CN105956560A - Vehicle model identification method based on pooling multi-scale depth convolution characteristics - Google Patents

Vehicle model identification method based on pooling multi-scale depth convolution characteristics Download PDF

Info

Publication number
CN105956560A
CN105956560A CN201610295487.XA CN201610295487A CN105956560A CN 105956560 A CN105956560 A CN 105956560A CN 201610295487 A CN201610295487 A CN 201610295487A CN 105956560 A CN105956560 A CN 105956560A
Authority
CN
China
Prior art keywords
depth convolution
degree
vehicle
feature
pondization
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201610295487.XA
Other languages
Chinese (zh)
Other versions
CN105956560B (en
Inventor
李鸿升
胡欢
曹滨
周辉
范峻铭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201610295487.XA priority Critical patent/CN105956560B/en
Publication of CN105956560A publication Critical patent/CN105956560A/en
Application granted granted Critical
Publication of CN105956560B publication Critical patent/CN105956560B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00624Recognising scenes, i.e. recognition of a whole field of perception; recognising scene-specific objects
    • G06K9/00791Recognising scenes perceived from the perspective of a land vehicle, e.g. recognising lanes, obstacles or traffic signs on road scenes
    • G06K9/00825Recognition of vehicle or traffic lights
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/36Image preprocessing, i.e. processing the image information without deciding about the identity of the image
    • G06K9/46Extraction of features or characteristics of the image
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6218Clustering techniques
    • G06K9/622Non-hierarchical partitioning techniques
    • G06K9/6221Non-hierarchical partitioning techniques based on statistics
    • G06K9/6223Non-hierarchical partitioning techniques based on statistics with a fixed number of clusters, e.g. K-means clustering
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computer systems based on biological models
    • G06N3/02Computer systems based on biological models using neural network models
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00Image coding
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K2209/00Indexing scheme relating to methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K2209/23Detecting or categorising vehicles
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20021Dividing image into blocks, subimages or windows

Abstract

The invention discloses a vehicle model identification method based on pooling multi-scale depth convolution characteristics, comprising the following steps: extracting the depth convolution characteristic of each vehicle model image in a vehicle model database according to different scales, wherein the first scale is not processed; carrying out PCA dimension reduction of the depth convolution characteristics of the remaining scales; carrying out coding using a local characteristic aggregation descriptor; carrying out PCA dimension reduction again to get the characteristic representation of the current scale; cascade-pooling the characteristics of all the scales to get the final characteristic representation of the current image; training a linear support vector machine using the characteristic representation of the vehicle model images to get a vehicle model identification system; and for a vehicle to be identified, acquiring the characteristic representation of the vehicle, and importing the vehicle into the identification system to identify the model of the vehicle. The traditional depth convolution characteristic lacks of geometrical invariability, which limits vehicle model image classification and identification in a variable scenario. The problem is well solved by using pooling multi-scale depth convolution characteristics of images. The vehicle model identification method of the invention is of high practicability and robustness.

Description

A kind of model recognizing method based on pondization multiple dimensioned degree of depth convolution feature
Technical field
The invention belongs to image procossing, pattern classification and identification technical field, multiple dimensioned based on pondization particularly to one The model recognizing method of degree of depth convolution feature.
Background of invention
Traditional vehicle cab recognition technology includes that vehicle detection segmentation, feature extraction and selection, pattern recognition etc. process.This kind of technology It is faced with many difficult points: under complex background, how to be partitioned into premise and base that complete target vehicle region is vehicle cab recognition Plinth;How in numerous features of automobile, to select representative feature, and it is the heaviest to convert it into effective parameter Want;After obtaining characteristic parameter, the most correctly select and design grader also to directly affect the accuracy rate of last identification.
The concept of degree of depth study originates from artificial neural network, refers to the neutral net with multiple structure.The degree of depth learns Mainly imitating neural hierarchical structure from bionic angle, low level represents details, the data that high-level expression is abstract Architectural feature, by the most abstract, the highly essential information of mining data, thus reach the destination of study.Convolutional neural networks By locally-attached mode, sharing weights, and then efficiently solve the problem being fully connected, this also makes convolutional neural networks exist Image processing method mask has the superiority of uniqueness.
At present, for different visual identity tasks, there has been proposed many various convolutional neural networks structures, And achieve remarkable result.But it is performed poor in terms of vehicle cab recognition, and convolutional neural networks feature based on the overall situation lacks Geometric invariance, limits the classification to variable scene and coupling.
In accordance with the above, the present invention proposes a kind of vehicle cab recognition side based on pondization multiple dimensioned degree of depth convolution feature Method, it is proposed that a degree of depth learning framework about vehicle cab recognition the most succinct, by Analysis On Multi-scale Features and the degree of depth of vehicle image Convolutional network feature combines, and have employed local feature polymerization and describes the coded system of son.Individually degree of depth convolution feature lacks Few geometric invariance, limits the classification to variable vehicle scene and coupling, and Analysis On Multi-scale Features utilizes the image of more horn of plenty Information, both combinations solve this problem well, use local feature polymerization to describe son and encode, improve computing speed Degree, reduces memory consumption, improves the accuracy rate of identification on the whole, have higher practicality and robustness.
Summary of the invention
Heretofore described method is the shortcoming in order to overcome above-mentioned prior art, carries out feature mainly for vehicle image The problem extracted and vehicle is finely divided identification, it is proposed that a kind of vehicle cab recognition based on pondization multiple dimensioned degree of depth convolution feature Method.Concrete technical scheme is as described below.
A kind of model recognizing method based on pondization multiple dimensioned degree of depth convolution feature, comprises the following steps:
Step 1: each vehicle image to vehicle image data base, extracts its degree of depth convolution feature by different scale;
Step 2: first yardstick does not processes, the degree of depth convolution feature of remaining each yardstick, carry out PCA dimensionality reduction, the spy after dimensionality reduction Levy vector;
Step 3: the characteristic vector after dimensionality reduction is carried out local feature polymerization and describes son coding, the characteristic vector after being encoded;
Step 4: the characteristic vector after coding is carried out PCA dimensionality reduction, obtains the character representation of current scale;
Step 5: the character representation of all yardsticks is cascaded pond, obtains present image pondization multiple dimensioned degree of depth convolution mark sheet Show;
Step 6: the pondization multiple dimensioned degree of depth convolution character representation of all vehicle images is used for linear SVM training, To model recognition system;
Step 7: to vehicle to be identified, same acquisition its pondization multiple dimensioned degree of depth convolution character representation, import identification system Identify its vehicle.
In technique scheme, described step 1 includes following step:
Step 1.1: ask for the average image U of all images in model data storehouse;
Step 1.2: by original vehicle image scaling to 256*256 size, then deducts average image U, then imports degree of depth convolution Neural network model carries out feature extraction, 4096 dimensional features of network model's layer 7 are rolled up as the degree of depth of first yardstick Long-pending feature, the most further processes;
Step 1.3: on the original image with window, step-length 32*32 is extracted m image block, according to the mode in step 1.2, is carried Take 4096 dimension degree of depth convolution features of m image block;
Step 1.4: the window size in step 1.3 is set to 64*64, extracts 4096 dimensions of n image block in the same way Degree of depth convolution feature.
In technique scheme, described step 2 includes following step:
Step 2.1: m the 4096 dimension degree of depth convolution feature that will generate in step 1.3, uses PCA dimensionality reduction to be dropped to 500 dimensions, Obtain the characteristic vector of m 500 dimension;
Step 2.2: n the 4096 dimension degree of depth convolution feature that will generate in step 1.4, uses PCA dimensionality reduction to be dropped to 500 dimensions, Obtain the characteristic vector of n 500 dimension.
In technique scheme, described step 3 includes the most several step:
Step 3.1: the individual characteristic vector after dimensionality reduction in step 2.1 carries out k-means cluster, generates the code of a 100*500 This;
Step 3.2: use local feature polymerization to describe son and each characteristic vector is encoded;
Step 3.3: after using two norm normalization, the character representation of 50000 dimensions after being encoded;
Step 3.4: equally n characteristic vector in step 2.2 is carried out the operation of step 3.1-3.3,50000 after being encoded The character representation of dimension.
In technique scheme, described step 4 includes the most several step:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as the The character representation of two yardsticks;
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as the The character representation of three yardsticks;
In technique scheme, the feature of all yardsticks is cascaded pond by described step 5, obtains present image pondization multiple dimensioned Degree of depth convolution character representation, including cascade each and every one yardstick of pondization first to the characteristic vector of the 3rd yardstick, obtains current vehicle The characteristic vector of the 3*4096 dimension that image is final represents.
In technique scheme, described step 6 includes following step:
Step 6.1: using the pondization multiple dimensioned degree of depth convolution feature of current class vehicle image as positive sample, other classification vehicles The pondization of image multiple dimensioned degree of depth convolution feature is as negative sample;
Step 6.2: use Linear SVM training the two sample, obtain the grader of current class vehicle image;
Step 6.3: repeat the step 6.1 operation to step 6.2, obtain the grader of all categories vehicle image, structure after associating Become model recognition system.
In technique scheme, described step 7 includes following step:
Step 7.1: to vehicle image to be identified, by the extraction of step 1 to step 5, obtain the pond of vehicle image to be identified Multiple dimensioned degree of depth convolution character representation;
Step 7.2: in the model recognition system trained in the character representation steps for importing 6 that will obtain, identify its vehicle.
Because the present invention uses technique scheme, therefore possess following beneficial effect:
Inventive concept is simple and clear, Analysis On Multi-scale Features and the degree of depth convolutional network feature of vehicle image is combined, and uses Local feature polymerization describes the coded system of son.Individually degree of depth convolution feature lacks geometric invariance, limits variable The classification of vehicle scene and coupling, and Analysis On Multi-scale Features utilizes the image information of more horn of plenty, both solve well in combination This problem, uses local feature polymerization to describe son and encodes, improve arithmetic speed, reduce memory consumption, carry on the whole The high accuracy rate identified, has higher practicality and robustness.
Accompanying drawing explanation
Fig. 1 is that algorithm based on pondization multiple dimensioned degree of depth convolution feature realizes schematic diagram.
Detailed description of the invention
By describing the technology contents of the present invention, structural feature in detail, being realized purpose and effect, below in conjunction with embodiment And coordinate accompanying drawing to be explained in detail.
The present invention proposes a kind of model recognizing method based on pondization multiple dimensioned degree of depth convolution feature, knows in vehicle vehicle Shang not obtain good effect.Whole algorithm realizes schematic diagram as it is shown in figure 1, include step:
Step 1: each vehicle image to vehicle image data base, extracts its degree of depth convolution feature by different scale, and yardstick 1 is not Process;
Specifically, to each vehicle image, extract the degree of depth convolution feature under three yardsticks here, and yardstick 1 is not entered Single stepping, only processes the degree of depth convolution feature of remaining two yardsticks, including following step:
Step 1.1: ask for the average image of all images in model data storehouse;
Step 1.2: by original vehicle image scaling to 256*256 size, then deducts average image, then imports degree of depth convolution god Feature extraction is carried out in network model, using 4096 dimensional features of network model's layer 7 as the degree of depth convolution feature of yardstick 1, The most further process;
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 4096 dimension degree of depth convolution features of m image block;
Step 1.4: the window size in step 1.3 is set to 64*64, extracts 4096 dimensions of n image block in the same way Degree of depth convolution feature.
Step 2: by the degree of depth convolution feature of remaining each yardstick, carry out PCA dimensionality reduction;
Specifically, the degree of depth convolution feature to yardstick 2 and yardstick 3, constitute owing to having a lot of image block, global dimension is very Height, so needing to process, including following step further:
Step 2.1: m the 4096 dimension degree of depth convolution feature that will generate in step 1.3, uses PCA dimensionality reduction to be dropped to 500 dimensions, Obtain the characteristic vector of m 500 dimension;
Step 2.2: n the 4096 dimension degree of depth convolution feature that will generate in step 1.4, uses PCA dimensionality reduction to be dropped to 500 dimensions, Obtain the characteristic vector of n 500 dimension.
Step 3: the characteristic vector after dimensionality reduction is carried out local feature polymerization and describes son coding;
Specifically, representing the characteristic vector after dimensionality reduction, need to encode further, take here is local feature polymerization Describe son to encode, including following step:
Step 3.1: m characteristic vector after dimensionality reduction in step 2.1 carries out k-means cluster, generates the code of a 100*500 This;
Step 3.2: use local feature polymerization to describe son and each characteristic vector is encoded;
Step 3.3: after using two norm normalization, the character representation of 50000 dimensions after being encoded;
Step 3.4: equally n characteristic vector in step 2.2 is carried out the operation of step 3.1-3.3,50000 after being encoded The character representation of dimension.
Step 4: the characteristic vector after coding is carried out PCA dimensionality reduction, obtains the character representation of current scale;
Specifically, the feature vector dimension after coding is the highest, and the complexity of calculating is at a relatively high, in addition it is also necessary to drop further Dimension, where like taking PCA dimensionality reduction to be dropped to 4096 dimensions, including following step:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as chi The character representation of degree 2;
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as chi The character representation of degree 3;
Step 5: the feature of all yardsticks is cascaded pond, obtains the multiple dimensioned degree of depth convolution character representation of present image pondization;
Specifically, yardstick 1 is represented to the characteristic vector of yardstick 3, contain space and the knot of original image under different scale Structure information, needs cascade pond further to get up, and the characteristic vector of the dimension that pie graph picture is final represents.
Step 6: the pondization multiple dimensioned degree of depth convolution character representation of all vehicle images is used for linear SVM instruction Practice, obtain model recognition system;
Specifically, the pondization multiple dimensioned degree of depth convolution character representation to all vehicle images obtained, take one vs here Rest linear SVM is trained.The detailed process of one vs rest linear SVM training is: when setting original training There is K kind vehicle classification to need to divide, when extracting training set, extract each independent class positive sample as training set respectively Collection, remaining all samples, as negative sample collection, obtain the linear SVM grader of K two classification, test by training Time, corresponding test vector to be tested with this K training result file respectively, each test has a scoring, final recognition result is exactly that classification that score value is the highest, also will vehicle classification to be identified for having That class of maximum classification function value, including following step:
Step 6.1: using the pondization multiple dimensioned degree of depth convolution feature of current class vehicle image as positive sample, other classification vehicles The pondization of image multiple dimensioned degree of depth convolution feature is as negative sample;
Step 6.2: use Linear SVM training the two sample, obtain the grader of current class vehicle image;
Step 6.3: repeat the step 6.1 operation to step 6.2, obtain the grader of all categories vehicle image, structure after associating Become model recognition system.
Step 7: to vehicle to be identified, same acquisition its pondization multiple dimensioned degree of depth convolution character representation, import identification system I.e. may recognize that its vehicle.
Specifically, to vehicle to be identified, obtain the multiple dimensioned degree of depth convolution character representation of its pondization by same step, then Import identification system and i.e. may recognize that its vehicle, including following step:
Step 7.1: to vehicle image to be identified, by the extraction of step 1 to step 5, obtain the pond of vehicle image to be identified Multiple dimensioned degree of depth convolution character representation;
Step 7.2: in the model recognition system trained in the character representation steps for importing 6 that will obtain, identify its vehicle.

Claims (8)

1. a model recognizing method based on pondization multiple dimensioned degree of depth convolution feature, comprises the following steps:
Step 1: each vehicle image to vehicle image data base, extracts its degree of depth convolution feature by different scale;
Step 2: first yardstick does not processes, the degree of depth convolution feature of remaining each yardstick, carry out PCA dimensionality reduction, the spy after dimensionality reduction Levy vector;
Step 3: the characteristic vector after dimensionality reduction is carried out local feature polymerization and describes son coding, the characteristic vector after being encoded;
Step 4: the characteristic vector after coding is carried out PCA dimensionality reduction, obtains the character representation of current scale;
Step 5: the character representation of all yardsticks is cascaded pond, obtains present image pondization multiple dimensioned degree of depth convolution mark sheet Show;
Step 6: the pondization multiple dimensioned degree of depth convolution character representation of all vehicle images is used for linear SVM training, To model recognition system;
Step 7: to vehicle to be identified, same acquisition its pondization multiple dimensioned degree of depth convolution character representation, import identification system Identify its vehicle.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 1, it is characterised in that institute State step 1 and include following step:
Step 1.1: ask for the average image of all images in model data storehouse;
Step 1.2: by original vehicle image scaling to 256*256 size, then deducts average image, then imports degree of depth convolution god Feature extraction is carried out, using 4096 dimensional features of network model's layer 7 as the degree of depth convolution of first yardstick in network model Feature, the most further processes;
Step 1.3: on the original image with window 128*128, step-length 32*32 is extractedmIndividual image block, according in step 1.2 Mode, extractsm4096 dimension degree of depth convolution features of individual image block;
Step 1.4: the window size in step 1.3 is set to 64*64, extracts in the same wayn4096 dimensions of individual image block Degree of depth convolution feature.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 2, it is characterised in that institute State step 2 and include following step:
Step 2.1: by generate in step 1.3mIndividual 4096 dimension degree of depth convolution features, use PCA dimensionality reduction to be dropped to 500 dimensions, ObtainmThe characteristic vector of individual 500 dimensions;
Step 2.2: by generate in step 1.4nIndividual 4096 dimension degree of depth convolution features, use PCA dimensionality reduction to be dropped to 500 dimensions, ObtainnThe characteristic vector of individual 500 dimensions.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 3, it is characterised in that institute State step 3 and include the most several step:
Step 3.1: after dimensionality reduction in step 2.1mIndividual characteristic vector carries out k-means cluster, generates the code of a 100*500 This;
Step 3.2: use local feature polymerization to describe son and each characteristic vector is encoded;
Step 3.3: after using two norm normalization, the character representation of 50000 dimensions after being encoded;
Step 3.4: same in step 2.2nIndividual characteristic vector carries out the operation of step 3.1-3.3,50000 after being encoded The character representation of dimension.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 3, it is characterised in that institute State step 4 and include the most several step:
Step 4.1: to 50000 dimensional feature vectors generated in step 3.3, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as the The character representation of two yardsticks;
Step 4.2: to 50000 dimensional feature vectors generated in step 3.4, uses PCA dimensionality reduction to be dropped to 4096 dimensions, as the The character representation of three yardsticks.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 1, it is characterised in that institute State in step 5 and the feature of all yardsticks is cascaded pond, obtain the multiple dimensioned degree of depth convolution character representation of present image pondization, including Cascade each and every one yardstick of pondization first, to the characteristic vector of the 3rd yardstick, obtains the spy of the final 3*4096 dimension of current vehicle image Levy vector representation.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 1, it is characterised in that institute State step 6 and include following step:
Step 6.1: using the pondization multiple dimensioned degree of depth convolution feature of current class vehicle image as positive sample, other classification vehicles The pondization of image multiple dimensioned degree of depth convolution feature is as negative sample;
Step 6.2: use Linear SVM training the two sample, obtain the grader of current class vehicle image;
Step 6.3: repeat the step 6.1 operation to step 6.2, obtain the grader of all categories vehicle image, structure after associating Become model recognition system.
Model recognizing method based on pondization multiple dimensioned degree of depth convolution feature the most according to claim 1, it is characterised in that institute State step 7 and include following step:
Step 7.1: to vehicle image to be identified, by the extraction of step 1 to step 5, obtain the pond of vehicle image to be identified Multiple dimensioned degree of depth convolution character representation;
Step 7.2: in the model recognition system trained in the character representation steps for importing 6 that will obtain, identify its vehicle.
CN201610295487.XA 2016-05-06 2016-05-06 A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization Expired - Fee Related CN105956560B (en)

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 true CN105956560A (en) 2016-09-21
CN105956560B 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)

Cited By (18)

* Cited by examiner, † Cited by third party
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
CN106599797A (en) * 2016-11-24 2017-04-26 北京航空航天大学 Infrared face identification method based on local parallel nerve network
CN106909938A (en) * 2017-02-16 2017-06-30 青岛科技大学 Viewing angle independence Activity recognition method based on deep learning network
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
CN107133570A (en) * 2017-04-07 2017-09-05 武汉睿智视讯科技有限公司 A kind of vehicle/pedestrian detection method and system
CN107169455A (en) * 2017-05-16 2017-09-15 中山大学 Face character recognition methods based on depth local feature
CN107220657A (en) * 2017-05-10 2017-09-29 中国地质大学(武汉) A kind of method of high-resolution remote sensing image scene classification towards small data set
CN107330463A (en) * 2017-06-29 2017-11-07 南京信息工程大学 Model recognizing method based on CNN multiple features combinings and many nuclear sparse expressions
CN107506729A (en) * 2017-08-24 2017-12-22 中国科学技术大学 A kind of visibility detecting method based on deep learning
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
CN108062754A (en) * 2018-01-19 2018-05-22 深圳大学 Segmentation, recognition methods and device based on dense network image
CN108154502A (en) * 2017-12-22 2018-06-12 王华锋 A kind of though-hole solder joint recognition methods based on convolutional neural networks
CN108154153A (en) * 2016-12-02 2018-06-12 北京市商汤科技开发有限公司 Scene analysis method and system, electronic equipment
CN109102010A (en) * 2018-07-27 2018-12-28 北京以萨技术股份有限公司 A kind of image classification method based on two way blocks structure
CN109410251A (en) * 2018-11-19 2019-03-01 南京邮电大学 Method for tracking target based on dense connection convolutional network
WO2019041360A1 (en) * 2017-09-04 2019-03-07 华为技术有限公司 Pedestrian attribute recognition and positioning method and convolutional neural network system
US11062453B2 (en) 2016-12-02 2021-07-13 Beijing Sensetime Technology Development Co., Ltd. Method and system for scene parsing and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
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

Patent Citations (2)

* Cited by examiner, † Cited by third party
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)

* Cited by examiner, † Cited by third party
Title
X ZENG 等: "Deep Learning of Scene-Specific Classifier for Pedestrian Detection", 《SPRINGER INTERNATIONAL PUBLISHING》 *
ZHEN DONG 等: "Vehicle Type Classification Using Unsupervised Convolutional Neural Network", 《22ND INTERNATIONAL CONFERENCE ON PATTERN RECOGNITON》 *
李梦涵 等: "多尺度级联行人检测算法的研究与实现", 《计算机技术与发展》 *
潘炜深 等: "基于多尺度梯度及深度神经网络的汉子识别", 《北京航空航天大学学报》 *
邓柳 等: "基于深度卷积神经网络的车型识别研究", 《计算机应用研究》 *

Cited By (26)

* Cited by examiner, † Cited by third party
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
CN106599797A (en) * 2016-11-24 2017-04-26 北京航空航天大学 Infrared face identification method based on local parallel nerve 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
CN108154153A (en) * 2016-12-02 2018-06-12 北京市商汤科技开发有限公司 Scene analysis method and system, electronic equipment
US11062453B2 (en) 2016-12-02 2021-07-13 Beijing Sensetime Technology Development Co., Ltd. Method and system for scene parsing and storage medium
CN106909938A (en) * 2017-02-16 2017-06-30 青岛科技大学 Viewing angle independence Activity recognition method based on deep learning network
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
CN107133570A (en) * 2017-04-07 2017-09-05 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
CN107220657A (en) * 2017-05-10 2017-09-29 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
CN107169455A (en) * 2017-05-16 2017-09-15 中山大学 Face character recognition methods based on depth local feature
CN107169455B (en) * 2017-05-16 2020-08-28 中山大学 Face attribute recognition method based on depth local features
CN107330463A (en) * 2017-06-29 2017-11-07 南京信息工程大学 Model recognizing method based on CNN multiple features combinings and many nuclear sparse expressions
CN107506729B (en) * 2017-08-24 2020-04-03 中国科学技术大学 Visibility detection method based on deep learning
CN107506729A (en) * 2017-08-24 2017-12-22 中国科学技术大学 A kind of visibility detecting method based on deep learning
WO2019041360A1 (en) * 2017-09-04 2019-03-07 华为技术有限公司 Pedestrian attribute recognition 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
CN108154502A (en) * 2017-12-22 2018-06-12 王华锋 A kind of though-hole solder joint recognition methods based on convolutional neural networks
CN108062754A (en) * 2018-01-19 2018-05-22 深圳大学 Segmentation, recognition methods and device based on dense network image
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
CN109102010A (en) * 2018-07-27 2018-12-28 北京以萨技术股份有限公司 A kind of image classification method based on two way blocks structure
CN109410251A (en) * 2018-11-19 2019-03-01 南京邮电大学 Method for tracking target based on dense connection convolutional network

Also Published As

Publication number Publication date
CN105956560B (en) 2019-07-09

Similar Documents

Publication Publication Date Title
CN105956560A (en) Vehicle model identification method based on pooling multi-scale depth convolution characteristics
CN106504233B (en) Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN
CN106682598B (en) Multi-pose face feature point detection method based on cascade regression
CN103679158B (en) Face authentication method and device
CN103605972B (en) Non-restricted environment face verification method based on block depth neural network
CN104599275B (en) The RGB-D scene understanding methods of imparametrization based on probability graph model
CN102938065B (en) Face feature extraction method and face identification method based on large-scale image data
CN105718889B (en) Based on GB (2D)2The face personal identification method of PCANet depth convolution model
CN106415594A (en) A method and a system for face verification
CN104484658A (en) Face gender recognition method and device based on multi-channel convolution neural network
CN103514456A (en) Image classification method and device based on compressed sensing multi-core learning
CN104090972A (en) Image feature extraction and similarity measurement method used for three-dimensional city model retrieval
CN108062543A (en) A kind of face recognition method and device
CN104408405A (en) Face representation and similarity calculation method
CN104050460B (en) The pedestrian detection method of multiple features fusion
CN105243139A (en) Deep learning based three-dimensional model retrieval method and retrieval device thereof
CN104200228B (en) Recognizing method and system for safety belt
CN107808129A (en) A kind of facial multi-characteristic points localization method based on single convolutional neural networks
CN107122712B (en) Palm print image identification method based on CNN and bidirectional VLAD
CN104143091A (en) Single-sample face recognition method based on improved mLBP
CN105095880A (en) LGBP encoding-based finger multi-modal feature fusion method
JP2016062610A (en) Feature model creation method and feature model creation device
CN104376312A (en) Face recognition method based on word bag compressed sensing feature extraction
CN104537353A (en) Three-dimensional face age classifying device and method based on three-dimensional point cloud
CN106548195A (en) A kind of object detection method based on modified model HOG ULBP feature operators

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20190709

Termination date: 20200506