CN109165636A - A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion - Google Patents
A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion Download PDFInfo
- Publication number
- CN109165636A CN109165636A CN201811143808.XA CN201811143808A CN109165636A CN 109165636 A CN109165636 A CN 109165636A CN 201811143808 A CN201811143808 A CN 201811143808A CN 109165636 A CN109165636 A CN 109165636A
- Authority
- CN
- China
- Prior art keywords
- image
- birds
- feature
- bird
- target frame
- 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
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (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)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of sparse recognition methods of the Rare Birds based on component-level multiple features fusion, obtain head, body and the target frame three parts information of birds first;Then respectively by three parts information extraction color, the overall situation and local feature, by these three Fusion Features, to obtain the feature vector for all parts information that can characterize birds in picture and feature vector formed to the sparse dictionary of thus training image;Rarefaction representation is carried out to test sample followed by training sample, the final property for utilizing rarefaction representation coefficient judges the generic of test sample, and the generic that judgement is obtained exports.The Global Dictionary without classification rule of dictionary data set composition more all for the characterization ability of certain kinds is strong in the method for the present invention; it can influence to avoid posture, illumination, shooting angle to birds image recognition; the accuracy rate and robustness that birds identification can be improved have practicability in the round-the-clock unattended video monitoring protection of Rare Birds.
Description
Technical field
The present invention relates to a kind of sparse recognition methods of the Rare Birds based on component-level multiple features fusion, belong to image procossing
Technical field.
Background technique
Currently, the living environment of some birds is affected, and causes it in the case where global environment destroys the background got worse
Endangered danger, therefore, to rare and endangered dirds carry out protection it is imperative.But in the situation that specific category is unknown
Under, it is difficult to take the measure of being effectively protected.Therefore, carry out birds identification to protection Endangered birds, there is biggish ecology and society
It can meaning.Initially, birds know method for distinguishing focus utilization sound identified, such as patent 2013105810072 propose one
Mobile birds recognition methods of the kind based on chirm, but because acquiring the influence of noise in distance and acquired sound,
Recognition result is difficult to keep higher accuracy.With the development and universal, more and more recognition methods of monitoring device and technology
It is conceived to and obtains Rare Birds image using video monitoring equipment, and obtains Precious, Rare, Endangered bird in certain region by identification technology
The type of class, to realize the effective protection to rare and endangered dirds.Therefore, carry out the image recognition research of rare birds, have
There are important learning value and realistic meaning.
It is exactly the sophisticated category problem of birds image on the identification question essence of birds image, passes through the inspection to the prior art
The method of Suo Faxian, birds image sophisticated category are divided into two kinds: one is Weakly supervised classification method.The starting point of such method
It is to be identified end to end without using other artificial markup informations other than class label to realize.The method of representative has
" the Bilinear that Lin et al. is delivered on " IEEE International Conference on Computer Vision "
CNN models for fine-grained visual recognition " is carried out end to end using bilinear model
Identification.The second is the classification method supervised by force, the method for representative has: Zhang N. et al. is in " IEEE International
Conference on Computer Vision " on " the Part-based R-CNNs for fine-grained that delivers
Category detection " proposes the R-CNN based on component for birds image-recognizing method, i.e., obtains bird using RCNN
Base part information;" the Bird species categorization using pose that Branson S. et al. is delivered
Normalized deep convolutional nets " proposes a kind of normalized recognition methods of posture, and the algorithm is first
The positioning of localized region is first carried out, and original image is cut according to callout box information, is believed according to the component extracted
Breath carries out the alignment operation of posture, extracts convolution feature again later, is classified using SVM.
In view of above-mentioned several method is all the research carried out under the method for deep learning, the raising of recognition accuracy
It is premised on largely calculate, this large amount of calculate needs powerful hardware supported.Also, it is based on convolutional Neural net
The deep learning algorithm of network needs to carry out the training of model with a large amount of training sample, and for some Rare Birds, it obtains big
The birds picture difficulty of amount is larger, the lazy weight of the picture of acquisition with the preferable convolutional neural networks model of training, even
Preparatory trained network model is finely adjusted, it is also difficult to avoid the problem that data over-fitting is brought, which limits identifications
The raising of rate.
Summary of the invention
Purpose: in order to overcome the deficiencies in the prior art, the present invention provides a kind of based on component-level multiple features fusion
The sparse recognition methods of Rare Birds.
Technical solution: in order to solve the above technical problems, the technical solution adopted by the present invention are as follows:
A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion, comprises the following steps that
Step 1: for all birds images in training image, using the head of birds known to training image, body and
The location information of entire bird three parts target frame, color, texture, the overlapping and size for obtaining three parts target frame respectively are similar
Above four weight coefficient heads, body and entire bird are grouped, and find out average value by the weight coefficient of degree, respectively
It obtains the head and marks frame color, texture, overlapping and size similarity weight coefficientPhysical target frame face
Color, texture, overlapping and size similarity weight coefficientEntire bird target frame color, overlaps texture
And the weight coefficient of size similarity
Step 2: for each birds image in test image, utilizing acquisition three parts target frame color, line in step 1
Reason, overlapping and size similarity weight coefficient, head, the body of birds in test image are obtained using the method for selective search
Body and entire bird three parts target frame.
Step 3: extracting the color characteristic of the three parts target frame of birds in training image, local feature and global spy
These features are carried out fusion and form training sample eigenvectors matrix and composition training sample dictionary by sign.
Step 4: extract in test image the color characteristic of three parts target frame of each birds image, local feature with
And global characteristics, and these features are subjected to fusion and form test sample feature.
Step 5: rarefaction representation coefficient of the test sample feature under training sample dictionary is obtained by sparse algorithm,
And final birds are obtained according to the property of rarefaction representation coefficient and classifies and exports.
Preferably, training image described in step 1 include 200 kinds of birds difference postures, different shooting angles and
The image of illumination condition, every kind of birds image have 30.
Preferably, in step 2 selective search method, the specific steps are as follows:
2.1: using area suggesting method respectively divides the three parts image of the head of test image, body, entire bird
It cuts, obtains original header point cut zoneRaw body partial segmentation region
Original entire bird partial segmentation region
2.2: calculating separately three parts cut zone R1、R2、R3Similarity S between adjacent area two-by-twon(ri n,rj n), and
It is added to similarity set SnIn, specific formula is as follows:
Wherein, Scolour(ri n,rj n) it is color similarity, Stexture(ri n,rj n) it is texture similarity, Sfill(ri n,rj n)
To overlap similarity, Ssize(ri n,rj n) it is size similarity,For three parts target frame color, texture, friendship
Folded and size similarity weight coefficient, n are { 1,2,3 } one of numerical value.
2.3: respectively from similarity set SnIn find out the maximum two region r of similarityi nAnd rj n, it is merged into being one
A region rt n;And region r is removed from similarity seti nAnd rj nSimilarity, calculate rt nThe similarity in region adjacent thereto,
Similarity set S is arrived by what its result was addednIn;Simultaneously by new region rt nIt is respectively added to regional ensemble R1、R2、R3In.
2.4: step 2.3 is repeated, until similarity set SnWhen for sky, obtained region is test sample birds figure
Head, body and the entire bird three parts target frame of picture.
Preferably, specific step is as follows for the step 3:
3.1: RGB (color), GIST (part), PHOG are extracted respectively to known training image three parts target block diagram picture
(overall situation) feature,Respectively indicate head, body, entire bird color characteristic,Respectively indicate head, body, entire bird local feature, Respectively indicate head, body, entire bird global characteristics.
3.2: three kinds of features grouping that training image three parts target frame extracts being merged, head subhead is respectively obtained
Mark the fusion feature of frameThe fusion feature of body part target frameThe fusion feature of entire bird partial target frameThe fusion feature of three parts target frame is merged again, is obtained
One fusion feature V corresponding with training imageR={ (Vhead)T,(Vbody)T,(Vbbox)T}TThat is training sample feature vector,
Fusion is carried out using concatenated mode.
3.3: by training sample feature vector VRIt is grouped by bird type, each type training sample eigenvectors matrix
For Di={ (VR)1,(VR)2,...(VR)n, wherein that i is represented is the type of bird, (VR)nRepresent n-th birds figure in such bird
The feature vector of picture.
3.4: the training sample eigenvectors matrix of all kinds bird is formed into training sample dictionary D={ D1, D2...,
DN, wherein N is the total quantity of the type of bird.
Preferably, specific step is as follows for the step 4:
4.1: to the head of picture each in test image, body, entire bird three parts target frame extract respectively RGB,
GIST, PHOG feature,Respectively indicate head, body, entire bird color characteristic; Respectively indicate head, body, entire bird local feature;
Respectively indicate head, body, entire bird global characteristics.
4.2: three kinds of features grouping that test image three parts target frame extracts being merged, head subhead is respectively obtained
Mark the fusion feature of frameThe fusion feature of body part target frameThe fusion feature of entire bird partial target frameThe fusion feature of three parts target frame is merged again, is surveyed
Sample eigen y={ (Thead)T,(Tbody)T,(Tbbox)T}T, merge and carried out using concatenated mode.
Preferably, in RGB feature extraction process, feature is extracted using triple channel, forms one 256 × 3 dimension
Color feature vector;
In GIST characteristic extraction procedure, is averaged using Gabor filter group to image be reconverted into gray value first
Then image carries out pre-filtering to gray level image, the scaling of the contrast of part is carried out by the division of sparse grid, is finally divided
Block calculates Gabor characteristic and extracts and be combined, and image zooming-out is one 512 feature vector tieed up;
In PHOG characteristic extraction procedure, the region being connected to one by one is divided the image into first, then acquires each connected region
These histograms are stitched together and constitute the target area by the direction histogram at the gradient of each pixel or edge in domain
The description of HOG feature;Inside same division scalogram, each region calculates a HOG feature, is spliced in order, is obtained
The HOG feature of current scale figure;The HOG feature that each layer scalogram extracts is spliced, the PHOG for just obtaining whole image is special
Sign, image zooming-out are characterized in the feature vector of one 680 dimension.
Preferably, specific step is as follows for step 5:
5.1: sparse solution being carried out according to the training sample dictionary D of acquisition, rarefaction representation coefficient x is sought, such as formula (2)
It is shown:
min||wij||1subject to||y-Dx||2< ε (2)
Wherein, formula is utilized | | y-Dx | |2< ε, the set of all x found out, selection meet wijThe smallest x of absolute value is
Rarefaction representation coefficient, wijCoefficient corresponding with jth picture in the i-th class of training image in x is represented, ε, which is represented, calculates residual error, x=
{(w11,w12,...,w1(n-1),w1n),(w21,w22,...,w2(g-1),w2g),...,(wm1,wm2,...,wm(h-1),wmh)}TIt represents
Required rarefaction representation coefficient, wm(h-1)Represent coefficient corresponding with the h-1 picture of training image m class in x.
5.2: the maximum value of every a kind of bird rarefaction representation coefficient average value is sought using following formula (3):
Wherein, w11Indicate the rarefaction representation of the 1st training sample image feature vector in the corresponding 1st class birds of test image
Coefficient value, wm(h-1)Indicate the rarefaction representation of h-1 training sample image feature vectors in the corresponding m class birds of test image
Coefficient value, n, g, h respectively represents the 1st, 2, the quantity of m class birds training image.
5.3: the type according to bird corresponding to the maximum value of rarefaction representation coefficient average value is exactly belonging to test image
The classification of bird.
The utility model has the advantages that the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion provided by the invention,
The component information for obtaining birds image, can overcome the influence of posture, shooting angle to recognition result;Class used in the present invention
Sample dictionary is better than the characterization ability of particular category the Global Dictionary of all data set generations;RGB, GIST, PHOG tri-
Kind feature combines, and can not only retain the global characteristics of image, can also protrude partial gradient feature and color characteristic;We
Method has preferable robustness and preferable discrimination, all to the target image to be sorted of the variation of illumination, posture, shooting angle
Obtain preferable classifying quality.
Detailed description of the invention
Fig. 1 is method flow schematic diagram of the invention.
Specific embodiment
The present invention will be further explained with reference to the accompanying drawing.
As shown in Figure 1, being the principle framework of one embodiment of the invention, which includes three parts, i.e. birds image
Head, body, the position portion of entire bird, feature extraction with merge part and rarefaction representation classified part.A given width is surveyed
Attempt picture, first with selective search obtain target object head, body, entire bird part;Then feature extraction and fusion
Module carries out extraction and the fusion of feature to the three parts image of extraction respectively;Finally training sample figure is obtained using sparse solution
As the sparse expression in the case where testing dictionary, and obtain final classification.Red triangular, Blue circles, yellow pentagon in figure
Respectively represent tri- kinds of features of extracted PHOG, GIST, RGB.
In feature 1:RGB characteristic extraction procedure, respectively by head, body, entire bird three parts target frame image zooming-out R,
G, B triple channel feature forms the color feature vector V of one 256 × 3 dimensionRGB, RGB feature calculation formula is as follows:
nmIt is the number of pixels fallen on color value m, N is total number of pixels.
In feature 2:GIST characteristic extraction procedure, flat are asked to three kinds of component original images using Gabor filter group first
Mean value is reconverted into gray-value image, then carries out pre-filtering to gray level image, carries out part by the division of sparse grid
The scaling of contrast, last piecemeal calculate Gabor characteristic and extract and be combined, and extraction is one 512 feature vector tieed up
VGIST。
In feature 3:PHOG characteristic extraction procedure, image of component is divided into the region of connection one by one first, is then adopted
Collect the direction histogram at the gradient of each pixel or edge in each connected region, these histograms are stitched together and are just constituted
The HOG feature of the target area describes.Inside same division scalogram, each region calculates a HOG feature, in order into
Row splicing, obtains the HOG feature of current scale figure;The HOG feature that each layer scalogram extracts is spliced, is just entirely schemed
PHOG feature this feature of picture is the feature vector V of one 680 dimensionPHOG, the formula used is as follows:
Wherein, IxAnd IyThe gradient value on both horizontally and vertically is represented, M (x, y) represents the range value of gradient, θ (x, y)
Represent the direction of gradient.
During solving sparse coefficient, orthogonal matching algorithm OMP is picked out in the update of every step to be had with current residue
There is the atom of highest correlation.After having selected these atoms, sky that signal is just opened by rectangular projection to these atoms
Between in, then recalculate signal residual error again, repeat this process until meeting stop condition.
The present invention preferably resolves the positioning parts problem of birds image, and completes the extraction to component feature and melt
It closes, and feature preferably describes the part of birds image of component and global characteristics.To preferably complete birds
The rarefaction representation of image, and then improve the accuracy of identification of Rare Birds image.
The above is only a preferred embodiment of the present invention, it should be pointed out that: for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (7)
1. a kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion, it is characterised in that: comprise the following steps that
Step 1: for all birds images in training image, utilizing the head of birds known to training image, body and entire
The location information of bird three parts target frame obtains the color of three parts target frame, texture, overlapping and size similarity respectively
Above four weight coefficient heads, body and entire bird are grouped, and find out average value by weight coefficient, respectively obtain
The head marks frame color, texture, overlapping and size similarity weight coefficientPhysical target frame color, line
Reason, overlapping and size similarity weight coefficientEntire bird target frame color, texture, it is overlapping and
The weight coefficient of size similarity
Step 2: for each birds image in test image, using obtained in step 1 three parts target frame color, texture,
Overlapping and size similarity weight coefficient, head, the body of birds in test image are obtained using the method for selective search
And entire bird three parts target frame;
Step 3: color characteristic, local feature and the global characteristics of the three parts target frame of birds in training image are extracted, it will
These features carry out fusion and form training sample eigenvectors matrix and composition training sample dictionary;
Step 4: extracting the color characteristic of three parts target frame of each birds image in test image, local feature and complete
Office's feature, and these features are subjected to fusion and form test sample feature;
Step 5: rarefaction representation coefficient of the test sample feature under training sample dictionary, and root are obtained by sparse algorithm
Final birds are obtained according to the property of rarefaction representation coefficient to classify and export.
2. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 1, special
Sign is: training image described in the step 1 includes 200 kinds of birds difference postures, different shooting angles and illumination condition
Image, every kind of birds image have 30.
3. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 1, special
Sign is: the method for selective search in the step 2, the specific steps are as follows:
2.1: using area suggesting method is respectively split the three parts image of the head of test image, body, entire bird, obtains
Divide cut zone to original headerRaw body partial segmentation regionIt is original
Entire bird partial segmentation region
2.2: calculating separately three parts cut zone R1、R2、R3Similarity S between adjacent area two-by-twon(ri n,rj n), and add
To similarity set SnIn, specific formula is as follows:
Wherein, Scolour(ri n,rj n) it is color similarity, Stexture(ri n,rj n) it is texture similarity, Sfill(ri n,rj n) it is to hand over
Folded similarity, Ssize(ri n,rj n) it is size similarity,For three parts target frame color, texture, it is overlapping with
And size similarity weight coefficient, n are { 1,2,3 } one of numerical value;
2.3: respectively from similarity set SnIn find out the maximum two region r of similarityi nAnd rj n, it is merged into as an area
Domain rt n;And region r is removed from similarity seti nAnd rj nSimilarity, calculate rt nThe similarity in region adjacent thereto, by it
That as a result adds arrives similarity set SnIn;Simultaneously by new region rt nIt is respectively added to regional ensemble R1、R2、R3In;
2.4: step 2.3 is repeated, until similarity set SnWhen for sky, obtained region is test sample birds image
Head, body and entire bird three parts target frame.
4. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 1, special
Sign is: specific step is as follows for the step 3:
3.1: to known training image three parts target block diagram picture extract respectively RGB (color), GIST (part),
PHOG (overall situation) feature,Respectively indicate head, body, entire bird color characteristic,Respectively indicate head, body, entire bird local feature, Respectively indicate head, body, entire bird global characteristics;
3.2: three kinds of features grouping that training image three parts target frame extracts being merged, head partial objectives for frame is respectively obtained
Fusion featureThe fusion feature of body part target frameThe fusion feature of entire bird partial target frameThe fusion feature of three parts target frame is merged again, is obtained
One fusion feature V corresponding with training imageR={ (Vhead)T,(Vbody)T,(Vbbox)T}TThat is training sample feature vector,
Fusion is carried out using concatenated mode;
3.3: by training sample feature vector VRIt is grouped by bird type, each type training sample eigenvectors matrix is Di
={ (VR)1,(VR)2,...(VR)n, wherein that i is represented is the type of bird, (VR)nRepresent in such bird n-th birds image
Feature vector;
3.4: the training sample eigenvectors matrix of all kinds bird is formed into training sample dictionary D={ D1, D2..., DN,
Middle N is the total quantity of the type of bird.
5. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 1, special
Sign is: specific step is as follows for the step 4:
4.1: to the head of picture each in test image, body, entire bird three parts target frame extract respectively RGB, GIST,
PHOG feature,Respectively indicate head, body, entire bird color characteristic;Respectively indicate head, body, entire bird local feature; Respectively indicate head, body, entire bird global characteristics;
4.2: three kinds of features grouping that test image three parts target frame extracts being merged, head partial objectives for frame is respectively obtained
Fusion featureThe fusion feature of body part target frameThe fusion feature of entire bird partial target frameThe fusion feature of three parts target frame is merged again, is surveyed
Sample eigen y={ (Thead)T,(Tbody)T,(Tbbox)T}T, merge and carried out using concatenated mode.
6. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 4 or 5,
It is characterized in that: in the RGB feature extraction process, feature being extracted using triple channel, forms the color characteristic of one 256 × 3 dimension
Vector;
In the GIST characteristic extraction procedure, is averaged using Gabor filter group to image be reconverted into gray value first
Then image carries out pre-filtering to gray level image, the scaling of the contrast of part is carried out by the division of sparse grid, is finally divided
Block calculates Gabor characteristic and extracts and be combined, and image zooming-out is one 512 feature vector tieed up;
In the PHOG characteristic extraction procedure, the region being connected to one by one is divided the image into first, then acquires each connected region
These histograms are stitched together and constitute the target area by the direction histogram at the gradient of each pixel or edge in domain
The description of HOG feature;Inside same division scalogram, each region calculates a HOG feature, is spliced in order, is obtained
The HOG feature of current scale figure;The HOG feature that each layer scalogram extracts is spliced, the PHOG for just obtaining whole image is special
Sign, image zooming-out are characterized in the feature vector of one 680 dimension.
7. the sparse recognition methods of a kind of Rare Birds based on component-level multiple features fusion according to claim 1, special
Sign is: specific step is as follows for the step 5:
5.1: sparse solution is carried out according to the training sample dictionary D of acquisition, seeks rarefaction representation coefficient x, as shown in formula 2:
min||wij||1subject to||y-Dx||2< ε (2)
Wherein, formula is utilized | | y-Dx | |2< ε, the set of all x found out, selection meet wijThe smallest x of absolute value is sparse
Indicate coefficient, wijCoefficient corresponding with jth picture in the i-th class of training image in x is represented, ε, which is represented, calculates residual error, x=
{(w11,w12,...,w1(n-1),w1n),(w21,w22,...,w2(g-1),w2g),...,(wm1,wm2,...,wm(h-1),wmh)}TIt represents
Required rarefaction representation coefficient, wm(h-1)Represent coefficient corresponding with h-1 picture in training image m class in x;
5.2: the maximum value of every a kind of bird rarefaction representation coefficient average value is sought using following equation 3:
Wherein, w11Indicate the rarefaction representation coefficient of the 1st training sample image feature vector in the corresponding 1st class birds of test image
Value, wm(h-1)Indicate the rarefaction representation coefficient of h-1 training sample image feature vectors in the corresponding m class birds of test image
Value, n, g, h respectively represents the 1st, 2, the quantity of m class birds training image;
5.3: the type according to bird corresponding to the maximum value of rarefaction representation coefficient average value is exactly bird belonging to test image
Classification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811143808.XA CN109165636A (en) | 2018-09-28 | 2018-09-28 | A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811143808.XA CN109165636A (en) | 2018-09-28 | 2018-09-28 | A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109165636A true CN109165636A (en) | 2019-01-08 |
Family
ID=64892805
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811143808.XA Pending CN109165636A (en) | 2018-09-28 | 2018-09-28 | A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109165636A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110008850A (en) * | 2019-03-15 | 2019-07-12 | 交通运输部天津水运工程科学研究所 | A kind of refuge plants screening technique based on birds color extraction |
CN110633727A (en) * | 2019-06-28 | 2019-12-31 | 中国人民解放军军事科学院国防科技创新研究院 | Deep neural network ship target fine-grained identification method based on selective search |
CN110837771A (en) * | 2019-09-17 | 2020-02-25 | 常熟理工学院 | Bird identification method and device based on convolutional neural network |
CN113887447A (en) * | 2021-10-08 | 2022-01-04 | 中国科学院半导体研究所 | Training method of object classification model, object classification prediction method and device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160042252A1 (en) * | 2014-08-05 | 2016-02-11 | Sri International | Multi-Dimensional Realization of Visual Content of an Image Collection |
CN105447503A (en) * | 2015-11-05 | 2016-03-30 | 长春工业大学 | Sparse-representation-LBP-and-HOG-integration-based pedestrian detection method |
CN105631469A (en) * | 2015-12-18 | 2016-06-01 | 华南理工大学 | Bird image recognition method by multilayer sparse coding features |
CN107220651A (en) * | 2017-06-05 | 2017-09-29 | 北京邮电大学 | A kind of method and device for extracting characteristics of image |
CN108197591A (en) * | 2018-01-22 | 2018-06-22 | 北京林业大学 | A kind of birds individual discrimination method based on multiple features fusion transfer learning |
CN108280421A (en) * | 2018-01-22 | 2018-07-13 | 湘潭大学 | Human bodys' response method based on multiple features Depth Motion figure |
-
2018
- 2018-09-28 CN CN201811143808.XA patent/CN109165636A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160042252A1 (en) * | 2014-08-05 | 2016-02-11 | Sri International | Multi-Dimensional Realization of Visual Content of an Image Collection |
CN105447503A (en) * | 2015-11-05 | 2016-03-30 | 长春工业大学 | Sparse-representation-LBP-and-HOG-integration-based pedestrian detection method |
CN105631469A (en) * | 2015-12-18 | 2016-06-01 | 华南理工大学 | Bird image recognition method by multilayer sparse coding features |
CN107220651A (en) * | 2017-06-05 | 2017-09-29 | 北京邮电大学 | A kind of method and device for extracting characteristics of image |
CN108197591A (en) * | 2018-01-22 | 2018-06-22 | 北京林业大学 | A kind of birds individual discrimination method based on multiple features fusion transfer learning |
CN108280421A (en) * | 2018-01-22 | 2018-07-13 | 湘潭大学 | Human bodys' response method based on multiple features Depth Motion figure |
Non-Patent Citations (3)
Title |
---|
JIXIN LIU 等: "Rare Bird Sparse Recognition via Part-Based Gist Feature Fusion and Regularized Intraclass Dictionary Learning", 《COMPUTERS, MATERIALS & CONTINUA》 * |
刘佶鑫 等: "可见光-近红外HSV图像融合的场景类字典稀疏识别方法", 《计算机应用》 * |
汪明明: "基于物体部件模型的细粒度图像识别", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110008850A (en) * | 2019-03-15 | 2019-07-12 | 交通运输部天津水运工程科学研究所 | A kind of refuge plants screening technique based on birds color extraction |
CN110008850B (en) * | 2019-03-15 | 2023-01-10 | 交通运输部天津水运工程科学研究所 | Refuge plant screening method based on bird color extraction |
CN110633727A (en) * | 2019-06-28 | 2019-12-31 | 中国人民解放军军事科学院国防科技创新研究院 | Deep neural network ship target fine-grained identification method based on selective search |
CN110837771A (en) * | 2019-09-17 | 2020-02-25 | 常熟理工学院 | Bird identification method and device based on convolutional neural network |
CN113887447A (en) * | 2021-10-08 | 2022-01-04 | 中国科学院半导体研究所 | Training method of object classification model, object classification prediction method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106845357B (en) | A kind of video human face detection and recognition methods based on multichannel network | |
CN108520226B (en) | Pedestrian re-identification method based on body decomposition and significance detection | |
CN109165636A (en) | A kind of sparse recognition methods of Rare Birds based on component-level multiple features fusion | |
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN105574550B (en) | A kind of vehicle identification method and device | |
CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
CN105740780B (en) | Method and device for detecting living human face | |
CN109558810B (en) | Target person identification method based on part segmentation and fusion | |
CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
CN108319953A (en) | Occlusion detection method and device, electronic equipment and the storage medium of target object | |
CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
CN110046599A (en) | Intelligent control method based on depth integration neural network pedestrian weight identification technology | |
CN109190475B (en) | Face recognition network and pedestrian re-recognition network collaborative training method | |
CN106610969A (en) | Multimodal information-based video content auditing system and method | |
CN110263768A (en) | A kind of face identification method based on depth residual error network | |
CN112069891B (en) | Deep fake face identification method based on illumination characteristics | |
CN109063649A (en) | Pedestrian's recognition methods again of residual error network is aligned based on twin pedestrian | |
CN109711309B (en) | Method for automatically identifying whether portrait picture is eye-closed | |
CN110443179A (en) | It leaves the post detection method, device and storage medium | |
CN105844245A (en) | Fake face detecting method and system for realizing same | |
CN109271932A (en) | Pedestrian based on color-match recognition methods again | |
CN112183438B (en) | Image identification method for illegal behaviors based on small sample learning neural network | |
CN112183472A (en) | Method for detecting whether test field personnel wear work clothes or not based on improved RetinaNet | |
CN109670457A (en) | A kind of driver status recognition methods and device | |
CN107944395A (en) | A kind of method and system based on neutral net verification testimony of a witness unification |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190108 |
|
RJ01 | Rejection of invention patent application after publication |