CN105574540A - Method for learning and automatically classifying pest image features based on unsupervised learning technology - Google Patents
Method for learning and automatically classifying pest image features based on unsupervised learning technology Download PDFInfo
- Publication number
- CN105574540A CN105574540A CN201510923441.3A CN201510923441A CN105574540A CN 105574540 A CN105574540 A CN 105574540A CN 201510923441 A CN201510923441 A CN 201510923441A CN 105574540 A CN105574540 A CN 105574540A
- Authority
- CN
- China
- Prior art keywords
- image
- dictionary
- insect
- block
- 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
Links
Classifications
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
-
- 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
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method for learning and automatically classifying pest image features based on an unsupervised learning technology. Compared with the prior art, the defect of poor pest recognition capability caused by covering and pollution on a pest image is solved. The method comprises the steps of randomly sampling pest image blocks in a large scale; learning the unsupervised feature dictionary of the pest images; coding the pest image features and performing pooling operation on the features; and recognizing a multi-class classifier. According to the method, the pest recognition accuracy is improved, the unsupervised dictionary is built by the unsupervised dictionary training mode, the features are subjected to the pooling operation through combining with the sparse coding mode, the relatively strong capability to discriminate the features is achieved, and the pest image is effectively expressed.
Description
Technical field
The present invention relates to image identification technical field, specifically a kind of study of the insect characteristics of image based on unsupervised learning technology and automatic classification method.
Background technology
Along with the rise of precision agriculture, computer image processing technology is that agricultural production provides new approaches, for the automatic identification realizing insect image provides advanced technological means.Because it has the advantages such as accuracy is high, speed fast, contain much information, in crop pests identification, obtain more application, crop pests can be identified efficiently, scientific management crop growth, improve Crop yield and quality.But actual farm environment is complicated, pest species is various, and captured insect image easily produces and blocks or pollute, and has certain restriction to raising discrimination.The method of rarefaction representation all has robustness to block and pixel pollution etc., and the prerequisite of rarefaction representation trains the complete dictionary preserving characteristic atomic preferably, therefore how more preferably to carry out feature learning and carry out effective expression to feature becoming problem demanding prompt solution.
Summary of the invention
The object of the invention is, in order to solve in prior art insect image because blocking and polluting the defect causing insect recognition capability difference, to provide a kind of insect characteristics of image based on unsupervised learning technology to learn to solve the problems referred to above with automatic classification method.
To achieve these goals, technical scheme of the present invention is as follows:
Based on the study of insect characteristics of image and the automatic classification method of unsupervised learning technology, comprise the following steps:
Carry out extensive insect image block stochastic sampling, random sampling is carried out to image, this sampling process is performed in all training sample image, carry out large-scale image block collection;
The non-supervisory characteristics dictionary study of insect image, uses unsupervised learning method construction feature dictionary D=[d
1, d
2..., d
m] ∈ R
n × M, wherein M represents the size of dictionary, and each arranges d
jrepresent the atom of dictionary;
Insect characteristics of image is encoded and is carried out feature pool operation, utilizes non-supervisory dictionary D
t, by insect image block vector y
icoding becomes proper vector x
i; Feature is converted into and preserves important information and abandon the feature of irrelevant information;
Multi classifier identification, carries out training study by multi classifier to positive and negative feature, realizes the judgement of training sample insect generic.
Described extensive insect image block stochastic sampling of carrying out comprises the following steps:
Gather training insect image/video frame, division block is carried out to the insect image of every frame, a block carries out random sampling wherein, to sample in block the image block of r × r, wherein n=r × r, n are the dimension of non-supervisory dictionary, produce the center point coordinate (x of r × r image block within the scope of block matrices with the form of random number, y), to sample within the scope of block the little image block that the length of side is r with this coordinate;
To onblock executing random sampling procedure each in frame, and each block is defined as the pond scope defined when sparse features carries out pondization operation;
The all frames of training insect image/video are all carried out to division block and carry out random sampling procedure, obtains the training sample feature of original dictionary, composition training sample eigenmatrix Y.
The non-supervisory characteristics dictionary study of described insect image comprises the following steps:
The random dictionary D of initialization, arranges degree of rarefication k,
to primitive character normalization operation, initialization dictionary is expressed as D=[d
1, d
2..., d
m] ∈ R
n × M;
Utilize orthogonal matching pursuit method, fixing random dictionary D carries out rarefaction representation to input data Y and obtains sparse vector matrix X,
Wherein inputting data Y is the training sample eigenmatrix extracted, and X represents sparse vector matrix, passes through || x
i||
0≤ k carries out coefficient restriction;
Upgrade random dictionary D by column, its formula is as follows:
Wherein d
krepresent kth row in D,
represent a kth row vector,
represent the contribute matrix of dictionary;
Whether error in judgement meets accuracy requirement or whether reaches the iterations of specifying, satisfied then terminate training, produce non-supervisory dictionary D
t, do not meet and then continue compute sparse vector matrix X and upgrade random dictionary D.
Described insect characteristics of image is encoded and is carried out feature poolization operation and comprises the following steps:
To insect characteristics of image Y and non-supervisory dictionary D
t, use orthogonal matching pursuit method to obtain coding characteristic
in at most k item be nonzero term, its formula is as follows:
Wherein
x
irepresent i-th sparse vector;
Perform the operation of maximum pond to each block p, wherein p ∈ P, P are total block counts, and each block p obtains and converges feature f=[f
1..., f
j..., f
m] ∈ R
m;
Perform in each block p
wherein
represent
in the jth of all sparse features of p block tie up all data, f
jfor the maximal value of jth dimension data.
Described multi classifier is identified as and convergence feature f is added support vector machine carries out learning training, obtains feature templates, carries out in the following ways;
For two class situations, kth class and m class are classified, and by solving optimization problem, its formula is as follows:
Wherein C is called punishment parameter,
represent the slack variable of non-negative,
a tolerance of training mistake, ω
kmfor lineoid parameter, then the decision function between kth class and m class as shown in the formula:
Wherein x
isupport vector,
for the Lagrange multiplier of correspondence, b
kmfor classification threshold values;
For multiclass situation, two class 1-V-1 situations are applied to multiclass, if there is s class, adopt directed acyclic drawing method, then total (s-1) × s/2 sorter is classified respectively, then takes the mode of competing to predict classification.
Also comprise and classifying to test sample book, it comprises the following steps:
Obtain test insect image/video, obtain test sample image frame;
To test pattern picture frame according to non-supervisory characteristics dictionary, obtain corresponding sparse coding and encode as characteristics of image;
The convergence of maximum pond is carried out to sparse features, reduces feature quantity, obtain the feature having more information;
In conjunction with the feature templates that multi classifier learns, the convergence feature of test sample book is predicted, realize classification.
Beneficial effect
A kind of study of the insect characteristics of image based on unsupervised learning technology of the present invention and automatic classification method, compared with prior art improve the accuracy rate of insect identification, non-supervisory dictionary training patterns is utilized to construct non-supervisory dictionary, in conjunction with sparse coding mode and to feature carry out pondization operation, achieve the resolving ability that feature is stronger, effectively represent insect image.
Accompanying drawing explanation
Fig. 1 is method flow diagram of the present invention;
Fig. 2 is the discrimination correlation curve figure of the present invention and prior art.
Embodiment
For making to have a better understanding and awareness architectural feature of the present invention and effect of reaching, coordinating detailed description in order to preferred embodiment and accompanying drawing, being described as follows:
As shown in Figure 1, a kind of study of the insect characteristics of image based on unsupervised learning technology of the present invention and automatic classification method, adopt generally first to train and know method for distinguishing afterwards, first in laboratory, recognition training is carried out to insect image, then carry out actual classification in practical application.It comprises the following steps:
The first step, carries out extensive insect image block stochastic sampling, can learn the dictionary that comprises insect characteristics of image.Carry out random sampling to image, this sampling process performed in all training sample image, carry out large-scale image block collection, it comprises the following steps:
(1) gather training insect image/video frame, this insect image/video frame is training sample.Carry out division block to the insect image of every frame, a two field picture is divided into several blocks, a block carries out random sampling wherein.The process of random sampling is the image block of r × r of sampling in block, and the image block of r × r is little image block, wherein n=r × r as far as possible, and n is the dimension of the non-supervisory dictionary in subsequent step.Produce the center point coordinate (x of r × r image block with the form of random number by the method for prior art within the scope of block matrices, y), to sample within the scope of block the little image block that the length of side is r with this coordinate, best mode is extracted the central spot of the image block of r × r at block, be convenient to like this calculate center point coordinate (x, y), be also convenient to calculate sampling length of side r.
(2) to onblock executing random sampling procedure each in frame, the sampling process of previous step is namely carried out.And each block is defined in the pond scope defined when pondization operation being carried out to all sparse features in the operating process of follow-up pond.
(3) all frames of training insect image/video all carried out to division block and carry out random sampling procedure, performing the operation of above two steps, to carry out large-scale image block collection.In actual applications, can extract 120,000 image blocks, extract 60 class pest images altogether from Sample Storehouse, every class 2000 image blocks, obtain the training sample feature of original dictionary with this, composition training sample eigenmatrix Y.
Second step, the non-supervisory characteristics dictionary study of insect image, constructs the complete dictionary that comprises multiclass insect characteristics of image atom.Use unsupervised learning method construction feature dictionary D=[d
1, d
2..., d
m] ∈ R
n × M, wherein M represents the size of dictionary, and each arranges d
jrepresent the atom of dictionary.And adopt K-SVD algorithm to carry out the training process of dictionary, K-SVD algorithm reduces the energy of residual error by continuous iteration and learns regeneration characteristics dictionary D, not only achieve reconstructed error minimum and meet sparse restriction, feature can have stronger rarefaction representation under characteristics dictionary D.It comprises the following steps:
(1) the random dictionary D of initialization, arranges degree of rarefication k,
to primitive character normalization operation, initialization dictionary is expressed as D=[d
1, d
2..., d
m] ∈ R
n × M.
(2) utility preferably orthogonal matching pursuit method (OMP), fixing random dictionary D carries out rarefaction representation to input data Y and obtains sparse vector matrix X,
Wherein inputting data Y is the training sample eigenmatrix extracted, and X represents sparse vector matrix, passes through || x
i||
0≤ k carries out coefficient restriction.
(3) upgrade random dictionary D by column, its formula is as follows:
Wherein d
krepresent kth row in D,
represent a kth row vector,
represent the contribute matrix of dictionary.At this, make d
kwith
zero setting, can remove this contribution.Minimize
only need E
kcarry out SVD decomposition, E
k=UWV
t, the often row of U and V are orthogonal basis, and W is singular value matrix, and main shaft distributes E
kenergy, descending arrangement, W
00for ceiling capacity.U
0as new dictionary row,
as the new row vector of X.It should be noted that as ensureing dictionary convergence, calculating E
kin time, needs to use
row corresponding to middle nonzero term.
(4) whether error in judgement meets accuracy requirement or whether reaches the iterations of specifying, satisfied then terminate training, produce non-supervisory dictionary D
t, do not meet and then continue compute sparse vector matrix X and upgrade random dictionary D.
3rd step, insect characteristics of image is encoded and is carried out feature pool operation.Utilize non-supervisory dictionary D
t, by insect image block vector y
icoding becomes proper vector x
i, this cataloged procedure is applied in the orthogonal matching pursuit algorithm used in dictionary building process, and this algorithm can ensure that the projection of atom is all orthogonal, and the result of iteration is all optimum and fast convergence rate.Pondization operation can ensure that the feature after converging saves preoperative information and eliminates incoherent information, feature is converted into and preserves important information and abandon the feature of irrelevant information.It comprises the following steps:
(1) to insect characteristics of image Y and non-supervisory dictionary D
t, use orthogonal matching pursuit method to obtain coding characteristic
in at most k item be nonzero term, its formula is as follows:
Wherein
x
irepresent i-th sparse vector, k is sparse restriction, y
irepresent uncoded insect feature,
represent the sparse features after sparse transferring frame coding.This step obtains i point patterns by separating optimization problem
ensure in match tracing process residual error with before each vector all orthogonal, therefore, this coding is effective.
(2) perform the operation of maximum pond to each block p, converge acquisition converge feature by this maximum pond, convergence feature has compared to sparse coding feature preserves principal character and maintains the constant feature with reducing feature quantity of dimension.Wherein p ∈ P, P are total block counts, and each block p obtains and converges feature f=[f
1..., f
j..., f
m] ∈ R
m.
Perform in each block p
feature is converged in acquisition, therefore converges feature and has preservation principal character compared to sparse coding feature and maintain the constant feature with reducing feature quantity of dimension, wherein
represent
in the jth of all sparse features of p block tie up all data, i.e. matrix
in the jth row data of p block, f
jfor the maximal value of jth dimension data.
4th step, multi classifier identification.By method of the prior art, by multi classifier, training study is carried out to positive and negative feature, realize the judgement of training sample insect generic.By support vector machine as sorter, structure based risk minimization criterion learns feature, carry out training the template obtaining and preserve tagsort to training insect characteristics of image, select different kernel functions according to actual conditions, comprise the kernel functions such as linear kernel, polynomial kernel, radial basis core.Its multi classifier is identified as and convergence feature f is added support vector machine carries out learning training, obtains feature templates, carries out in the following ways:
(1) for two class situations, kth class and m class are classified, and by solving optimization problem, its formula is as follows:
Wherein C is called punishment parameter,
represent the slack variable of non-negative,
a tolerance of training mistake, ω
kmfor lineoid parameter, then the decision function between kth class and m class as shown in the formula:
Wherein x
isupport vector,
for the Lagrange multiplier of correspondence, b
kmfor classification threshold values.
(2) for multiclass situation, two class 1-V-1 situations are applied to multiclass, if there is s class, adopt directed acyclic drawing method, then total (s-1) × s/2 sorter is classified respectively, then takes the mode of competing to predict classification.
Above process completes training study to image pattern and assorting process, completes the training to multi classifier, to realize direct use in actual applications.When reality uses, then directly to classify to test sample book, carry out the above-described first step equally to FOUR EASY STEPS, just directly apply in the 4th step multi classifier, it comprises the following steps:
A, acquisition test insect image/video, obtain test sample image frame.
B, to test pattern picture frame according to non-supervisory characteristics dictionary, obtain corresponding sparse coding and encode as characteristics of image.
C, the convergence of maximum pond is carried out to sparse features, reduce feature quantity, obtain the feature having more information;
D, the feature templates learnt in conjunction with multi classifier, predict the convergence feature of test sample book, realizes classification.
Carry out the inventive method validation verification for common 15 kinds of crop pests, take to increase insect image number gradually to learn non-supervisory dictionary, improve the characteristic differentiation of each dictionary row atom in non-supervisory dictionary gradually.Showing the inventive method in Fig. 2 is applied in insect image recognition when directly adopting disaggregated model (feature is without non-supervisory dictionary training) to carry out discriminator with tradition, the contrast situation of discrimination.As can be seen from Figure 2, the discrimination situation of the inventive method is always higher than other three kinds of classic methods, when training sample number increases, the advantage of non-supervisory dictionary training is more obvious, about 8.8% is exceeded than neural network model, exceed about 13.67% than support vector machine, more exceeded about 17.87% than k nearest neighbor.
Contrast (%) with average recognition rate of the present invention under the different disaggregated model of table 1
Be at training sample in table 1
less (20-50) and
each model average recognition rate situation time more (120-150), as can be seen from table also, when training sample is less, the inventive method just has higher recognition performance, when training sample is more, recognition performance is more obvious, and therefore the inventive method has higher recognition performance compared with the sorting technique of existing insect.
More than show and describe ultimate principle of the present invention, principal character and advantage of the present invention.The technician of the industry should understand; the present invention is not restricted to the described embodiments; the just principle of the present invention described in above-described embodiment and instructions; the present invention also has various changes and modifications without departing from the spirit and scope of the present invention, and these changes and improvements all fall in claimed scope of the present invention.The protection domain of application claims is defined by appending claims and equivalent thereof.
Claims (6)
1., based on the study of insect characteristics of image and the automatic classification method of unsupervised learning technology, it is characterized in that, comprise the following steps:
11) carry out extensive insect image block stochastic sampling, random sampling is carried out to image, this sampling process is performed in all training sample image, carry out large-scale image block collection;
12) the non-supervisory characteristics dictionary study of insect image, uses unsupervised learning method construction feature dictionary D=[d
1, d
2..., d
m] ∈ R
n × M, wherein M represents the size of dictionary, and each arranges d
jrepresent the atom of dictionary;
13) insect characteristics of image is encoded and is carried out feature pool operation, utilizes non-supervisory dictionary D
t, by insect image block vector y
icoding becomes proper vector x
i; Feature is converted into and preserves important information and abandon the feature of irrelevant information;
14) multi classifier identification, carries out training study by multi classifier to positive and negative feature, realizes the judgement of training sample insect generic.
2. a kind of study of the insect characteristics of image based on unsupervised learning technology according to claim 1 and automatic classification method, it is characterized in that, described extensive insect image block stochastic sampling of carrying out comprises the following steps:
21) training insect image/video frame is gathered, division block is carried out to the insect image of every frame, a block carries out random sampling wherein, to sample in block the image block of r × r, wherein n=r × r, n are the dimension of non-supervisory dictionary, produce the center point coordinate (x of r × r image block within the scope of block matrices with the form of random number, y), to sample within the scope of block the little image block that the length of side is r with this coordinate;
22) to onblock executing random sampling procedure each in frame, and each block is defined as the pond scope defined when sparse features carries out pondization operation;
23) all frames of training insect image/video all carried out to division block and carry out random sampling procedure, obtaining the training sample feature of original dictionary, composition training sample eigenmatrix Y.
3. a kind of study of the insect characteristics of image based on unsupervised learning technology according to claim 1 and automatic classification method, is characterized in that, the non-supervisory characteristics dictionary study of described insect image comprises the following steps:
31) the random dictionary D of initialization, arranges degree of rarefication k,
to primitive character normalization operation, initialization dictionary is expressed as D=[d
1, d
2..., d
m] ∈ R
n × M;
32) utilize orthogonal matching pursuit method, fixing random dictionary D carries out rarefaction representation to input data Y and obtains sparse vector matrix X,
Wherein inputting data Y is the training sample eigenmatrix extracted, and X represents sparse vector matrix, passes through || x
i||
0≤ k carries out coefficient restriction;
33) upgrade random dictionary D by column, its formula is as follows:
Wherein d
krepresent kth row in D,
represent a kth row vector,
represent the contribute matrix of dictionary;
34) whether error in judgement meets accuracy requirement or whether reaches the iterations of specifying, satisfied then terminate training, produce non-supervisory dictionary D
t, do not meet and then continue compute sparse vector matrix X and upgrade random dictionary D.
4. a kind of study of the insect characteristics of image based on unsupervised learning technology according to claim 1 and automatic classification method, is characterized in that, described insect characteristics of image is encoded and carried out feature poolization operation and comprises the following steps:
41) to insect characteristics of image Y and non-supervisory dictionary D
t, use orthogonal matching pursuit method to obtain coding characteristic
in at most k item be nonzero term, its formula is as follows:
Wherein
x
irepresent i-th sparse vector;
42) perform the operation of maximum pond to each block p, wherein p ∈ P, P are total block counts, and each block p obtains and converges feature f=[f
1..., f
j..., f
m] ∈ R
m;
Perform in each block p
wherein
represent
in the jth of all sparse features of p block tie up all data, f
jfor the maximal value of jth dimension data.
5. a kind of study of the insect characteristics of image based on unsupervised learning technology according to claim 1 and automatic classification method, it is characterized in that, described multi classifier is identified as and convergence feature f is added support vector machine carries out learning training, obtains feature templates, carries out in the following ways;
51) for two class situations, kth class and m class are classified, and by solving optimization problem, its formula is as follows:
Wherein C is called punishment parameter,
represent the slack variable of non-negative,
a tolerance of training mistake, ω
kmfor lineoid parameter, then the decision function between kth class and m class as shown in the formula:
Wherein x
isupport vector,
for the Lagrange multiplier of correspondence, b
kmfor classification threshold values;
52) for multiclass situation, two class 1-V-1 situations are applied to multiclass, if there is s class, adopt directed acyclic drawing method, then total (s-1) × s/2 sorter is classified respectively, then takes the mode of competing to predict classification.
6. a kind of study of the insect characteristics of image based on unsupervised learning technology according to claim 1 and automatic classification method, it is characterized in that, also comprise and classifying to test sample book, it comprises the following steps:
61) obtain test insect image/video, obtain test sample image frame;
62) to test pattern picture frame according to non-supervisory characteristics dictionary, obtain corresponding sparse coding and encode as characteristics of image;
63) convergence of maximum pond is carried out to sparse features, reduce feature quantity, obtain the feature having more information;
64) in conjunction with the feature templates that multi classifier learns, the convergence feature of test sample book is predicted, realize classification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510923441.3A CN105574540B (en) | 2015-12-10 | 2015-12-10 | A kind of study of pest characteristics of image and automatic classification method based on unsupervised learning technology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510923441.3A CN105574540B (en) | 2015-12-10 | 2015-12-10 | A kind of study of pest characteristics of image and automatic classification method based on unsupervised learning technology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105574540A true CN105574540A (en) | 2016-05-11 |
CN105574540B CN105574540B (en) | 2019-03-15 |
Family
ID=55884646
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510923441.3A Active CN105574540B (en) | 2015-12-10 | 2015-12-10 | A kind of study of pest characteristics of image and automatic classification method based on unsupervised learning technology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105574540B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106250856A (en) * | 2016-08-03 | 2016-12-21 | 武汉大学 | A kind of high-definition picture scene classification method based on non-supervisory feature learning |
CN106407996A (en) * | 2016-06-30 | 2017-02-15 | 华南理工大学 | Machine learning based detection method and detection system for the fall of the old |
CN106446915A (en) * | 2016-09-30 | 2017-02-22 | 广东石油化工学院 | System for quickly identifying fruit-piercing moths based on transcendental analysis |
CN106570509A (en) * | 2016-11-04 | 2017-04-19 | 天津大学 | Dictionary learning and coding method for extracting digital image feature |
CN106682681A (en) * | 2016-08-19 | 2017-05-17 | 江苏电力信息技术有限公司 | Recognition algorithm automatic improvement method based on relevance feedback |
CN107025431A (en) * | 2017-02-24 | 2017-08-08 | 中国科学院合肥物质科学研究院 | The insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study |
CN109376802A (en) * | 2018-12-12 | 2019-02-22 | 浙江工业大学 | A kind of gastroscope organ classes method dictionary-based learning |
CN117036832A (en) * | 2023-10-09 | 2023-11-10 | 之江实验室 | Image classification method, device and medium based on random multi-scale blocking |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102819748A (en) * | 2012-07-19 | 2012-12-12 | 河南工业大学 | Classification and identification method and classification and identification device of sparse representations of destructive insects |
US20130208977A1 (en) * | 2011-11-02 | 2013-08-15 | Nec Laboratories America, Inc. | Receptive field learning for pooled image features |
-
2015
- 2015-12-10 CN CN201510923441.3A patent/CN105574540B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130208977A1 (en) * | 2011-11-02 | 2013-08-15 | Nec Laboratories America, Inc. | Receptive field learning for pooled image features |
CN102819748A (en) * | 2012-07-19 | 2012-12-12 | 河南工业大学 | Classification and identification method and classification and identification device of sparse representations of destructive insects |
Non-Patent Citations (1)
Title |
---|
乃科: "基于块结构化字典学习的稀疏表示图像识别", 《中南大学2014硕士学位论文》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106407996A (en) * | 2016-06-30 | 2017-02-15 | 华南理工大学 | Machine learning based detection method and detection system for the fall of the old |
CN106250856A (en) * | 2016-08-03 | 2016-12-21 | 武汉大学 | A kind of high-definition picture scene classification method based on non-supervisory feature learning |
CN106682681A (en) * | 2016-08-19 | 2017-05-17 | 江苏电力信息技术有限公司 | Recognition algorithm automatic improvement method based on relevance feedback |
CN106446915A (en) * | 2016-09-30 | 2017-02-22 | 广东石油化工学院 | System for quickly identifying fruit-piercing moths based on transcendental analysis |
CN106570509A (en) * | 2016-11-04 | 2017-04-19 | 天津大学 | Dictionary learning and coding method for extracting digital image feature |
CN106570509B (en) * | 2016-11-04 | 2019-09-27 | 天津大学 | A kind of dictionary learning and coding method for extracting digital picture feature |
CN107025431A (en) * | 2017-02-24 | 2017-08-08 | 中国科学院合肥物质科学研究院 | The insect image-recognizing method being combined based on depth characteristic with multinuclear Boosting study |
CN107025431B (en) * | 2017-02-24 | 2019-06-18 | 中国科学院合肥物质科学研究院 | Learn the pest image-recognizing method combined with multicore Boosting based on depth characteristic |
CN109376802A (en) * | 2018-12-12 | 2019-02-22 | 浙江工业大学 | A kind of gastroscope organ classes method dictionary-based learning |
CN109376802B (en) * | 2018-12-12 | 2021-08-03 | 浙江工业大学 | Gastroscope organ classification method based on dictionary learning |
CN117036832A (en) * | 2023-10-09 | 2023-11-10 | 之江实验室 | Image classification method, device and medium based on random multi-scale blocking |
CN117036832B (en) * | 2023-10-09 | 2024-01-05 | 之江实验室 | Image classification method, device and medium based on random multi-scale blocking |
Also Published As
Publication number | Publication date |
---|---|
CN105574540B (en) | 2019-03-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105574540A (en) | Method for learning and automatically classifying pest image features based on unsupervised learning technology | |
CN109272500B (en) | Fabric classification method based on adaptive convolutional neural network | |
CN103699523A (en) | Product classification method and device | |
CN105740758A (en) | Internet video face recognition method based on deep learning | |
CN105469047A (en) | Chinese detection method based on unsupervised learning and deep learning network and system thereof | |
CN104778481A (en) | Method and device for creating sample library for large-scale face mode analysis | |
CN107145889A (en) | Target identification method based on double CNN networks with RoI ponds | |
CN109446985B (en) | Multi-angle plant identification method based on vector neural network | |
CN103366180A (en) | Cell image segmentation method based on automatic feature learning | |
CN108764242A (en) | Off-line Chinese Character discrimination body recognition methods based on deep layer convolutional neural networks | |
CN108021947A (en) | A kind of layering extreme learning machine target identification method of view-based access control model | |
CN101599126A (en) | Utilize the support vector machine classifier of overall intercommunication weighting | |
CN106909946A (en) | A kind of picking system of multi-modal fusion | |
CN104239902A (en) | Hyper-spectral image classification method based on non-local similarity and sparse coding | |
CN111738367B (en) | Part classification method based on image recognition | |
CN101833667A (en) | Pattern recognition classification method expressed based on grouping sparsity | |
CN101968852A (en) | Entropy sequencing-based semi-supervision spectral clustering method for determining clustering number | |
CN105046714A (en) | Unsupervised image segmentation method based on super pixels and target discovering mechanism | |
CN109033169A (en) | Mobile traffic classification method based on multistage weight conversion and convolutional neural networks | |
CN107679550A (en) | A kind of appraisal procedure of data set classification availability | |
CN106529397A (en) | Facial feature point positioning method and system in unconstrained environment | |
CN105550712A (en) | Optimized convolution automatic encoding network-based auroral image sorting method | |
CN105205460A (en) | Face expression feature extraction and recognition method based on maximum direction encoding | |
CN104537381A (en) | Blurred image identification method based on blurred invariant feature | |
CN114743133A (en) | Lightweight small sample video classification and identification method and system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |