CN107025431A - The insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study - Google Patents

The insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study Download PDF

Info

Publication number
CN107025431A
CN107025431A CN201710105763.6A CN201710105763A CN107025431A CN 107025431 A CN107025431 A CN 107025431A CN 201710105763 A CN201710105763 A CN 201710105763A CN 107025431 A CN107025431 A CN 107025431A
Authority
CN
China
Prior art keywords
image
training
insect
multinuclear
block
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
CN201710105763.6A
Other languages
Chinese (zh)
Other versions
CN107025431B (en
Inventor
陈天娇
谢成军
余健
张洁
李�瑞
陈红波
王儒敬
宋良图
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.)
Hefei Institutes of Physical Science of CAS
Original Assignee
Hefei Institutes of Physical Science of CAS
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 Hefei Institutes of Physical Science of CAS filed Critical Hefei Institutes of Physical Science of CAS
Priority to CN201710105763.6A priority Critical patent/CN107025431B/en
Publication of CN107025431A publication Critical patent/CN107025431A/en
Application granted granted Critical
Publication of CN107025431B publication Critical patent/CN107025431B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • 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

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)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The present invention relates to the insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study, the low defect of insect image recognition rate is solved compared with prior art.The present invention comprises the following steps：Obtain nature image set；Collection, mark and the pretreatment of insect image；Utilize the sample formation block level feature of training set；Multinuclear taxonomy model is trained using the multinuclear Boosting SVM learnt；After the sample formation block level feature of test set, in the multinuclear taxonomy model after the completion of input training, the automatic identification of insect image is carried out.Learn to be combined the image recognition for carrying out insect present invention utilizes the multilayer depth characteristic of insect image and with multinuclear Boosting, improve the accuracy rate of insect identification.

Description

The insect image recognition being combined based on depth characteristic with multinuclear Boosting study Method
Technical field
The present invention relates to image identification technical field, specifically learnt based on depth characteristic and multinuclear Boosting The insect image-recognizing method being combined.
Background technology
Insect is always the basic problem for perplexing crop growth, because plant insect symptom is initially very fuzzy, is increased The difficulty manually estimated.The identification of insect image due to its floristic diversity, the polytropy of variety classes plant insect, So that traditional automatic identifying method discrimination is not high, robustness is poor, and it can be only present in the experimental stage.Therefore, how The accuracy for enough improving insect image recognition has become the technical problem for being badly in need of solving.
The content of the invention
The invention aims to solve the low defect of insect image recognition rate in the prior art to be based on deeply there is provided one kind Degree feature solves the above problems with the insect image-recognizing method that multinuclear Boosting study is combined.
To achieve these goals, technical scheme is as follows：
A kind of insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study, including following step Suddenly：
Nature image set is obtained, nature image set is obtained and is divided into natural image block, use unsupervised feature learning side Method trains study dictionary from unlabelled natural image block；
All insect images are shot by collection, mark and the pretreatment of insect image under conditions of maintaining uniform illumination, and Shooting is focused in insect symptom, by all insect image normalizations and size is zoomed to for 200 × 200 pixels, if obtaining Dry sample；The sample of every kind of insect is randomly divided into training set and test set；
Using the sample formation block level feature of training set, marked training image is extracted into image from different scale Block, the training image blocks collection under certain yardstick is carried out after rarefaction representation by learning dictionary, then is operated by spatial alignment pondization Form block level feature；
Multinuclear taxonomy model is trained using the multinuclear Boosting SVM learnt, the block level feature under different scale is passed through Multinuclear taxonomy model is trained based on the multinuclear Boosting SVMs learnt；
After the sample formation block level feature of test set, in the multinuclear taxonomy model after the completion of input training, insect is carried out The automatic identification of image.
Described acquisition natural image comprises the following steps：
Obtain nature image set Y, Y=[y1,y2,...yi...], yiRepresent i-th natural image；Utilize uniform grid Concentrated from natural image and extract N number of natural image block x altogetheri, xi∈Rn, (i=1,2 ..., N), each of which natural image K blocks are extracted, and to natural image block xiDo normalized；
Using singular value decomposition algorithm training study dictionary, solve following optimization problem and obtain study dictionary：
D=[d1,d2,...,dM]∈Rn×MStudy dictionary is represented,It is nature figure As block xiCorresponding sparse coding,It is natural image block xiRarefaction representation vector.
The sample formation block level feature of described utilization training set comprises the following steps：
Marked training image is extracted into image block from different scale, from coarseness level, middle granular level and thin Image block is extracted to training image respectively on granular level different scale；
For a training image yi, K insect image block is obtained using the uniform grid segmentation of some scale size [x1,x2,...,xK]；
Training image blocks under certain yardstick are subjected to rarefaction representation by learning dictionary；
Image block sparse coding is comprised the following steps using dictionary：
For given insect image yi, overlapping square [x is segmented the image into using uniform grid1,x2,...,xK]∈ Rn×K；
Study dictionary D, K insect image block [x are obtained using singular value decomposition algorithm study1,x2,...,xK] corresponding Sparse coding is respectivelyFind each piece of xiSparse codeFollowing optimization problem is solved,
Wherein | | | |0It isThe number of middle nonzero element；
Take l1Norm minimum is substituted, and solves following optimization problem：
WhereinSolution be corresponding topography's block vector xiCorresponding characteristic vector,Correspond to The sparse coding of all topography's blocks in one insect image；
Block level feature is formed by the operation of spatial alignment pond, it comprises the following steps：
I-th piece of sparse codingJ sections are divided into,WhereinTable Show coefficientJth section；
ThenIt is weighted, its formula is as follows：
Wherein, vector viCorresponding to i-th of localized mass；
The corresponding all localized masses of one training image distinguish corresponding characteristic vector viForm matrix V；
Pondization operation is done to matrix V, takes the diagonal element of matrix V to produce final block level feature, its formula is as follows：
S=diag (V),
Wherein：S represents the characteristic vector extracted from an a certain yardstick of training image.
The SVM training multinuclear taxonomy models of described utilization multinuclear Boosting study comprise the following steps：
Training set is inputted in SVMFeature set { the f of different scale1,f2,...fN, nuclear matrix { K1, K2,...KM, initialization strong classifier F (x)=0,
Wherein, xiRepresent training sample, yiRepresent that mark, the D of training sample represent number of training；
Each is trained to include f for each n ∈ N, m ∈ MNAnd KMCorresponding single SVMmnParameter amnAnd bmn,
hmn=Kmn(x,xi)Tamn+bmn；
Initialization sample weightBy training sample assignment identical initial weight；In each iteration, select Select the minimum SVM classifier of error in classification in data set；
If iterations l=1 to L；
Its error in classification ε is calculated for each SVMsm,n,
Wherein U (x) is a function, is otherwise -1 when x ＞ 0 are 1；
Select hl(x)=argmin εmn；
Calculate the weight of selection SVM classifier
If β1The then iteration ends of ＜ 0；Otherwise decision function F (x) ← β is updatedlhl, the grader of selection is added to In decision function；
WithThe weight of training sample is updated, the sample of classification error is assigned in next iteration Bigger weighted value；
Grader isExport the strong classifier being made up of multiple monokaryon SVM classifiers.
Beneficial effect
The method being combined based on depth characteristic with multinuclear Boosting study of the present invention, is utilized compared with prior art The multilayer depth characteristic of insect image simultaneously learns to be combined the image recognition for carrying out insect with multinuclear Boosting, improves evil The accuracy rate of worm identification, enhances the robustness of insect recognizer.
Brief description of the drawings
Fig. 1 is method precedence diagram of the invention.
Embodiment
To make to have a better understanding and awareness to architectural feature of the invention and the effect reached, to preferably Embodiment and accompanying drawing coordinate detailed description, are described as follows：
As shown in figure 1, a kind of insect being combined based on depth characteristic with multinuclear Boosting study of the present invention Image-recognizing method, comprises the following steps：
The first step, obtains nature image set.Obtain nature image set and be divided into natural image block, use unsupervised feature Learning method trains study dictionary from unlabelled natural image block.It is comprised the following steps that：
(1) nature image set Y, Y=[y are obtained1,y2,...yi...], yiRepresent i-th natural image.
Concentrated using uniform grid from natural image and extract N number of natural image block x altogetheri, xi∈Rn, (i=1,2 ..., N), each of which natural image extracts K blocks, and to natural image block xiDo the natural image point of normalized, i.e., one It is divided into multiple K blocks into K blocks, multiple natural images, its sum is N number of.
(2) application unsupervised-learning algorithm singular value decomposition algorithm training study dictionary, solves following optimization problem and obtains Learn dictionary：
D=[d1,d2,...,dM]∈Rn×MStudy dictionary is represented,It is nature figure As block xiCorresponding sparse coding,It is natural image block xiRarefaction representation vector.
Second step, collection, mark and the pretreatment of insect image.To all insect images under conditions of maintaining uniform illumination Shoot, to eliminate potential negative effect of the illumination change to classification performance, and shooting is focused in insect symptom.For calculating Efficiency considers, by all insect image normalizations and zooms to size for 200 × 200 pixels, obtains several samples；Will be every The sample for planting insect is randomly divided into training set and test set.
3rd step, utilizes the sample formation block level feature of training set.Marked training image is above carried from different scale Image block is taken, the training image blocks collection under certain yardstick is carried out after rarefaction representation by learning dictionary, then pass through spatial alignment pond Change operation and form block level feature.
When insect image undergoes big cosmetic variation, relatively small yardstick, larger image block level can be provided More preferable geometric properties, and less image block can obtain finer feature.In order to obtain the greater compactness of table of insect image Show, image is sampled from different scale herein：Coarseness level, middle granular level and fine granularity level.Each yardstick Tile size it is different, its scope is, for example, 3 × 3,5 × 5,7 × 7.Marked training insect image is passed through into difference Uniform grid under yardstick is divided into overlapping square, important in order to retain by image block using the dictionary sparse coding learnt Characteristic information remove incoherent redundancy, then block level feature is formed by spatial alignment pond.It is comprised the following steps that：
(1) marked training image is extracted into image block from different scale, from coarseness level, middle granular level and Image block is extracted to training image respectively on the horizontal different scale of fine granularity, i.e., training image entered from 3 different scales Row sampling, ultimately forms multiple characteristic vector s of 3 different scales, so as to constitute the feature set { f of different scale1,f2, ...fN}。
For a training image yi, utilize some yardstick (coarseness level, middle granular level or fine granularity level) The uniform grid segmentation of size obtains K insect image block [x1,x2,...,xK]。
(2) training image blocks under certain yardstick are subjected to rarefaction representation by learning dictionary；
Image block sparse coding is comprised the following steps using dictionary：
A, the insect image y for givingi, overlapping square [x is segmented the image into using uniform grid1,x2,...,xK] ∈Rn×K；
B, using singular value decomposition algorithm study obtain study dictionary D, K insect image block [x1,x2,...,xK] correspondence Sparse coding be respectivelyFind each piece of xiSparse codeFollowing optimization problem is solved,
Wherein | | | |0It isThe number of middle nonzero element；
Because l0Norm minimum is the problem of NP is difficult, therefore takes l1Norm minimum, solves following optimization problem：
WhereinSolution be corresponding topography's block vector xiCorresponding characteristic vector,Correspond to The sparse coding of all topography's blocks in one insect image.
(3) block level feature is formed by the operation of spatial alignment pond, it comprises the following steps：
A, i-th piece of sparse codingJ sections are divided into,Wherein Represent coefficientJth section；
B, thenIt is weighted, its formula is as follows：
Wherein, vector viCorresponding to i-th of localized mass；
The corresponding all localized masses of one training image distinguish corresponding characteristic vector viForm matrix V；
C, pondization operation is done to matrix V, take the diagonal element of matrix V to produce final block level feature, its formula is as follows：
S=diag (V),
Wherein：S represents the characteristic vector extracted from an a certain yardstick of training image.
4th step, multinuclear taxonomy model is trained using the multinuclear Boosting SVM learnt, and the block level under different scale is special Levy and multinuclear taxonomy model is trained by the SVMs learnt based on multinuclear Boosting.Complete insect image different scale After depth block level feature extraction, the training of insect image is realized using multinuclear Boosting study classification methods, in order to merge The depth characteristic of different scale, the adaptive learning of weights under different characteristic is completed by combining multiple kernel function modes.It has Body step is as follows：
(1) training set is inputted in SVMFeature set { the f of different scale1,f2,...fN, conventional nuclear matrix {K1,K2,...KM, initialization strong classifier F (x)=0, wherein, xiRepresent training sample, yiRepresent mark, the D of training sample Represent number of training.
(2) each is trained to include f for each n ∈ N, m ∈ MNAnd KMCorresponding single SVMmnParameter amnWith bmn,
hmn=Kmn(x,xi)Tamn+bmn。
(3) initialization sample weightBy training sample assignment identical initial weight；In iteration each time In, the minimum SVM classifier of error in classification in selection data set；
(4) iterations l=1 to L is set；
A, its error in classification ε is calculated for each SVMsm,n,
Wherein U (x) is a function, is otherwise -1 when x ＞ 0 are 1.
B, selection hl(x)=argmin εmn；
C, the weight for calculating selection SVM classifier
If D, βlThe then iteration ends of ＜ 0；Otherwise decision function F (x) ← β is updatedlhl, the grader of selection is added Into decision function；
E, useThe weight of training sample is updated, by the sample of classification error in next iteration Assign bigger weighted value.
(5) grader isExport the strong classifier being made up of multiple monokaryon SVM classifiers.
5th step, after the sample formation block level feature of test set, in the multinuclear taxonomy model after the completion of input training, enters The automatic identification of row insect image, identifies the classification of insect image to be measured.
General principle, principal character and the advantages of the present invention of the present invention has been shown and described above.The technology of the industry Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and that described in above-described embodiment and specification is the present invention Principle, various changes and modifications of the present invention are possible without departing from the spirit and scope of the present invention, these change and Improvement is both fallen within the range of claimed invention.The protection domain of application claims by appended claims and its Equivalent is defined.

Claims (4)

1. a kind of insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study, it is characterised in that Comprise the following steps：
11) nature image set is obtained, nature image set is obtained and is divided into natural image block, use unsupervised feature learning method The training study dictionary from unlabelled natural image block；
12) all insect images are shot by the collection of insect image, mark and pretreatment under conditions of maintaining uniform illumination, and Shooting is focused in insect symptom, by all insect image normalizations and size is zoomed to for 200 × 200 pixels, if obtaining Dry sample；The sample of every kind of insect is randomly divided into training set and test set；
13) using the sample formation block level feature of training set, marked training image is extracted into image block from different scale, Training image blocks collection under certain yardstick is carried out after rarefaction representation by learning dictionary, then formation is operated by spatial alignment pondization Block level feature；
14) using the SVM training multinuclear taxonomy models of multinuclear Boosting study, the block level feature under different scale is passed through into base Multinuclear taxonomy model is trained in the multinuclear Boosting SVMs learnt；
15) by after the sample formation block level feature of test set, in the multinuclear taxonomy model after the completion of input training, insect figure is carried out The automatic identification of picture.
2. a kind of insect image knowledge being combined based on depth characteristic with multinuclear Boosting study according to claim 1 Other method, it is characterised in that described acquisition natural image comprises the following steps：
21) nature image set Y, Y=[y are obtained1,y2,...yi...], yiRepresent i-th natural image；Using uniform grid from Natural image is concentrated and extracts N number of natural image block x altogetheri, xi∈Rn, (i=1,2 ..., N), each of which natural image carries K blocks are taken, and to natural image block xiDo normalized；
22) application singular value decomposition algorithm training study dictionary, solves following optimization problem and obtains study dictionary：
D=[d1,d2,...,dM]∈Rn×MStudy dictionary is represented,It is natural image block xi Corresponding sparse coding,It is natural image block xiRarefaction representation vector.
3. a kind of insect image knowledge being combined based on depth characteristic with multinuclear Boosting study according to claim 1 Other method, it is characterised in that the sample formation block level feature of described utilization training set comprises the following steps：
31) marked training image is extracted into image block from different scale, from coarseness level, middle granular level and particulate Spend on horizontal different scale and image block is extracted to training image respectively；
For a training image yi, K insect image block [x is obtained using the uniform grid segmentation of some scale size1, x2,...,xK]；
32) training image blocks under certain yardstick are subjected to rarefaction representation by learning dictionary；
Image block sparse coding is comprised the following steps using dictionary：
321) for given insect image yi, overlapping square [x is segmented the image into using uniform grid1,x2,...,xK]∈Rn ×K；
322) obtained study dictionary D, K insect image block [x is learnt using singular value decomposition algorithm1,x2,...,xK] correspondence Sparse coding be respectivelyFind each piece of xiSparse codeFollowing optimization problem is solved,
Wherein | | | |0It isThe number of middle nonzero element；
Take l1Norm minimum is substituted, and solves following optimization problem：
WhereinSolution be corresponding topography's block vector xiCorresponding characteristic vector,Corresponding to one The sparse coding of all topography's blocks in insect image；
33) block level feature is formed by the operation of spatial alignment pond, it comprises the following steps：
331) i-th piece of sparse codingJ sections are divided into,WhereinTable Show coefficientJth section；
332) thenIt is weighted, its formula is as follows：
Wherein, vector viCorresponding to i-th of localized mass；
The corresponding all localized masses of one training image distinguish corresponding characteristic vector viForm matrix V；
333) pondization operation is done to matrix V, takes the diagonal element of matrix V to produce final block level feature, its formula is as follows：
S=diag (V),
Wherein：S represents the characteristic vector extracted from an a certain yardstick of training image.
4. a kind of insect image knowledge being combined based on depth characteristic with multinuclear Boosting study according to claim 1 Other method, it is characterised in that the SVM training multinuclear taxonomy models of described utilization multinuclear Boosting study include following step Suddenly：
41) training set is inputted in SVMFeature set { the f of different scale1,f2,...fN, nuclear matrix { K1,K2, ...KM, initialization strong classifier F (x)=0,
Wherein, xiRepresent training sample, yiRepresent that mark, the D of training sample represent number of training；
42) each is trained to include f for each n ∈ N, m ∈ MNAnd KMCorresponding single SVMmnParameter amnAnd bmn,
hmn=Kmn(x,xi)Tamn+bmn；
43) initialization sample weightBy training sample assignment identical initial weight；In each iteration, select Select the minimum SVM classifier of error in classification in data set；
44) iterations l=1 to L is set；
441) its error in classification ε is calculated for each SVMsm,n,
Wherein U (x) is a function, is otherwise -1 when x ＞ 0 are 1；
442) h is selectedl(x)=argmin εmn；
443) weight of selection SVM classifier is calculated
If 444) βlThe then iteration ends of ＜ 0；Otherwise decision function F (x) ← β is updatedlhl, the grader of selection is added to In decision function；
445) useThe weight of training sample is updated, the sample of classification error is assigned in next iteration Bigger weighted value；
45) grader isExport the strong classifier being made up of multiple monokaryon SVM classifiers.
CN201710105763.6A 2017-02-24 2017-02-24 Learn the pest image-recognizing method combined with multicore Boosting based on depth characteristic Active CN107025431B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710105763.6A CN107025431B (en) 2017-02-24 2017-02-24 Learn the pest image-recognizing method combined with multicore Boosting based on depth characteristic

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710105763.6A CN107025431B (en) 2017-02-24 2017-02-24 Learn the pest image-recognizing method combined with multicore Boosting based on depth characteristic

Publications (2)

Publication Number Publication Date
CN107025431A true CN107025431A (en) 2017-08-08
CN107025431B CN107025431B (en) 2019-06-18

Family

ID=59525886

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710105763.6A Active CN107025431B (en) 2017-02-24 2017-02-24 Learn the pest image-recognizing method combined with multicore Boosting based on depth characteristic

Country Status (1)

Country Link
CN (1) CN107025431B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104077612A (en) * 2014-07-15 2014-10-01 中国科学院合肥物质科学研究院 Pest image recognition method based on multi-feature sparse representation technology
CN104102922A (en) * 2014-07-15 2014-10-15 中国科学院合肥物质科学研究院 Pest image classification method based on context sensing dictionary learning
CN105488536A (en) * 2015-12-10 2016-04-13 中国科学院合肥物质科学研究院 Agricultural pest image recognition method based on multi-feature deep learning technology
CN105574540A (en) * 2015-12-10 2016-05-11 中国科学院合肥物质科学研究院 Method for learning and automatically classifying pest image features based on unsupervised learning technology

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104077612A (en) * 2014-07-15 2014-10-01 中国科学院合肥物质科学研究院 Pest image recognition method based on multi-feature sparse representation technology
CN104102922A (en) * 2014-07-15 2014-10-15 中国科学院合肥物质科学研究院 Pest image classification method based on context sensing dictionary learning
CN105488536A (en) * 2015-12-10 2016-04-13 中国科学院合肥物质科学研究院 Agricultural pest image recognition method based on multi-feature deep learning technology
CN105574540A (en) * 2015-12-10 2016-05-11 中国科学院合肥物质科学研究院 Method for learning and automatically classifying pest image features based on unsupervised learning technology

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
胡庆辉等: ""基于Boosting框架的非稀疏多核学习方法"", 《计算机应用研究》 *

Also Published As

Publication number Publication date
CN107025431B (en) 2019-06-18

Similar Documents

Publication Publication Date Title
CN105488536B (en) A kind of agricultural pests image-recognizing method based on multiple features depth learning technology
CN105184312B (en) A kind of character detecting method and device based on deep learning
CN102609681B (en) Face recognition method based on dictionary learning models
CN110348399B (en) Hyperspectral intelligent classification method based on prototype learning mechanism and multidimensional residual error network
CN107194418B (en) Rice aphid detection method based on antagonistic characteristic learning
CN108984745A (en) A kind of neural network file classification method merging more knowledge mappings
CN107506740A (en) A kind of Human bodys' response method based on Three dimensional convolution neutral net and transfer learning model
CN107341499A (en) It is a kind of based on non-formaldehyde finishing and ELM fabric defect detection and sorting technique
CN105069481B (en) Natural scene multiple labeling sorting technique based on spatial pyramid sparse coding
CN110232341A (en) Based on convolution-stacking noise reduction codes network semi-supervised learning image-recognizing method
CN105224942A (en) A kind of RGB-D image classification method and system
CN106845528A (en) A kind of image classification algorithms based on K means Yu deep learning
CN109522924A (en) A kind of broad-leaf forest wood recognition method based on single photo
CN110197205A (en) A kind of image-recognizing method of multiple features source residual error network
Ma et al. Linearization to nonlinear learning for visual tracking
CN110490227A (en) A kind of few sample image classification method based on Feature Conversion
CN103793694B (en) Human face recognition method based on multiple-feature space sparse classifiers
CN105787517B (en) Classification of Polarimetric SAR Image method based on the sparse self-encoding encoder of small echo
CN106650667A (en) Pedestrian detection method and system based on support vector machine
CN109902662A (en) A kind of pedestrian recognition methods, system, device and storage medium again
CN105976397B (en) A kind of method for tracking target
CN109614866A (en) Method for detecting human face based on cascade deep convolutional neural networks
CN104063713A (en) Semi-autonomous on-line studying method based on random fern classifier
CN109344898A (en) Convolutional neural networks image classification method based on sparse coding pre-training
CN105574540A (en) Method for learning and automatically classifying pest image features based on unsupervised learning technology

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