源电气股份有限公司 A kind of real-time train number identification method and device
CN108805057A (en) * 2018-05-29 2018-11-13 北京师范大学 A kind of SAR image oil depot area detection method based on joint significance analysis

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102043958A (en) * 2010-11-26 2011-05-04 华中科技大学 High-definition remote sensing image multi-class target detection and identification method
US20110142282A1 (en) * 2009-12-14 2011-06-16 Indian Institute Of Technology Bombay Visual object tracking with scale and orientation adaptation
CN104463248A (en) * 2014-12-09 2015-03-25 西北工业大学 High-resolution remote sensing image airplane detecting method based on high-level feature extraction of depth boltzmann machine

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110142282A1 (en) * 2009-12-14 2011-06-16 Indian Institute Of Technology Bombay Visual object tracking with scale and orientation adaptation
CN102043958A (en) * 2010-11-26 2011-05-04 华中科技大学 High-definition remote sensing image multi-class target detection and identification method
CN104463248A (en) * 2014-12-09 2015-03-25 西北工业大学 High-resolution remote sensing image airplane detecting method based on high-level feature extraction of depth boltzmann machine

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108734158A (en) * 2017-04-14 2018-11-02 成都唐源电气股份有限公司 A kind of real-time train number identification method and device
CN108734158B (en) * 2017-04-14 2020-05-19 成都唐源电气股份有限公司 Real-time train number identification method and device
CN108805057A (en) * 2018-05-29 2018-11-13 北京师范大学 A kind of SAR image oil depot area detection method based on joint significance analysis
CN108805057B (en) * 2018-05-29 2020-11-17 北京师范大学 SAR image reservoir area detection method based on joint significance analysis

Also Published As

Publication number Publication date
CN106557740B (en) 2019-05-03

Similar Documents

Publication Publication Date Title
Xie et al. Multilevel cloud detection in remote sensing images based on deep learning
CN106874894B (en) Human body target detection method based on regional full convolution neural network
CN104166841B (en) The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network
CN104809443B (en) Detection method of license plate and system based on convolutional neural networks
Shao et al. Cloud detection in remote sensing images based on multiscale features-convolutional neural network
Huang et al. A multidirectional and multiscale morphological index for automatic building extraction from multispectral GeoEye-1 imagery
CN105095856B (en) Face identification method is blocked based on mask
CN104050471B (en) Natural scene character detection method and system
CN106778835B (en) Remote sensing image airport target identification method fusing scene information and depth features
Brandtberg et al. Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis
CN103824070B (en) A kind of rapid pedestrian detection method based on computer vision
CN102542289B (en) Pedestrian volume statistical method based on plurality of Gaussian counting models
Huang et al. Morphological building/shadow index for building extraction from high-resolution imagery over urban areas
CN102915438B (en) The extracting method of a kind of video caption and device
US6661907B2 (en) Face detection in digital images
KR20160143494A (en) Saliency information acquisition apparatus and saliency information acquisition method
CN107563355B (en) Hyperspectral anomaly detection method based on generation of countermeasure network
CN107341488B (en) SAR image target detection and identification integrated method
CN106909941A (en) Multilist character recognition system and method based on machine vision
CN105404861A (en) Training and detecting methods and systems for key human facial feature point detection model
Nguyen et al. Robust minutiae extractor: Integrating deep networks and fingerprint domain knowledge
CN107153817B (en) Pedestrian re-identification data labeling method and device
TWI281126B (en) Image detection method based on region
CN102609720B (en) Pedestrian detection method based on position correction model
Hu et al. Automatic segmentation of high-resolution satellite imagery by integrating texture, intensity, and color features

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