CN108197584A - A kind of recognition methods again of the pedestrian based on triple deep neural network - Google Patents
A kind of recognition methods again of the pedestrian based on triple deep neural network Download PDFInfo
- Publication number
- CN108197584A CN108197584A CN201810031436.5A CN201810031436A CN108197584A CN 108197584 A CN108197584 A CN 108197584A CN 201810031436 A CN201810031436 A CN 201810031436A CN 108197584 A CN108197584 A CN 108197584A
- Authority
- CN
- China
- Prior art keywords
- triple
- pedestrian
- feature
- scale
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of recognition methods again of the pedestrian based on triple deep neural network, it is characterised in that:Pedestrian's feature is extracted in the following ways, to realize that pedestrian identifies again, it is included on training set and randomly selects a pictures, the picture of three kinds of different scales is generated by original image, including original image fullScale, half scale original image halfScale and upper part of the body scale picture topscale;Then the picture of each scale is found out with its same class and inhomogeneous sample, to form triple;The triple sample under different scale is separately input in the different depth convolutional neural networks of three kinds of structures again, finally the output feature of three kinds of convolutional neural networks is cascaded, passes through normalized output final result.The present invention can efficiently extract pedestrian's feature and finally improve the result precision that pedestrian identifies again.
Description
Technical field
The present invention relates to a kind of pedestrian weight identification technologies, and in particular to a kind of pedestrian based on triple deep neural network
Recognition methods again.
Background technology
Pedestrian identifies again, that is, refers to that Auto-matching is the same as the skill of a group traveling together under the non-overlapping multi-cam picture of irradiation area
Art, rapidly and accurately to find moving frame and track of pedestrian's object under multi-cam.Pedestrian's weight identification technology at present
It has obtained widely paying close attention to and applying in academia and industrial quarters.However it can be faced with many when implementing the technology and choose
War, for example, pedestrian's object can by illumination, block, the factors such as visual angle are influenced, this causes the interior class (same pedestrian) between pedestrian
Difference is even greater than (different pedestrians) difference between class, so as to cause the failure of the problem.Practical pedestrian's weight Study of recognition work
In be broadly divided into three steps:Feature extraction (external appearance characteristic of pedestrian's object represents), distance metric (the similitude ratio between pedestrian
Compared with) and sorting consistence (optimization to ranking results).Present invention is generally directed to the first two steps, i.e. feature extraction and apart from degree
Amount.Last decade also achieves good success in the research in the two steps.Currently how in true monitor video scene
In to extract the Feature Descriptor with robustness and identification be a huge challenge.Particularly, in the method for deep learning
Pedestrian is successfully applied to again behind identification field, many methods all employ the training method of scale fusion.But in real scene
Many methods are difficult in adapt to the different scale picture in same pedestrian down.
Invention content
In view of the above problems, the present invention proposes a kind of pedestrian's weight based on multiple dimensioned triple depth convolutional neural networks
Recognition methods, it is intended to obtain pedestrian's Feature Descriptor more with robustness.
Technical solution of the present invention provides a kind of recognition methods again of the pedestrian based on triple deep neural network, and use is following
Mode extracts pedestrian's feature, to realize that pedestrian identifies again,
A pictures are randomly selected on training set, the picture of three kinds of different scales are generated by original image, including original
Beginning picture fullScale, half scale original image halfScale and upper part of the body scale picture topscale;Then to each
The picture of scale find out with its same class and inhomogeneous sample, to form triple;Again the triple sample under different scale
Originally it is separately input in the different depth convolutional neural networks of three kinds of structures, finally the output feature of three kinds of convolutional neural networks
It is cascaded, passes through normalized output final result.
Moreover, the composition triple realization method is as follows,
To original image fullScale, half scale original image halfScale and topscale points of upper part of the body scale picture
Not Zuo Wei anchor point sample Anchor, randomly select one in training set and Anchor belong to of a sort sample and inhomogeneous
Sample, the two samples it is corresponding be known as Positive and Negative, thus form one (Anchor, Positive,
Negative) triple.
Moreover, the triple sample under different scale is separately input to the different depth convolutional neural networks of three kinds of structures
In, train these three sub-networks using triple loss.
Moreover, setting Anchor is denoted as xi, Positive is denoted as xj, Negative is denoted as xk, use gW(x) each ruler is represented
Outputs of the picture x in neural network is spent, problem, training triple are identified again for pedestrian, final feature should will meet
Following formula,
||gW(xi)-gW(xj)||<||gW(xi)-gW(xk)||
It is calculated in order to facilitate partial derivative, being write as quadratic form is,
||gW(xi)-gW(xj)||2<||gW(xi)-gW(xk)||2
Finally for a given training set X, triple constraints conversion into following object function,
Wherein,
D (W, X) represents the sum of the deviations of distance between the distance of same person and different people;
N represents the sum of triple sample;
C represents constraint factor.
Moreover, the normalization, output are represented with following formula:
Wherein,
Y represents the feature after normalization;
G=[g1,g2,…,gk] represent the feature that series connection layer exports, wherein gpRepresent p-th of feature;
The dimension that k expressions are characterized.
Poor for the pedestrian's feature robustness extracted in existing pedestrian again recognition methods, the present invention proposes a kind of feasible
Method be used to efficiently extract pedestrian's feature and finally improve the result precision that identifies again of pedestrian, i.e., under multiple dimensioned, adopt
Realize that pedestrian identifies again with the depth convolutional neural networks based on triple.
Description of the drawings
Fig. 1 is the flow chart of the embodiment of the present invention;
Fig. 2 is the MST-CNN structure charts based on triple loss of the embodiment of the present invention;
Specific embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, with reference to the accompanying drawings and examples to this hair
It is bright to be described in further detail.
Pedestrian is successfully applied to again behind identification field in deep learning, and many methods all employ the training side of scale fusion
Formula.But many methods are difficult in adapt to the different scale picture in same pedestrian under real scene.The skill that the present invention uses
Art scheme is a kind of depth convolutional neural networks based on multiple dimensioned triple.When it is implemented, by under different scale
Triple is input in corresponding convolutional neural networks, and the output of each network obtains final line by L2normalization
The result that people identifies again.
A pictures are randomly selected on training set, the picture of three kinds of different scales is generated by original image, i.e., it is former
Beginning picture, half scale original image and upper part of the body scale picture.Then the picture of each scale is found out and it is of a sort
Sample and inhomogeneous sample, triple is formed under each scale.It is again that the triple difference under three kinds of different scales is defeated
Enter into the different convolutional neural networks of three kinds of structures, finally the output feature of three kinds of convolutional neural networks be cascaded,
Pass through L2 normalized output final results.
Referring to Fig. 1, embodiment specific implementation includes the following steps:
Step 1:By the way that the sequence of maneuvers such as original image are obtained with the picture of three kinds of different scales, i.e. fullScale,
HalfScale and topscale, and triple is constructed to each scale picture;Its specific steps includes:
Step 1.1:A pictures are randomly choosed in training set, by obtaining three kinds of different scales to the processing of the picture
Picture:Original picture (fullScale), half scale picture (halfScale) and upper part of the body picture (topScale).
Step 1.2:For three kinds of above-mentioned different scale pictures, each scale picture is referred to as Anchor (anchors
Point sample).Then one and Anchor is randomly selected in training set again and (is denoted as xi) belong to of a sort sample and inhomogeneity
Sample, the two samples it is corresponding be known as Positive (be denoted as xj) and Negative (be denoted as xk), thus form one
(Anchor, Positive, Negative) triple.Positive, Negative identify positive sample respectively, and (and anchor point sample is
Same person) and negative sample (and anchor point sample is not same person).
Step 2:In the case where step 1 obtains three scales after different triples, each triple is input to corresponding depth
It spends in convolutional neural networks, therefore altogether containing there are three the MST-CNN networks of shared parameter;
Corresponding convolution operation can be used to following formula sublist and reach:
Here ai (l)And ai (l-1)Represent that l layers of i-th of output channel and i-th of l-1 layers output are logical respectively
Road;Represent the convolution kernel between i-th and j-th of feature map (characteristic pattern) in l layers;bi (l)It represents in l layers
I-th of feature map (characteristic pattern) bias term.Used herein is relu activation primitives, i.e. relu (x)=max (0,
X), x is variable.
It implements step and includes:
Step 2.1:The original picture size that embodiment is used is 227 × 227, and other two kinds of scale pictures of generation are all
114×114.Each sub-network structure realize step approximately as:
Step 2.1.1:For the triple of original image, it is input in first sub-network.There are five the networks
Convolutional layer, the convolution kernel size of each convolutional layer is 11 × 11, wherein carries out Maxpooling after first three convolution operation, most
The vector tieed up eventually by two full articulamentum outputs into one 1000.
Step 2.1.2:For the triple that the upper part of the body picture that extracts is formed, the network is there are three convolutional layer, three
Pond layer, the convolution kernel size of each convolutional layer is 5 × 5, eventually passes through what two full articulamentum outputs were tieed up into one 1000
Vector.
Step 2.1.3:For the triple that the half scale picture that extracts is formed, the network is there are two convolutional layer, two
Pond layer, the convolution kernel size of each convolutional layer is 3 × 3, eventually passes through what two full articulamentum outputs were tieed up into one 1000
Vector.
Step 2.2:(triplet loss) is lost using triple to train above three sub-network.Triplet loss
Purpose be by study Anchor is allowed (to be denoted as xi) and Positive (be denoted as xj) the distance between feature representation is as small as possible,
And Anchor (is denoted as xi) and Negative (be denoted as xk) distance between feature representation person is as big as possible.Use gW(x) each is represented
Outputs of the scale picture x in neural network.Problem is identified again for pedestrian, to train triple, final feature should be wanted
Meet following formula:
||gW(xi)-gW(xj)||<||gW(xi)-gW(xk)||
It is calculated in order to facilitate partial derivative, it, which is write as quadratic form, is:
||gW(xi)-gW(xj)||2<||gW(xi)-gW(xk)||2
Finally for a given training set X, triple constraint has been converted into following object function:
Wherein,
D (W, X) represents the sum of the deviations of distance between the distance of same person and different people;
N represents the sum of triple sample;
C represents constraint factor, is in the present invention 1, means the size for constraining above-mentioned error.
Referring to Fig. 2, wherein:
Image triplet:Image triple (anchor point sample, positive sample, negative sample)
feature triplet:For the character triple corresponding to above-mentioned image triple
gw(Xj):The character representation of positive sample
gw(Xj):The character representation of anchor point sample
gw(Xk):The character representation of negative sample
MST-CNN:Multiple dimensioned triple convolutional neural networks (Multi-scale Triplet Convolutional
Neural Network)
pull:Expression makes the distance between same person small as much as possible
push:Expression makes the distance between different pedestrians as large as possible
Step 3:By the way that the output feature in three networks is cascaded, final output result is obtained.In order to allow
Network is quickly restrained, and using L2 Normalization (normalization) layer, output is represented with following formula:
Wherein,
Y represents the feature after normalization;
G=[g1,g2,…,gk] represent the feature that series connection layer exports, wherein gpRepresent p-th of feature;
The dimension that k expressions are characterized.
L2Normalization layers of importance is that it can be ensured that the distance calculated from each triple will not
More than formula | | gW(xi)-gW(xj)||<||gW(xi)-gW(xk) | | the range given, i.e. positive sample to the distance between be less than it is negative
Sample to the distance between.
When it is implemented, the automatic running that computer software technology realizes more than flow can be used in those skilled in the art.
It should be understood that the part that this specification does not elaborate belongs to the prior art.It is above-mentioned to implement for preferable
The description of example is more detailed, therefore can not be considered the limitation to scope of patent protection of the present invention, the common skill of this field
Art personnel, in the case where not departing from the ambit that the claims in the present invention are protected, can also make and replace under the enlightenment of the present invention
It changes or deforms, each fall within protection scope of the present invention, it is of the invention range is claimed to be determined by the appended claims.
Claims (5)
1. a kind of recognition methods again of the pedestrian based on triple deep neural network, it is characterised in that:It extracts in the following ways
Pedestrian's feature, to realize that pedestrian identifies again,
A pictures are randomly selected on training set, the picture of three kinds of different scales are generated by original image, including original graph
Piece fullScale, half scale original image halfScale and upper part of the body scale picture topscale;Then to each scale
Picture find out with its same class and inhomogeneous sample, to form triple;Again the triple sample point under different scale
It is not input in the different depth convolutional neural networks of three kinds of structures, finally the output feature of three kinds of convolutional neural networks is connected
Together, pass through normalized output final result.
2. pedestrian's recognition methods again according to claim 1 based on triple deep neural network, it is characterised in that:It is described
It is as follows to form triple realization method,
Original image fullScale, half scale original image halfScale and upper part of the body scale picture topscale are made respectively
For anchor point sample Anchor, one and Anchor is randomly selected in training set and belongs to of a sort sample and inhomogeneous sample
This, the two samples it is corresponding be known as Positive and Negative, thus form one (Anchor, Positive,
Negative) triple.
3. pedestrian's recognition methods again according to claim 1 based on triple deep neural network, it is characterised in that:Not
It is separately input in the different depth convolutional neural networks of three kinds of structures with the triple sample under scale, is lost using triple
To train these three sub-networks.
4. pedestrian's recognition methods again according to claim 3 based on triple deep neural network, it is characterised in that:If
Anchor is denoted as xi, Positive is denoted as xj, Negative is denoted as xk, use gW(x) each scale picture x is represented in nerve net
Output in network identifies problem, training triple again for pedestrian, and final feature should will meet following formula,
||gW(xi)-gW(xj)||<||gW(xi)-gW(xk)||
It is calculated in order to facilitate partial derivative, being write as quadratic form is,
||gW(xi)-gW(xj)||2<||gW(xi)-gW(xk)||2
Finally for a given training set X, triple constraints conversion into following object function,
Wherein,
D (W, X) represents the sum of the deviations of distance between the distance of same person and different people;
N represents the sum of triple sample;
C represents constraint factor.
5. according to the recognition methods again of the pedestrian based on triple deep neural network, feature described in claims 1 or 2 or 3 or 4
It is:The normalization, output are represented with following formula:
Wherein,
Y represents the feature after normalization;
G=[g1,g2,…,gk] represent the feature that series connection layer exports, wherein gpRepresent p-th of feature;
The dimension that k expressions are characterized.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810031436.5A CN108197584A (en) | 2018-01-12 | 2018-01-12 | A kind of recognition methods again of the pedestrian based on triple deep neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810031436.5A CN108197584A (en) | 2018-01-12 | 2018-01-12 | A kind of recognition methods again of the pedestrian based on triple deep neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108197584A true CN108197584A (en) | 2018-06-22 |
Family
ID=62588930
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810031436.5A Pending CN108197584A (en) | 2018-01-12 | 2018-01-12 | A kind of recognition methods again of the pedestrian based on triple deep neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108197584A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109214430A (en) * | 2018-08-15 | 2019-01-15 | 天津大学 | A kind of recognition methods again of the pedestrian based on feature space topology distribution |
CN109344787A (en) * | 2018-10-15 | 2019-02-15 | 浙江工业大学 | A kind of specific objective tracking identified again based on recognition of face and pedestrian |
CN109544204A (en) * | 2018-10-15 | 2019-03-29 | 浙江工业大学 | A kind of shopping guide's behavior analysis method based on lightweight multitask convolutional neural networks |
CN109948425A (en) * | 2019-01-22 | 2019-06-28 | 中国矿业大学 | A kind of perception of structure is from paying attention to and online example polymerize matched pedestrian's searching method and device |
CN109977798A (en) * | 2019-03-06 | 2019-07-05 | 中山大学 | The exposure mask pond model training identified again for pedestrian and pedestrian's recognition methods again |
CN110084215A (en) * | 2019-05-05 | 2019-08-02 | 上海海事大学 | A kind of pedestrian of the twin network model of binaryzation triple recognition methods and system again |
CN111401202A (en) * | 2020-03-11 | 2020-07-10 | 西南石油大学 | Pedestrian mask wearing real-time detection method based on deep learning |
CN111582107A (en) * | 2020-04-28 | 2020-08-25 | 浙江大华技术股份有限公司 | Training method and recognition method of target re-recognition model, electronic equipment and device |
CN112492323A (en) * | 2019-09-12 | 2021-03-12 | 上海哔哩哔哩科技有限公司 | Live broadcast mask generation method, readable storage medium and computer equipment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140134581A1 (en) * | 2012-11-09 | 2014-05-15 | Hogan Assessment Systems, Inc. | Assessment for identifying derailers of interpersonal behavior |
CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
CN106600577A (en) * | 2016-11-10 | 2017-04-26 | 华南理工大学 | Cell counting method based on depth deconvolution neural network |
CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
-
2018
- 2018-01-12 CN CN201810031436.5A patent/CN108197584A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140134581A1 (en) * | 2012-11-09 | 2014-05-15 | Hogan Assessment Systems, Inc. | Assessment for identifying derailers of interpersonal behavior |
CN105631413A (en) * | 2015-12-23 | 2016-06-01 | 中通服公众信息产业股份有限公司 | Cross-scene pedestrian searching method based on depth learning |
CN106600577A (en) * | 2016-11-10 | 2017-04-26 | 华南理工大学 | Cell counting method based on depth deconvolution neural network |
CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
Non-Patent Citations (4)
Title |
---|
JIAWEI LIU 等: "Multi-Scale Triplet CNN for Person Re-Identification", 《COMPUTER SCIENCE》 * |
SHENGYONG DING 等: "Deep Feature Learning with Relative Distance Comparison for Person Re-identification", 《COMPUTER SCIENCE》 * |
XIAO WANG 等: "Multi-Pedestrian Detection from Effective Proposal in crowd Scene", 《PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON INTERNET MULTIMEDIA COMPUTING AND SERVICE》 * |
XIAO WANG 等: "Saliency-Based Deformable Model for Pedestrian Detection", 《PACIFIC RIM CONFERENCE ON MULTIMEDIA》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109214430A (en) * | 2018-08-15 | 2019-01-15 | 天津大学 | A kind of recognition methods again of the pedestrian based on feature space topology distribution |
CN109214430B (en) * | 2018-08-15 | 2021-10-22 | 天津大学 | Pedestrian re-identification method based on feature space topological distribution |
CN109344787A (en) * | 2018-10-15 | 2019-02-15 | 浙江工业大学 | A kind of specific objective tracking identified again based on recognition of face and pedestrian |
CN109544204A (en) * | 2018-10-15 | 2019-03-29 | 浙江工业大学 | A kind of shopping guide's behavior analysis method based on lightweight multitask convolutional neural networks |
CN109344787B (en) * | 2018-10-15 | 2021-06-08 | 浙江工业大学 | Specific target tracking method based on face recognition and pedestrian re-recognition |
CN109948425A (en) * | 2019-01-22 | 2019-06-28 | 中国矿业大学 | A kind of perception of structure is from paying attention to and online example polymerize matched pedestrian's searching method and device |
CN109948425B (en) * | 2019-01-22 | 2023-06-09 | 中国矿业大学 | Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching |
CN109977798B (en) * | 2019-03-06 | 2021-06-04 | 中山大学 | Mask pooling model training and pedestrian re-identification method for pedestrian re-identification |
CN109977798A (en) * | 2019-03-06 | 2019-07-05 | 中山大学 | The exposure mask pond model training identified again for pedestrian and pedestrian's recognition methods again |
CN110084215A (en) * | 2019-05-05 | 2019-08-02 | 上海海事大学 | A kind of pedestrian of the twin network model of binaryzation triple recognition methods and system again |
CN112492323A (en) * | 2019-09-12 | 2021-03-12 | 上海哔哩哔哩科技有限公司 | Live broadcast mask generation method, readable storage medium and computer equipment |
CN112492323B (en) * | 2019-09-12 | 2022-07-19 | 上海哔哩哔哩科技有限公司 | Live broadcast mask generation method, readable storage medium and computer equipment |
CN111401202A (en) * | 2020-03-11 | 2020-07-10 | 西南石油大学 | Pedestrian mask wearing real-time detection method based on deep learning |
CN111582107A (en) * | 2020-04-28 | 2020-08-25 | 浙江大华技术股份有限公司 | Training method and recognition method of target re-recognition model, electronic equipment and device |
CN111582107B (en) * | 2020-04-28 | 2023-09-29 | 浙江大华技术股份有限公司 | Training method and recognition method of target re-recognition model, electronic equipment and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108197584A (en) | A kind of recognition methods again of the pedestrian based on triple deep neural network | |
CN109359559B (en) | Pedestrian re-identification method based on dynamic shielding sample | |
DE112019005750T5 (en) | Learning to generate synthetic data sets for training neural networks | |
CN108416307A (en) | A kind of Aerial Images road surface crack detection method, device and equipment | |
DE102020002153A1 (en) | Use of an object attribute detecting models for the automatic selection of versions of detected objects in images | |
CN109584251A (en) | A kind of tongue body image partition method based on single goal region segmentation | |
CN108710863A (en) | Unmanned plane Scene Semantics dividing method based on deep learning and system | |
DE112016004535T5 (en) | Universal Compliance Network | |
CN112464911A (en) | Improved YOLOv 3-tiny-based traffic sign detection and identification method | |
CN105678231A (en) | Pedestrian image detection method based on sparse coding and neural network | |
CN106778852A (en) | A kind of picture material recognition methods for correcting erroneous judgement | |
CN109344699A (en) | Winter jujube disease recognition method based on depth of seam division convolutional neural networks | |
CN105426882B (en) | The method of human eye is quickly positioned in a kind of facial image | |
CN104392233B (en) | A kind of image saliency map extracting method based on region | |
CN106910188A (en) | The detection method of airfield runway in remote sensing image based on deep learning | |
CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning | |
CN109726748A (en) | A kind of GL-CNN remote sensing images scene classification method based on frequency band feature fusion | |
CN108537181A (en) | A kind of gait recognition method based on the study of big spacing depth measure | |
CN106683102A (en) | SAR image segmentation method based on ridgelet filters and convolution structure model | |
CN105095867A (en) | Rapid dynamic face extraction and identification method based deep learning | |
Li et al. | Gaofen-3 sea ice detection based on deep learning | |
Ma et al. | Image-based air pollution estimation using hybrid convolutional neural network | |
CN110163567A (en) | Classroom roll calling system based on multitask concatenated convolutional neural network | |
CN106886745A (en) | A kind of unmanned plane reconnaissance method based on the generation of real-time online map | |
CN107766828A (en) | UAV Landing Geomorphological Classification method based on wavelet convolution neutral net |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180622 |