CN105956560B - A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization - Google Patents
A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization Download PDFInfo
- Publication number
- CN105956560B CN105956560B CN201610295487.XA CN201610295487A CN105956560B CN 105956560 B CN105956560 B CN 105956560B CN 201610295487 A CN201610295487 A CN 201610295487A CN 105956560 B CN105956560 B CN 105956560B
- Authority
- CN
- China
- Prior art keywords
- feature
- depth convolution
- pondization
- image
- multiple dimensioned
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of model recognizing methods based on the multiple dimensioned depth convolution feature of pondization, first each vehicle image to model data library, extract its depth convolution feature by different scale, first scale is not handled；By the depth convolution feature of remaining each scale, PCA dimensionality reduction is carried out；Local feature polymerization description son coding is carried out again；Then again by PCA dimensionality reduction, the character representation of current scale is obtained；The feature of all scales is cascaded into pond, obtains the final character representation of present image；The character representation of vehicle image is used for linear SVM training, obtains model recognition system；To vehicle to be identified, its character representation is equally obtained, importing identifying system may recognize that its vehicle.Traditional depth convolution feature lacks geometric invariance, limits to the vehicle image classification of variable scene and identification, and the present invention takes the multiple dimensioned depth convolution feature of the pondization of image, has well solved this problem, practicability with higher and robustness.
Description
Technical field
The invention belongs to image procossing, pattern classification and identification technology fields, in particular to a kind of multiple dimensioned based on pondization
The model recognizing method of depth convolution feature.
Background technique
Traditional vehicle cab recognition technology includes the processing such as vehicle detection segmentation, feature extraction and selection, pattern-recognition.It is this kind of
Technology is faced with many difficult points: how to be partitioned under complex background complete target vehicle region be vehicle cab recognition premise and
Basis；How in numerous features of automobile representative feature is selected, and converts it into effective parameter also especially
It is important；After obtaining characteristic parameter, how correctly to select and design classifier and also directly affect the accuracy rate finally identified.
The concept of deep learning originates from artificial neural network, refers to neural network with multi-layer structure.Deep learning
The hierarchical structure of nervous system is mainly imitated from bionic angle, low level indicates details, the abstract data of high-level expression
Structure feature, by being successively abstracted, the essential information of height mining data, to reach the destination of study.Convolutional neural networks
By locally-attached mode, weight is shared, and then efficiently solves the problem of being fully connected, this also makes convolutional neural networks exist
Image processing method face has unique superiority.
Currently, it is directed to different visual identity tasks, there has been proposed many various convolutional neural networks structures,
And achieve remarkable result.But it performs poor in terms of vehicle cab recognition, is lacked based on global convolutional neural networks feature
Geometric invariance limits classification and matching to variable scene.
According to the above, the invention proposes a kind of vehicle cab recognition sides based on the multiple dimensioned depth convolution feature of pondization
Method proposes a very succinct deep learning frame about vehicle cab recognition, by the Analysis On Multi-scale Features and depth of vehicle image
Convolutional network feature combines, and the coding mode of description is polymerize using local feature.Individual depth convolution feature lacks
Few geometric invariance limits the classification and matching to variable vehicle scene, and Analysis On Multi-scale Features utilize more abundant image
Information, the two combine and have well solved this problem, polymerize description using local feature and are encoded, improve operation speed
Degree reduces memory consumption, improves the accuracy rate of identification, practicability with higher and robustness on the whole.
Summary of the invention
Heretofore described method is in order to overcome the disadvantages of the above prior art, to carry out feature mainly for vehicle image
The problem of extracting and being finely divided identification to vehicle, proposes a kind of vehicle cab recognition based on the multiple dimensioned depth convolution feature of pondization
Method.Specific technical solution is as described below.
A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization, comprising the following steps:
Step 1: to each vehicle image of vehicle image data base, extracting its depth convolution feature by different scale；
Step 2: first scale is not handled, and the depth convolution feature of remaining each scale carries out PCA dimensionality reduction, dropped
Feature vector after dimension；
Step 3: local feature polymerization description son coding carried out to the feature vector after dimensionality reduction, feature after being encoded to
Amount；
Step 4: PCA dimensionality reduction being carried out to the feature vector after coding, obtains the character representation of current scale；
Step 5: the character representation of all scales being cascaded into pond, obtains the multiple dimensioned depth convolution feature of present image pondization
It indicates；
Step 6: the multiple dimensioned depth convolution character representation of the pondization of all vehicle images is used for linear SVM instruction
Practice, obtains model recognition system；
Step 7: to vehicle to be identified, equally obtaining the multiple dimensioned depth convolution character representation of its pondization, import vehicle cab recognition
System may recognize that its vehicle.
In above-mentioned technical proposal, the step 1 including the following steps:
Step 1.1: by original vehicle image scaling to 256*256 size；
Step 1.2: the mean value image U of all images in model data library after seeking scaling, then by the image after scaling
Mean value image U is subtracted, then imports in depth convolutional neural networks model and carries out feature extraction, by the 4096 of network model layer 7
Depth convolution feature of the dimensional feature as first scale, is no longer further processed；
Step 1.3: on the original image with window, step-length 32*32 extracts m image block, according to the side in step 1.2
Formula extracts the 4096 dimension depth convolution features of m image block；
Step 1.4: the window size in step 1.3 being set as 64*64, extracts n image block in the same way
4096 dimension depth convolution features.
In above-mentioned technical proposal, the step 2 including the following steps:
Step 2.1: by the m generated in step 1.3 4096 dimension depth convolution feature, being dropped to 500 using PCA dimensionality reduction
Dimension obtains the feature vector of m 500 dimension；
Step 2.2: by the n generated in step 1.4 4096 dimension depth convolution feature, being dropped to 500 using PCA dimensionality reduction
Dimension obtains the feature vector of n 500 dimension.
In above-mentioned technical proposal, the step 3 including the following steps:
Step 3.1: k-means cluster being carried out to a feature vector after dimensionality reduction in step 2.1, generates a 100*500
Code book；
Step 3.2: description being polymerize using local feature, each feature vector is encoded；
Step 3.3: after the normalization of two norms, the character representation of 50000 dimensions after being encoded；
Step 3.4: step 3.1-3.3 operation equally being carried out to n feature vector in step 2.2, after being encoded
The character representation of 50000 dimensions.
In above-mentioned technical proposal, the step 4 including the following steps:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3,4096 dimensions being dropped to using PCA dimensionality reduction, are made
For the character representation of second scale；
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4,4096 dimensions being dropped to using PCA dimensionality reduction, are made
For the character representation of third scale；
In above-mentioned technical proposal, the feature of all scales is cascaded into pond in the step 5, it is more to obtain present image pondization
Scale depth convolution character representation obtains working as front truck including cascade first scale of pondization to the feature vector of third scale
The feature vector of the final 3*4096 dimension of type image indicates.
In above-mentioned technical proposal, the step 6 including the following steps:
Step 6.1: using the multiple dimensioned depth convolution feature of the pondization of current class vehicle image as positive sample, other classifications
The multiple dimensioned depth convolution feature of the pondization of vehicle image is as negative sample；
Step 6.2: using the two samples of Linear SVM training, obtaining the classifier of current class vehicle image；
Step 6.3: repeating the operation that step 6.1 arrives step 6.2, obtain the classifier of all categories vehicle image, combine
After constitute model recognition system.
In above-mentioned technical proposal, the step 7 including the following steps:
Step 7.1: to vehicle image to be identified, the extraction of step 5 is arrived by step 1, obtains vehicle image to be identified
The multiple dimensioned depth convolution character representation of pondization；
Step 7.2: in trained model recognition system, will identify its vehicle in obtained character representation steps for importing 6
Type.
Because the present invention by adopting the above technical scheme, have it is following the utility model has the advantages that
Inventive concept is simple and clear, and the Analysis On Multi-scale Features of vehicle image and depth convolutional network feature are combined,
It polymerize the coding mode of description using local feature.Individual depth convolution feature lacks geometric invariance, limits pair
The classification and matching of variable vehicle scene, and Analysis On Multi-scale Features utilize more abundant image information, the two is combined and is solved well
It has determined this problem, description is polymerize using local feature and is encoded, arithmetic speed is improved, reduces memory consumption, it is whole
On improve the accuracy rate of identification, practicability with higher and robustness.
Detailed description of the invention
Fig. 1 is that the algorithm based on the multiple dimensioned depth convolution feature of pondization realizes schematic diagram.
Specific embodiment
In order to describe the technical content, the structural feature, the achieved object and the effect of this invention in detail, below in conjunction with embodiment
And attached drawing is cooperated to be explained in detail.
The invention proposes a kind of model recognizing methods based on the multiple dimensioned depth convolution feature of pondization, know in vehicle vehicle
Good effect Shang not obtained.Entire algorithm realize schematic diagram as shown in Figure 1, comprising steps of
Step 1: to each vehicle image of vehicle image data base, extracting its depth convolution feature, ruler by different scale
Degree 1 is not handled；
Specifically, to each vehicle image, extract the depth convolution feature under three scales here, and to scale 1 not into
Traveling single stepping only handles the depth convolution feature of remaining two scales, including the following steps:
Step 1.1: by original vehicle image scaling to 256*256 size；
Step 1.2: the mean value image of all images in model data library after seeking scaling；Then the image after scaling is subtracted
Mean value image is removed, then imports in depth convolutional neural networks model and carries out feature extraction, by 4096 dimensions of network model layer 7
Depth convolution feature of the feature as scale 1, is no longer further processed；
Step 1.3: on the original image with window 128*128, step-length 32*32 extracts m image block, according to step 1.2
In mode, depth convolution features are tieed up in extract m image block 4096；
Step 1.4: the window size in step 1.3 being set as 64*64, extracts n image block in the same way
4096 dimension depth convolution features.
Step 2: by the depth convolution feature of remaining each scale, carrying out PCA dimensionality reduction；
Specifically, to the depth convolution feature of scale 2 and scale 3, due to constituting with many image blocks, global dimension is non-
Chang Gao, so need to be further processed, including the following steps:
Step 2.1: by the m generated in step 1.3 4096 dimension depth convolution feature, being dropped to 500 using PCA dimensionality reduction
Dimension obtains the feature vector of m 500 dimension；
Step 2.2: by the n generated in step 1.4 4096 dimension depth convolution feature, being dropped to 500 using PCA dimensionality reduction
Dimension obtains the feature vector of n 500 dimension.
Step 3: local feature polymerization description son coding is carried out to the feature vector after dimensionality reduction；
Specifically, the feature vector after dimensionality reduction is indicated, needs further progress to encode, what is taken here is local feature
Polymerization description is encoded, including the following steps:
Step 3.1: k-means cluster being carried out to m feature vector after dimensionality reduction in step 2.1, generates a 100*500
Code book；
Step 3.2: description being polymerize using local feature, each feature vector is encoded；
Step 3.3: after the normalization of two norms, the character representation of 50000 dimensions after being encoded；
Step 3.4: step 3.1-3.3 operation equally being carried out to n feature vector in step 2.2, after being encoded
The character representation of 50000 dimensions.
Step 4: PCA dimensionality reduction being carried out to the feature vector after coding, obtains the character representation of current scale；
Specifically, the feature vector dimension after coding is still very high, and the complexity of calculating is quite high, it is also necessary to further into
Row dimensionality reduction equally takes PCA dimensionality reduction to be dropped to 4096 dimensions here, including the following steps:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3,4096 dimensions being dropped to using PCA dimensionality reduction, are made
For the character representation of scale 2；
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4,4096 dimensions being dropped to using PCA dimensionality reduction, are made
For the character representation of scale 3；
Step 5: the feature of all scales being cascaded into pond, obtains the multiple dimensioned depth convolution mark sheet of present image pondization
Show；
Specifically, the feature vector of scale 1 to scale 3 is indicated, contains the space of the original image under different scale
And structural information, need further to cascade pond, the feature vector for constituting the final dimension of image indicates.
Step 6: the multiple dimensioned depth convolution character representation of the pondization of all vehicle images is used for linear SVM instruction
Practice, obtains model recognition system；
Specifically, to the multiple dimensioned depth convolution character representation of the pondization of obtained all vehicle images, one is taken here
The training of vs rest linear SVM.The detailed process of one vs rest linear SVM training is: setting original instruction
There is K kind vehicle classification to need to divide when practicing, when extracting training set, extracts positive sample of each independent class as training set respectively
This collection, remaining all samples obtain the linear SVM classifier of K two classification as negative sample collection, by training, survey
When examination, corresponding test vector is tested with this K training result file respectively, each test has a scoring
(S1,...,Sk), final recognition result is exactly that highest classification of score value, also i.e. by vehicle classification to be identified be with
That of maximum classification function value is a kind of, including the following steps:
Step 6.1: using the multiple dimensioned depth convolution feature of the pondization of current class vehicle image as positive sample, other classifications
The multiple dimensioned depth convolution feature of the pondization of vehicle image is as negative sample；
Step 6.2: using the two samples of Linear SVM training, obtaining the classifier of current class vehicle image；
Step 6.3: repeating the operation that step 6.1 arrives step 6.2, obtain the classifier of all categories vehicle image, combine
After constitute model recognition system.
Step 7: to vehicle to be identified, equally obtaining the multiple dimensioned depth convolution character representation of its pondization, import identifying system
It may recognize that its vehicle.
Specifically, to vehicle to be identified, the multiple dimensioned depth convolution character representation of its pondization is obtained by same step, then
Importing identifying system may recognize that its vehicle, including the following steps:
Step 7.1: to vehicle image to be identified, the extraction of step 5 is arrived by step 1, obtains vehicle image to be identified
The multiple dimensioned depth convolution character representation of pondization；
Step 7.2: in trained model recognition system, will identify its vehicle in obtained character representation steps for importing 6
Type.
Claims (8)
1. a kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization, comprising the following steps:
Step 1: to each vehicle image of vehicle image data base, extracting its depth convolution feature by different scale；
Step 2: first scale is not handled, and the depth convolution feature of remaining each scale carries out PCA dimensionality reduction, after obtaining dimensionality reduction
Feature vector；
Step 3: local feature polymerization description son coding being carried out to the feature vector after dimensionality reduction, the feature vector after being encoded；
Step 4: PCA dimensionality reduction being carried out to the feature vector after coding, obtains the character representation of current scale；
Step 5: the character representation of all scales being cascaded into pond, obtains the multiple dimensioned depth convolution mark sheet of present image pondization
Show；
Step 6: the multiple dimensioned depth convolution character representation of the pondization of all vehicle images being used for linear SVM training, is obtained
To model recognition system；
Step 7: to vehicle to be identified, equally obtaining the multiple dimensioned depth convolution character representation of its pondization, import model recognition system
It may recognize that its vehicle.
2. the model recognizing method according to claim 1 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
State step 1 including the following steps:
Step 1.1: by original vehicle image scaling to 256*256 size；
Step 1.2: the mean value image of all images in model data library after seeking scaling；Then the image after scaling is subtracted
It is worth image, then imports in depth convolutional neural networks model and carry out feature extraction, by 4096 dimensional feature of network model layer 7
As the depth convolution feature of first scale, no longer it is further processed；
Step 1.3: on the original image with window 128*128, step-length 32*32 extracts m image block, according in step 1.2
Mode extracts the 4096 dimension depth convolution features of m image block；
Step 1.4: the window size in step 1.3 being set as 64*64, extracts 4096 dimensions of n image block in the same way
Depth convolution feature.
3. the model recognizing method according to claim 2 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
Step 2 including the following steps:
Step 2.1: by the m generated in step 1.3 4096 dimension depth convolution feature, 500 dimensions are dropped to using PCA dimensionality reduction,
Obtain the feature vector of m 500 dimension；
Step 2.2: by the n generated in step 1.4 4096 dimension depth convolution feature, 500 dimensions are dropped to using PCA dimensionality reduction,
Obtain the feature vector of n 500 dimension.
4. the model recognizing method according to claim 3 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
Step 3 including the following steps:
Step 3.1: k-means cluster being carried out to m feature vector after dimensionality reduction in step 2.1, generates the code of a 100*500
This；
Step 3.2: description being polymerize using local feature, each feature vector is encoded；
Step 3.3: after the normalization of two norms, the character representation of 50000 dimensions after being encoded；
Step 3.4: step 3.1-3.3 operation equally is carried out to n feature vector in step 2.2,50000 after being encoded
The character representation of dimension.
5. the model recognizing method according to claim 4 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
Step 4 including the following steps:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3,4096 dimensions being dropped to using PCA dimensionality reduction, as
The character representation of two scales；
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4,4096 dimensions being dropped to using PCA dimensionality reduction, as
The character representation of three scales.
6. the model recognizing method according to claim 1 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
The feature of all scales is cascaded into pond in step 5, obtains the multiple dimensioned depth convolution character representation of present image pondization, is wrapped
Cascade first scale of pondization is included to the feature vector of third scale, obtains the spy of the final 3*4096 dimension of current vehicle image
Levying vector indicates.
7. the model recognizing method according to claim 1 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
Step 6 including the following steps:
Step 6.1: using the multiple dimensioned depth convolution feature of the pondization of current class vehicle image as positive sample, other classification vehicles
The multiple dimensioned depth convolution feature of the pondization of image is as negative sample；
Step 6.2: using the two samples of Linear SVM training, obtaining the classifier of current class vehicle image；
Step 6.3: repeating the operation that step 6.1 arrives step 6.2, obtain the classifier of all categories vehicle image, structure after joint
At model recognition system.
8. the model recognizing method according to claim 1 based on the multiple dimensioned depth convolution feature of pondization, which is characterized in that institute
It states
Step 7 including the following steps:
Step 7.1: to vehicle image to be identified, the extraction of step 5 is arrived by step 1, obtains the pond of vehicle image to be identified
Multiple dimensioned depth convolution character representation；
Step 7.2: in trained model recognition system, will identify its vehicle in obtained character representation steps for importing 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610295487.XA CN105956560B (en) | 2016-05-06 | 2016-05-06 | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610295487.XA CN105956560B (en) | 2016-05-06 | 2016-05-06 | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105956560A CN105956560A (en) | 2016-09-21 |
| CN105956560B true CN105956560B (en) | 2019-07-09 |
Family
ID=56914761
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610295487.XA Expired - Fee Related CN105956560B (en) | 2016-05-06 | 2016-05-06 | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105956560B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503729A (en) * | 2016-09-29 | 2017-03-15 | 天津大学 | A kind of generation method of the image convolution feature based on top layer weights |
| CN106529446A (en) * | 2016-10-27 | 2017-03-22 | 桂林电子科技大学 | Vehicle type identification method and system based on multi-block deep convolutional neural network |
| CN106599797B (en) * | 2016-11-24 | 2019-06-07 | 北京航空航天大学 | A kind of infrared face recognition method based on local parallel neural network |
| CN108154153B (en) * | 2016-12-02 | 2022-02-22 | 北京市商汤科技开发有限公司 | Scene analysis method and system and electronic equipment |
| WO2018099473A1 (en) | 2016-12-02 | 2018-06-07 | 北京市商汤科技开发有限公司 | Scene analysis method and system, and electronic device |
| CN106909938B (en) * | 2017-02-16 | 2020-02-21 | 青岛科技大学 | Visual angle independence behavior identification method based on deep learning network |
| CN107133570B (en) * | 2017-04-07 | 2018-03-13 | 武汉睿智视讯科技有限公司 | A kind of vehicle/pedestrian detection method and system |
| CN107220657B (en) * | 2017-05-10 | 2018-05-18 | 中国地质大学（武汉） | A kind of method of high-resolution remote sensing image scene classification towards small data set |
| CN107122653A (en) * | 2017-05-11 | 2017-09-01 | 湖南星汉数智科技有限公司 | A kind of picture validation code processing method and processing device |
| CN107169455B (en) * | 2017-05-16 | 2020-08-28 | 中山大学 | Face attribute recognition method based on depth local features |
| CN107330463B (en) * | 2017-06-29 | 2020-12-08 | 南京信息工程大学 | Vehicle type identification method based on CNN multi-feature union and multi-kernel sparse representation |
| CN107506729B (en) * | 2017-08-24 | 2020-04-03 | 中国科学技术大学 | Visibility detection method based on deep learning |
| CN111052126B (en) | 2017-09-04 | 2024-06-04 | 华为技术有限公司 | Pedestrian attribute identification and positioning method and convolutional neural network system |
| CN107958219A (en) * | 2017-12-06 | 2018-04-24 | 电子科技大学 | Image scene classification method based on multi-model and Analysis On Multi-scale Features |
| CN108154502B (en) * | 2017-12-22 | 2022-01-11 | 王华锋 | Through hole welding spot identification method based on convolutional neural network |
| CN108062754B (en) * | 2018-01-19 | 2020-08-25 | 深圳大学 | Segmentation and identification method and device based on dense network image |
| CN109102010B (en) * | 2018-07-27 | 2021-06-04 | 北京以萨技术股份有限公司 | Image classification method based on bidirectional neural network structure |
| CN109410251B (en) * | 2018-11-19 | 2022-05-03 | 南京邮电大学 | Target tracking method based on dense connection convolution network |
| CN111583320B (en) * | 2020-03-17 | 2023-04-07 | 哈尔滨医科大学 | Breast cancer ultrasonic image typing method and system fusing deep convolutional network and image omics characteristics and storage medium |
| CN111753713B (en) * | 2020-06-23 | 2022-05-24 | 菏泽学院 | Electrocardiosignal identity recognition method and system based on sparse representation and deep cascading |
| CN112418168B (en) * | 2020-12-10 | 2024-04-02 | 深圳云天励飞技术股份有限公司 | Vehicle identification method, device, system, electronic equipment and storage medium |
| CN114022686A (en) * | 2021-12-07 | 2022-02-08 | 中国人民公安大学 | Pedestrian re-identification method oriented to occlusion scene |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184271A (en) * | 2015-09-18 | 2015-12-23 | 苏州派瑞雷尔智能科技有限公司 | Automatic vehicle detection method based on deep learning |
| CN105404859A (en) * | 2015-11-03 | 2016-03-16 | 电子科技大学 | Vehicle type recognition method based on pooling vehicle image original features |
-
2016
- 2016-05-06 CN CN201610295487.XA patent/CN105956560B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184271A (en) * | 2015-09-18 | 2015-12-23 | 苏州派瑞雷尔智能科技有限公司 | Automatic vehicle detection method based on deep learning |
| CN105404859A (en) * | 2015-11-03 | 2016-03-16 | 电子科技大学 | Vehicle type recognition method based on pooling vehicle image original features |
Non-Patent Citations (5)
| Title |
|---|
| Deep Learning of Scene-Specific Classifier for Pedestrian Detection;X Zeng 等;《Springer International Publishing》;20141231;第472-487页 * |
| Vehicle Type Classification Using Unsupervised Convolutional Neural Network;Zhen Dong 等;《22nd International Conference on Pattern Recogniton》;20141231;第172-177页 * |
| 基于多尺度梯度及深度神经网络的汉子识别;潘炜深 等;《北京航空航天大学学报》;20150430;第41卷(第4期);第751-756页 * |
| 基于深度卷积神经网络的车型识别研究;邓柳 等;《计算机应用研究》;20160331;第33卷(第3期);第930-932页 * |
| 多尺度级联行人检测算法的研究与实现;李梦涵 等;《计算机技术与发展》;20140831;第24卷(第8期);第10-13页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105956560A (en) | 2016-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105956560B (en) | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization | |
| CN107564025B (en) | Electric power equipment infrared image semantic segmentation method based on deep neural network | |
| CN107563385B (en) | License plate character recognition method based on depth convolution production confrontation network | |
| CN105389550B (en) | It is a kind of based on sparse guide and the remote sensing target detection method that significantly drives | |
| CN102938065B (en) | Face feature extraction method and face identification method based on large-scale image data | |
| CN110175613A (en) | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models | |
| CN106682569A (en) | Fast traffic signboard recognition method based on convolution neural network | |
| CN106570521B (en) | Multilingual scene character recognition method and recognition system | |
| CN110852182B (en) | Depth video human body behavior recognition method based on three-dimensional space time sequence modeling | |
| CN107590489A (en) | Object detection method based on concatenated convolutional neutral net | |
| CN105160310A (en) | 3D (three-dimensional) convolutional neural network based human body behavior recognition method | |
| CN107748873A (en) | A kind of multimodal method for tracking target for merging background information | |
| CN103514456A (en) | Image classification method and device based on compressed sensing multi-core learning | |
| CN105718889A (en) | Human face identity recognition method based on GB(2D)2PCANet depth convolution model | |
| CN105139004A (en) | Face expression identification method based on video sequences | |
| CN105373777A (en) | Face recognition method and device | |
| CN105046197A (en) | Multi-template pedestrian detection method based on cluster | |
| CN106778796A (en) | Human motion recognition method and system based on hybrid cooperative model training | |
| CN103440471B (en) | The Human bodys' response method represented based on low-rank | |
| CN109325507A (en) | A kind of image classification algorithms and system of combination super-pixel significant characteristics and HOG feature | |
| CN103400154A (en) | Human body movement recognition method based on surveillance isometric mapping | |
| CN104200228A (en) | Recognizing method and system for safety belt | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN103065158A (en) | Action identification method of independent subspace analysis (ISA) model based on relative gradient | |
| CN104298977A (en) | Low-order representing human body behavior identification method based on irrelevance constraint |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190709 Termination date: 20200506 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |