CN107766890A - The improved method that identification segment learns in a kind of fine granularity identification - Google Patents
The improved method that identification segment learns in a kind of fine granularity identification Download PDFInfo
- Publication number
- CN107766890A CN107766890A CN201711040828.XA CN201711040828A CN107766890A CN 107766890 A CN107766890 A CN 107766890A CN 201711040828 A CN201711040828 A CN 201711040828A CN 107766890 A CN107766890 A CN 107766890A
- Authority
- CN
- China
- Prior art keywords
- identification
- segment
- detector
- notable
- vector
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The improved method that identification segment learns in a kind of fine granularity identification：Extracting has the segment for differentiating property in original image, including：Original image obtains a characteristic pattern by the convolution pond layer in convolutional neural networks, and the vector of each space fixed position in characteristic pattern is considered as into the detector corresponding to relevant position segment in beginning image；Assuming that learnt in original image an identification region there is the detector that highest responds, detector and characteristic pattern are subjected to convolution algorithm, obtain new response diagram；The position with maximum is selected in new response diagram, is obtained with the segment for differentiating property；Feature of the study with the segment for differentiating property simultaneously is used to classify, including：Local notable figure is obtained according to the segment for differentiating property；Local notable figure is encoded using spatial weighting Fei Sheer vectors.The identification feature for being more suitable for fine granularity identification mission is arrived in present invention study, and the interference of background information is to improve nicety of grading in reduction identification segment.
Description
Technical field
Identification segment learns in being identified the present invention relates to a kind of fine granularity.More particularly to one kind according to response diagram to figure
The improvement that identification segment learns in the fine granularity identification of spatial weighting Fei Sheer vectors is obtained as describer carries out spatial weighting
Method.
Background technology
In recent years, fine granularity identification has attracted increasing concern in field of target recognition, and it is to a certain big classification
Target subclass is identified, such as flower class, birds, dog class, automobile, and generally they all have an identical structure, therefore how
Acquiring, there is the feature for differentiating property to turn into the main task of fine granularity identification in image.
In past research, fine granularity identification field mainly includes two tasks：Local positioning and character representation.Particulate
Degrees of data collection is also typically provided the mark of extra object boundary frame and part, before in addition to image category label
Many work more or less all rely on these extra marks, but fine-grained classification usually requires knowing for expert level
Know, and ordinary people is difficult to complete this task, the cost that this allows for manually marking is sufficiently expensive.In recent years, more researchs
The method for not needing any extra markup information is concentrated on, method of the present invention just only needs image category label, without
Local mark is needed, the local feature with identification is arrived with a kind of Weakly supervised method study.
Described for the feature of image, CNN features achieve breakthrough on many benchmark.Conventional method is by local message
Coding, global feature expression is then fused into, CNN features are different, and it can directly learn the overall situation, and need not be artificial
Design feature extractor, current fine granularity recognition methods are all on CNN basis, learn image by other algorithms
Trickle unique feature.
The content of the invention
The technical problem to be solved by the invention is to provide one kind more accurately to learn minutia, gives up in thumbnail
Mixed and disorderly background information, so as to improve nicety of grading, and no longer need identification in the fine granularity identification of global characteristics auxiliary
The improved method of segment study.
The technical solution adopted in the present invention is：The improved method that identification segment learns in a kind of fine granularity identification, bag
Include following steps：
1) extracting has the segment for differentiating property in original image, including：
(1) original image obtains the characteristic pattern of a C × H × W sizes by the convolution pond layer in convolutional neural networks,
Wherein, C is pipe number, and H is height, and W is width, and the vector of C × 1 of each space fixed position in characteristic pattern × 1 is considered as
Corresponding to the detector of relevant position segment in beginning image；
(2) assume learnt in original image an identification region have highest respond detector, by C ×
The detector of 1 × 1 size carries out convolution algorithm with the characteristic pattern of C × H × W sizes, obtains the response diagram of new H × W sizes；
(3) position with maximum is selected in the response diagram of new H × W sizes, obtains having for 1 × 1 size and sentence
The segment of other property；
2) study has the feature for the segment for differentiating property and for classifying, including：
(1) segment for differentiating property according to having obtains local notable figure；
(2) local notable figure is encoded using spatial weighting Fei Sheer vectors.
Study described in step 1) (2) step in original image an identification region there is the detection that highest responds
The step of device is：
(1) assume to be set to n per the identification segment detector number of class image, image shares M classes, required detector
Quantity is exactly nM；
(2) detector of nM C × 1 × 1 and the characteristic pattern of C × H × W sizes are subjected to convolution algorithm respectively, obtained new
Characteristic pattern, global maximum pond is carried out to new characteristic pattern, obtain the characteristic vector of a nM dimension；
(3) each category feature vector in the characteristic vector of nM dimensions is averaged, obtains the vector of a M dimension；
(4) vector of M dimensions is passed to Softmax loss functions, the detector of C × 1 × 1 carried out with back-propagation algorithm
Training, after the completion of training, obtain the detector that there is highest to respond in an identification region in original image；
Acquisition part notable figure described in step 2) (1) step is：Calculated according to the notable figure S obtained from original image
Local notable figure Q, represent as follows：
Wherein, p is the pixel of identification segment, and i is detection position, when i-th of detection position includes pixel p, then Di
(p)=1, otherwise Di(p)=0, S (p) is the notable figure of whole image, and Q (p) is local notable figure, and Z is one normalized normal
Amount causes maxQ (p)=1.
Step 2) (2) step includes：
Assuming that vectorial I=(z1,…,zN) it is a series of D dimensional feature vectors from image zooming-out, the Fei Sheer of a pictures I
Vector coding φ (I)=(u1,v1,…,uk,vk) it is mean square deviation ukWith covariance vkAccumulation, uk, vkIt is written as following form：
Wherein, j=1 ..., D represents vector dimension, defines o=(μk,σk,πk:K=1 ..., K) it is Gaussian Mixture mould
The parameter of type, qikIt is each vectorial z of pattern k in mixed modeliPosterior probability, wherein, i=1 ..., N.
For each vectorial ziIntroduce spatial weighting item Q (pi), ujkAnd vjkResult after weighting is expressed as：
Wherein, Q (pi) it is local notable figure, uijk、vijkIt is formula (2) (3) respectively, by introducing space weight, so that it may
Important feature is arrived with study.
The improved method that identification segment learns in a kind of fine granularity identification of the present invention, by CNN features and Fei Sheer to
Amount is combined, study to the identification feature for being more suitable for fine granularity identification mission, and background information is dry in reduction identification segment
Disturb to improve nicety of grading.The improvement that the present invention learns primarily directed to existing identification segment, for the segment learnt
Identification region detected by detector, except the identification feature of object, unnecessary background information be present toward contact, because
This present invention improves this limitation using local notable figure and Fei Sheer vector codings, makes full use of identification provincial characteristics, with
Effective for classification task.Its advantage is mainly reflected in：
1) novelty：Current most effective most popular character representation method is CNN, but the present invention will take for particular problem
She Er vector codings and CNN features are combined.Because the above all study of identification feature in fine granularity identification problem, and
And the background generally in data set is all much like, thus novelty of the present invention introduce Fei Sheer vectors, can efficiently reduce
The interference of background information in identification segment.
2) validity：Compared with original method, what the present invention designed is compiled based on local notable figure and Fei Sheer vectors
The method of code can effectively learn local feature, and traditional CNN usually requires the rectangle of fixed size as input, and this will
Invalid information comprising background, and the present invention can efficiently reduce the interference of ambient noise, so that what is learnt is local special
Sign more has identification, and then improves nicety of grading.
3) practicality：Simple possible, it is an end-to-end network, can be identified effective for fine granularity.
Brief description of the drawings
Fig. 1 is the flow chart of the improved method that identification segment learns in a kind of fine granularity identification of the present invention.
Embodiment
1. the improved method that identification segment learns in a kind of fine granularity identification, it is characterised in that comprise the following steps：
1) extracting has the segment for differentiating property in original image, including：
(1) original image obtains the characteristic pattern of a C × H × W sizes by the convolution pond layer in convolutional neural networks,
Wherein, C is pipe number, and H is height, and W is width, and the vector of C × 1 of each space fixed position in characteristic pattern × 1 is considered as
Corresponding to the detector of relevant position segment in beginning image；
(2) assume learnt in original image an identification region have highest respond detector, by C ×
The detector of 1 × 1 size carries out convolution algorithm with the characteristic pattern of C × H × W sizes, obtains the response diagram of new H × W sizes；
(3) position with maximum is selected in the response diagram of new H × W sizes, obtains having for 1 × 1 size and sentence
The segment of other property；
2) study has the feature for the segment for differentiating property and for classifying, including：
(1) segment for differentiating property according to having obtains local notable figure；
(2) local notable figure is encoded using spatial weighting Fei Sheer vectors；
2. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, its feature
It is, the identification region in original image of the study described in step 1) (2) step has the detector of highest response
Step is：
(1) assume to be set to n per the identification segment detector number of class image, image shares M classes, required detector
Quantity be exactly nM；
(2) detector of nM C × 1 × 1 and the characteristic pattern of C × H × W sizes are subjected to convolution algorithm respectively, obtained new
Characteristic pattern, global maximum pond is carried out to new characteristic pattern, obtain the characteristic vector of a nM dimension；
(3) each category feature vector in the characteristic vector of nM dimensions is averaged, obtains the vector of a M dimension；
(4) vector of M dimensions is passed to Softmax loss functions, the detector of C × 1 × 1 carried out with back-propagation algorithm
Training, after the completion of training, obtain the detector that there is highest to respond in an identification region in original image；
3. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, its feature
It is, the acquisition part notable figure described in step 2) (1) step is：According to the notable figure S calculating office obtained from original image
Portion notable figure Q, represent as follows：
Wherein, p is the pixel of identification segment, and i is detection position, when i-th of detection position includes pixel p, then Di
(p)=1, otherwise Di(p)=0, S (p) is the notable figure of whole image, and Q (p) is local notable figure, and Z is one normalized normal
Amount causes maxQ (p)=1.
4. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, its feature
It is, step 2) (2) step includes：
Assuming that vectorial I=(z1,…,zN) it is a series of D dimensional feature vectors from image zooming-out, the Fei Sheer of a pictures I
Vector coding φ (I)=(u1,v1,…,uk,vk) it is mean square deviation ukWith covariance vkAccumulation, uk, vkIt is written as following form：
Wherein, j=1 ..., D represents vector dimension, defines o=(μk,σk,πk:K=1 ..., K) it is Gaussian Mixture mould
The parameter of type, qikIt is each vectorial z of pattern k in mixed modeliPosterior probability, wherein, i=1 ..., N.
For each vectorial ziIntroduce spatial weighting item Q (pi), ujkAnd vjkResult after weighting is expressed as：
Wherein, Q (pi) it is local notable figure, uijk、vijkIt is formula (2) (3) respectively, by introducing space weight, so that it may
Important feature is arrived with study.
Instantiation is provided with reference to Fig. 1：
Fig. 1 describes the structure flow chart of the present invention.Structure of the present invention is mainly made up of three parts, as in Fig. 1 1., 2.,
It is 3. shown.This method is to be based on VGG-16 models, and the model shares 16 layers.Implementation process is divided into two stages：Training stage
And test phase.
In the training stage, mainly learn the parameter of detector, in its process such as Fig. 1 1., it is 2. shown.
(1) input picture first is by the way that in the good convolutional neural networks VGG-16 of pre-training, conv4-3 output sizes are
512 × 28 × 28 characteristic pattern, therefore, the size of each detector is 512 × 1 × 1.By the number of every a kind of detector
It is set to 10, then share 2000 detectors for CUB200-2011 data sets；
(2) each detector is subjected to convolution with 512 × 28 × 28 resulting characteristic patterns and obtains 28 × 28 sizes
Response diagram；
(3) response diagram is obtained to the characteristic vector of one 2000 dimension behind global maximum pond,
(4) to being averaged in 2000 dimensional vectors per a kind of characteristic vector, the vector of one 200 dimension is obtained, by this
Vector behind average pond is passed to Softmax loss functions and is trained by back-propagation algorithm, it is possible to obtains for every
One kind can extract the detector of identification segment, for fine granularity identification, identification segment that this method detects
It is sufficiently small, be advantageous to the identification of image.
Test phase, as in Fig. 1 1., it is 3. shown.For the detector trained, (1)-(3) step in repetition training stage,
The response diagram that size is 1 × 1 is can obtain, can identify that every pictures have the position for differentiating property.Then image is calculated
Local notable figure, local notable figure are obtained by Local map and image saliency map two parts, the identification extracted from original image
Segment is multiplied with global notable figure, as formula (1) obtains local notable figure.Local notable figure is for showing that pixel belongs to prospect
Possibility, can effectively reduce the interference of background.According to the Fei Sheer vector design weights that local notable figure is image, obtain
Spatial weighting Fei Sheer vectors, by introducing weight, it is possible to the study feature important into fine granularity identification mission, it is final real
The classification of existing fine granularity image.
Claims (4)
1. the improved method that identification segment learns in a kind of fine granularity identification, it is characterised in that comprise the following steps：
1) extracting has the segment for differentiating property in original image, including：
(1) original image obtains the characteristic pattern of a C × H × W sizes by the convolution pond layer in convolutional neural networks, its
In, C is pipe number, and H is height, and W is width, and the vector of C × 1 of each space fixed position in characteristic pattern × 1 is considered as pair
Should in beginning image relevant position segment detector；
(2) assume learnt in original image an identification region have highest respond detector, by C × 1 × 1
The detector of size carries out convolution algorithm with the characteristic pattern of C × H × W sizes, obtains the response diagram of new H × W sizes；
(3) position with maximum is selected in the response diagram of new H × W sizes, obtain 1 × 1 size has identification
The segment of matter；
2) study has the feature for the segment for differentiating property and for classifying, including：
(1) segment for differentiating property according to having obtains local notable figure；
(2) local notable figure is encoded using spatial weighting Fei Sheer vectors.
2. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, it is characterised in that
Study described in step 1) (2) step in original image an identification region there is the step of detector that highest responds
It is：
(1) assume to be set to n per the identification segment detector number of class image, image shares M classes, the quantity of required detector
It is exactly nM；
(2) detector of nM C × 1 × 1 and the characteristic pattern of C × H × W sizes are subjected to convolution algorithm respectively, obtain new feature
Figure, global maximum pond is carried out to new characteristic pattern, obtain the characteristic vector of a nM dimension；
(3) each category feature vector in the characteristic vector of nM dimensions is averaged, obtains the vector of a M dimension；
(4) vector of M dimensions is passed to Softmax loss functions, the detector of C × 1 × 1 instructed with back-propagation algorithm
Practice, after the completion of training, obtain the detector that there is highest to respond in an identification region in original image.
3. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, it is characterised in that
Acquisition part notable figure described in step 2) (1) step is：Calculated according to the notable figure S obtained from original image local notable
Scheme Q, represent as follows：
Wherein, p is the pixel of identification segment, and i is detection position, when i-th of detection position includes pixel p, then Di(p)=
1, otherwise Di(p)=0, S (p) is the notable figure of whole image, and Q (p) is local notable figure, and Z is that a normalized constant makes
Obtain maxQ (p)=1.
4. the improved method that identification segment learns in a kind of fine granularity identification according to claim 1, it is characterised in that
Step 2) (2) step includes：
Assuming that vectorial I=(z1,…,zN) it is a series of D dimensional feature vectors from image zooming-out, the Fei Sheer vectors of a pictures I
Encode φ (I)=(u1,v1,…,uk,vk) it is mean square deviation ukWith covariance vkAccumulation, uk, vkIt is written as following form：
Wherein, j=1 ..., D represents vector dimension, defines o=(μk,σk,πk:K=1 ..., K) it is gauss hybrid models
Parameter, qikIt is each vectorial z of pattern k in mixed modeliPosterior probability, wherein, i=1 ..., N.
For each vectorial ziIntroduce spatial weighting item Q (pi), ujkAnd vjkResult after weighting is expressed as：
Wherein, Q (pi) it is local notable figure, uijk、vijkIt is formula (2) (3) respectively, by introducing space weight, it is possible to learn
To important feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711040828.XA CN107766890B (en) | 2017-10-31 | 2017-10-31 | Improved method for discriminant graph block learning in fine-grained identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711040828.XA CN107766890B (en) | 2017-10-31 | 2017-10-31 | Improved method for discriminant graph block learning in fine-grained identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107766890A true CN107766890A (en) | 2018-03-06 |
| CN107766890B CN107766890B (en) | 2021-09-14 |
Family
ID=61271840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711040828.XA Active CN107766890B (en) | 2017-10-31 | 2017-10-31 | Improved method for discriminant graph block learning in fine-grained identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107766890B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409384A (en) * | 2018-09-30 | 2019-03-01 | 内蒙古科技大学 | Image-recognizing method, device, medium and equipment based on fine granularity image |
| CN109815973A (en) * | 2018-12-07 | 2019-05-28 | 天津大学 | A kind of deep learning method suitable for the identification of fish fine granularity |
| CN109948628A (en) * | 2019-03-15 | 2019-06-28 | 中山大学 | A kind of object detection method excavated based on identification region |
| CN110019915A (en) * | 2018-07-25 | 2019-07-16 | 北京京东尚科信息技术有限公司 | Detect the method, apparatus and computer readable storage medium of picture |
| CN110197202A (en) * | 2019-04-30 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of local feature fine granularity algorithm of target detection |
| CN110309858A (en) * | 2019-06-05 | 2019-10-08 | 大连理工大学 | Based on the fine granularity image classification algorithms for differentiating study |
| CN110363233A (en) * | 2019-06-28 | 2019-10-22 | 西安交通大学 | A kind of the fine granularity image-recognizing method and system of the convolutional neural networks based on block detector and Fusion Features |
| CN110796183A (en) * | 2019-10-17 | 2020-02-14 | 大连理工大学 | Weak supervision fine-grained image classification algorithm based on relevance-guided discriminant learning |
| CN111062438A (en) * | 2019-12-17 | 2020-04-24 | 大连理工大学 | Weak supervision fine-grained image classification algorithm based on graph propagation of correlation learning |
| CN112154451A (en) * | 2018-05-18 | 2020-12-29 | 悟图索知 | Method, apparatus and computer program for extracting representative features of objects in an image |
| CN112927221A (en) * | 2020-12-09 | 2021-06-08 | 广州市玄武无线科技股份有限公司 | Image fine-grained feature-based reproduction detection method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573744A (en) * | 2015-01-19 | 2015-04-29 | 上海交通大学 | Fine granularity classification recognition method and object part location and feature extraction method thereof |
| WO2016168235A1 (en) * | 2015-04-17 | 2016-10-20 | Nec Laboratories America, Inc. | Fine-grained image classification by exploring bipartite-graph labels |
| CN106778705A (en) * | 2017-02-04 | 2017-05-31 | 中国科学院自动化研究所 | A kind of pedestrian's individuality dividing method and device |
| CN106778807A (en) * | 2016-11-22 | 2017-05-31 | 天津大学 | The fine granularity image classification method of dictionary pair is relied on based on public dictionary pair and class |
| CN106778804A (en) * | 2016-11-18 | 2017-05-31 | 天津大学 | The zero sample image sorting technique based on category attribute transfer learning |
| CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
-
2017
- 2017-10-31 CN CN201711040828.XA patent/CN107766890B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104573744A (en) * | 2015-01-19 | 2015-04-29 | 上海交通大学 | Fine granularity classification recognition method and object part location and feature extraction method thereof |
| WO2016168235A1 (en) * | 2015-04-17 | 2016-10-20 | Nec Laboratories America, Inc. | Fine-grained image classification by exploring bipartite-graph labels |
| CN106778804A (en) * | 2016-11-18 | 2017-05-31 | 天津大学 | The zero sample image sorting technique based on category attribute transfer learning |
| CN106778807A (en) * | 2016-11-22 | 2017-05-31 | 天津大学 | The fine granularity image classification method of dictionary pair is relied on based on public dictionary pair and class |
| CN106778705A (en) * | 2017-02-04 | 2017-05-31 | 中国科学院自动化研究所 | A kind of pedestrian's individuality dividing method and device |
| CN106951872A (en) * | 2017-03-24 | 2017-07-14 | 江苏大学 | A kind of recognition methods again of the pedestrian based on unsupervised depth model and hierarchy attributes |
Non-Patent Citations (3)
| Title |
|---|
| XIAO LIU 等: "Fully Convolutional Attention Localization Networks: Efficient Attention Localization for Fine-Grained Recognition", 《ARXIV》 * |
| XIAOPENG ZHANG 等: "Picking Deep Filter Responses for Fine-grained Image Recognition", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| 冯语姗 等: "自上而下注意图分割的细粒度图像分类", 《中国图象图形学报》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112154451A (en) * | 2018-05-18 | 2020-12-29 | 悟图索知 | Method, apparatus and computer program for extracting representative features of objects in an image |
| CN110019915A (en) * | 2018-07-25 | 2019-07-16 | 北京京东尚科信息技术有限公司 | Detect the method, apparatus and computer readable storage medium of picture |
| CN109409384A (en) * | 2018-09-30 | 2019-03-01 | 内蒙古科技大学 | Image-recognizing method, device, medium and equipment based on fine granularity image |
| CN109815973A (en) * | 2018-12-07 | 2019-05-28 | 天津大学 | A kind of deep learning method suitable for the identification of fish fine granularity |
| CN109948628A (en) * | 2019-03-15 | 2019-06-28 | 中山大学 | A kind of object detection method excavated based on identification region |
| CN109948628B (en) * | 2019-03-15 | 2023-01-03 | 中山大学 | Target detection method based on discriminant region mining |
| CN110197202A (en) * | 2019-04-30 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of local feature fine granularity algorithm of target detection |
| CN110309858B (en) * | 2019-06-05 | 2022-07-01 | 大连理工大学 | Fine-grained image classification method based on discriminant learning |
| CN110309858A (en) * | 2019-06-05 | 2019-10-08 | 大连理工大学 | Based on the fine granularity image classification algorithms for differentiating study |
| CN110363233A (en) * | 2019-06-28 | 2019-10-22 | 西安交通大学 | A kind of the fine granularity image-recognizing method and system of the convolutional neural networks based on block detector and Fusion Features |
| CN110363233B (en) * | 2019-06-28 | 2021-05-28 | 西安交通大学 | Fine-grained image recognition method and system of convolutional neural network based on block detector and feature fusion |
| CN110796183A (en) * | 2019-10-17 | 2020-02-14 | 大连理工大学 | Weak supervision fine-grained image classification algorithm based on relevance-guided discriminant learning |
| CN111062438A (en) * | 2019-12-17 | 2020-04-24 | 大连理工大学 | Weak supervision fine-grained image classification algorithm based on graph propagation of correlation learning |
| CN111062438B (en) * | 2019-12-17 | 2023-06-16 | 大连理工大学 | Image propagation weak supervision fine granularity image classification algorithm based on correlation learning |
| CN112927221A (en) * | 2020-12-09 | 2021-06-08 | 广州市玄武无线科技股份有限公司 | Image fine-grained feature-based reproduction detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107766890B (en) | 2021-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107766890A (en) | The improved method that identification segment learns in a kind of fine granularity identification | |
| CN107292298B (en) | Ox face recognition method based on convolutional neural networks and sorter model | |
| CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
| Kang et al. | Extended random walker for shadow detection in very high resolution remote sensing images | |
| CN105069481B (en) | Natural scene multiple labeling sorting technique based on spatial pyramid sparse coding | |
| CN109949316A (en) | A kind of Weakly supervised example dividing method of grid equipment image based on RGB-T fusion | |
| CN107194418B (en) | Rice aphid detection method based on antagonistic characteristic learning | |
| CN107506740A (en) | A kind of Human bodys' response method based on Three dimensional convolution neutral net and transfer learning model | |
| CN108304873A (en) | Object detection method based on high-resolution optical satellite remote-sensing image and its system | |
| CN106991382A (en) | A kind of remote sensing scene classification method | |
| CN107134144A (en) | A kind of vehicle checking method for traffic monitoring | |
| CN106909902A (en) | A kind of remote sensing target detection method based on the notable model of improved stratification | |
| CN109271991A (en) | A kind of detection method of license plate based on deep learning | |
| CN110163236A (en) | The training method and device of model, storage medium, electronic device | |
| CN107145889A (en) | Target identification method based on double CNN networks with RoI ponds | |
| CN103049763A (en) | Context-constraint-based target identification method | |
| CN105279519A (en) | Remote sensing image water body extraction method and system based on cooperative training semi-supervised learning | |
| CN105825233B (en) | A kind of pedestrian detection method based on on-line study random fern classifier | |
| CN111382727B (en) | Dog face recognition method based on deep learning | |
| CN107767416A (en) | The recognition methods of pedestrian's direction in a kind of low-resolution image | |
| CN107545571A (en) | A kind of image detecting method and device | |
| CN107301376A (en) | A kind of pedestrian detection method stimulated based on deep learning multilayer | |
| CN107025442A (en) | A kind of multi-modal fusion gesture identification method based on color and depth information | |
| CN103810500A (en) | Place image recognition method based on supervised learning probability topic model | |
| CN105956610B (en) | A kind of remote sensing images classification of landform method based on multi-layer coding structure |
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 |