CN104537393B - A kind of traffic sign recognition method based on multiresolution convolutional neural networks - Google Patents
A kind of traffic sign recognition method based on multiresolution convolutional neural networks Download PDFInfo
- Publication number
- CN104537393B CN104537393B CN201510002850.XA CN201510002850A CN104537393B CN 104537393 B CN104537393 B CN 104537393B CN 201510002850 A CN201510002850 A CN 201510002850A CN 104537393 B CN104537393 B CN 104537393B
- Authority
- CN
- China
- Prior art keywords
- neural networks
- mrow
- convolutional neural
- training
- input
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 4
- 230000000644 propagated effect Effects 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000004913 activation Effects 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract description 10
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000010801 machine learning Methods 0.000 abstract description 4
- 238000001514 detection method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000007935 neutral effect Effects 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 241000165940 Houjia Species 0.000 description 1
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
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
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention belongs to Computer Applied Technology field machine Learning Theory and using subdomains, the Traffic Sign Recognition problem focused in intelligent transport technology.It is characterized in that use a kind of multiresolution convolutional neural networks traffic sign recognition method, it is slower using speed during convolutional neural networks progress Traffic Sign Recognition for solving the problems, such as, input is used as using the two dimensional image of different resolution, the mutually isostructural convolutional neural networks of concurrent operation two carry out Feature Mapping and extraction, and then the weight threshold based on network training carries out precise classification and identification.By the present invention in that it instead of basic CNN structures with two CNN with different resolution branch, full resolution pricture input can map out global and profile feature, the image of low resolution can map out part and minutia, ensure that the resolution ratio of identification, improve model training speed.
Description
Technical field
Focused on the invention belongs to Computer Applied Technology field machine Learning Theory and using subdomains, patent of the present invention
Traffic Sign Recognition problem in intelligent transport technology.It is proposed a kind of multiresolution convolutional neural networks Traffic Sign Recognition side
Method, it is slower using speed during convolutional neural networks progress Traffic Sign Recognition for solving the problems, such as, use different resolution
Two dimensional image carries out Feature Mapping and extraction as input, the mutually isostructural convolutional neural networks of concurrent operation two, then base
Precise classification and identification are carried out in the weight threshold of network training.Both the diversity of the characteristics of image of extraction had been ensure that, had been lifted again
The arithmetic speed of network, has effectively taken into account accuracy of identification and recognition speed.
Background technology
Traffic sign is an important component of road traffic system, and it, which is acted on, mainly includes display current road segment
Traffic, prompt the danger in driving environment and difficult, warning driver, shown the way for driver, for driving for safety and comfort
Offer useful information is provided.If paying attention to by driver oneself completely and finding traffic mark and make correct reaction, unavoidable meeting
Increase drives burden, accelerates fatigue, seriously may result in traffic accident.Therefore, safe and reliable quickly traffic mark is known
Other system seems more and more important, and numerous studies grind the research that personnel are also devoted to this field.
Earliest Traffic Sign Recognition starts from the beginning of the seventies in last century, but after the mid-80, just enters at a high speed
Developing period.
1987, Japan expanded detection, the research of identification to road signs.2000, Japanese Universities research and development
Traffic Sign Recognition System (Miura J, Kanda T, the Shirai Y.An active vision of detectable 60 frames per second
system for real-time traffic sign recognition[C].//Intelligent Transportation
Systems,2000.Proceedings.2000 IEEE.IEEE,2000:52-57.), system that employs two video cameras,
It is separately fixed in automobile, to collect the image with high quality, one is used for coarse positioning, and one is used to gather, and passes through
Threshold segmentation and template matches have reached very high recognition accuracy.
In the early 1990s, Lutz Priese of German Mercedes Benz Co. et al. investigated technology at that time
Automatic recognition of traffic signs system (Priese L, Klieber J, Lakmann R, et al.New results leading
on traffic sign recognition[C]//Intelligent Vehicles'94 Symposium,Proceedings
of the.IEEE,1994:249-254), the system includes CSC (Color Structure Code) and TSR (Traffic
Sign Recognition), achieve certain achievement in research in color coding and shape facility and template identification etc..
2008, the Traffic Sign Recognition System that Mobileye companies develop cooperatively with Continental companies was applied to
The series automobiles of BMW 7, the real-time prison of high accuracy is realized to speed(-)limit sign for the S level automobiles that run quickly, the system again within 2009
Survey, mainly employ the method that front camera and application specific processor are combined.
Traffic Sign Recognition generally comprises two key steps：Detection and identification.In detection-phase, according to the feature of mark
(such as color and shape) is pre-processed and split to image, removes interference of the invalid information to identification, only target may be deposited
Region handled, reduce amount of calculation.In cognitive phase, to target region, feature is extracted with different methods, and
These regions are classified with suitable sorting algorithm, obtain the type information of traffic mark.
Currently, the algorithm of conventional road traffic sign detection mainly has three classes:Method based on color, the method based on shape
With the method being combined based on color with shape.
Method based on color is most widely used, and its main algorithm mainly has four kinds according to the difference of color space：
The Threshold Segmentation Algorithm of RGB color space；The Threshold Segmentation Algorithm of HIS color spaces；The Threshold segmentation of HSV color spaces is calculated
Method；In addition to above-mentioned three kinds of conventional color spaces, also YCBCr color spaces and LAB color spaces etc. is of little use colour
Space.
Because traffic sign typically has fixed and simple shape, the detection algorithm based on shape facility is also studied people
Member's extensive use, conventional method have：Template matching method；HOUGH converter techniques；Corner detection method；Neutral net；Mathematical morphology
Method etc..
Detection based on color and the detection based on shape respectively have advantage and disadvantage, and the detection based on color is simple, and amount of calculation is small,
But easily by such environmental effects such as illumination, rainy weather, picture noises.Shandong based on the method for shape to external environment conditions such as illumination
Rod is more preferable, and the shape of every country traffic sign is substantially similar, and the versatility of this method is stronger, the method based on shape
Shortcoming be easily by the similar object contributions of shape, such as square window, billboard etc. and when apart from traffic sign farther out when,
It is not easy to extract shape.Method that therefore, many researchers are combined using color and shape feature (Torresen J,
Bakke J W,Sekanina L.Efficient recognition of speed limit signs[C]//
Intelligent Transportation Systems.2004:652-656.) determine final result.
The conventional method of traffic mark identification substantially has two classes：One kind is feature extraction algorithm, and another kind of is machine learning
Algorithm.Typically have in feature extraction algorithm：Principal component analytical method PCA；Gabor characteristic extraction algorithm；The small bauds of Haar
Levy extraction algorithm；Like-Fenton Oxidation extraction algorithm etc..The main algorithm that machine learning presently, there are has：KNN algorithms；Decision tree；
Neuroid；Support vector machines；Adaboost algorithm etc..
With the development of neutral net, in recent years, convolutional neural networks (LeCun Y, Bengio based on deep learning
Y.Convolutional networks for images,speech,and time series[J].The handbook of
Brain theory and neural networks, 1995,3361.) it is widely used in voice, image and natural language
Identification, and achieve good achievement.And in Traffic Sign Recognition field,The how vertical convolution that D et al. is proposed
Neutral net (Traffic Sign Recognition based on more longitudinal degree neutral nets,D,Meier U,Masci J,et
al.Multi-column deep neural network for traffic sign classification[J].Neural
Networks,2012,32:333-338.) and the multiple dimensioned convolutional neural networks that propose of Sermanet P and LeCun Y et al.
(Traffic Sign Recognition Sermanet P, the LeCun Y.Traffic sign based on multiple dimensioned convolutional neural networks
recognition with multi-scale convolutional networks[C]//Neural Networks
(IJCNN),The 2011 International Joint Conference on.IEEE,2011:2809-2813.), fully
Extraction characteristics of image, achieve higher discrimination, achieved very in GTSRB (German Traffic Sign Recognition benchmark) contest
Good achievement, has been even more than the discrimination of the mankind.
Although convolutional neural networks achieve good effect on recognition accuracy, its arithmetic speed becomes system
About the network Development an important factor for, although computer hardware is being constantly updated in recent years, arithmetic speed is constantly being lifted, instruction
Practice the time that a high network model of the degree of accuracy still needs to dozens of hour.
The content of the invention
The technical problem to be solved in the present invention is can not to be taken into account using convolutional neural networks when carrying out Traffic Sign Recognition
A kind of the problem of speed and accuracy rate, it is proposed that traffic sign recognition method based on multiresolution convolutional neural networks.
Technical scheme is as follows：
Multiresolution convolutional neural networks are proposed, this is a kind of method for accelerating network, and the method taken in the past is usually
Reduce the quantity of hidden layer and neuron；Network performance may so be reduced.Instead, we are not using reduction network
The method of size, but the picture of low resolution is used, still, it be have lost while so improving the speed of service in picture
High frequency detail and the accuracy for reducing network.Therefore, as Fig. 2, the present invention have carried out the improvement in structure to original CNN, made
Original CNN structures are instead of with the CNN of Liang Ge branches, full resolution pricture input maps out global and profile feature, and low
The image of resolution ratio maps out part and minutia, ensure that the resolution ratio of identification, during training, is carried out simultaneously using Liang Ge branches
Row computing, the input of low resolution cause the operand of network to reduce, improve the speed of computing.
The beneficial effects of the invention are as follows model training speed on the premise of it ensure that recognition accuracy, is improved, pass through
Experiment, identification error rate of the invention can reach 6.08%, than original CNN structures in the case of training iteration 100 times
6.19% slightly improves.On training speed, the multiple dimensioned convolutional neural networks time mentioned above is 37 hours, of the invention
Training time is only with 5.8 hours.
Whole model is divided into image preprocessing, MRCNN network trainings, accuracy rate checking and image recognition four-stage.1、
Image pre-processing phase
Advance processing is carried out to the image in all training sets, is required according to network training, training picture is entered first
Row normalization, is handled as gray level image, is unified for A*A resolution ratio as an input, and it is B*B to intercept this image resolution ratio
Core as another input, such as Fig. 1.What the multiresolution convolutional neural networks that the present invention uses performed is to have supervision
Training, so vector is to composition one by one for its sample set, corresponding one preferable output of an input.Before training is started,
All weights are all initialized with some different small random numbers.
2nd, the MRCNN network trainings stage
(1) the propagated forward stage
1. two samples are taken from sample set to (xh, yh)、(xl, yl), by xh、xlRespectively as two convolutional neural networks
Input.
2. two convolutional neural networks carry out first time convolution respectively.Convolution process includes：With a trainable filtering
Device fxDeconvolute the image of input, and the first stage is the image of input, and the stage below is exactly convolution feature map, Ran Houjia
One biasing bx, obtain convolutional layer C1。
3. by convolutional layer C1Obtained characteristic pattern is sampled as input.Sampling process includes：Per four pixels of neighborhood
Summation is changed into a pixel, then passes through scalar Wx+1Weighting, it is further added by biasing bx, letter is then activated by a sigmoid
Number, produce a Feature Mapping figure S for probably reducing four timesx+1, obtain the layer S that sample2。
4. repeat the C that the 2nd, 3 steps respectively obtain each convolutional neural networks3Layer and S4Layer.
5. calculate corresponding reality output Oh、Ol.In the propagated forward stage, information is by input layer, hidden layer
Convolution and sampling, be ultimately delivered to output layer.It is exported is represented with formula (1)：
O=Fn(......(F2(F1(x*w1)w2)......)wn) (1)
(2) the back-propagation stage
1. calculate reality output Oh、OlWith corresponding preferable output Yh、YlDifference；
2. the weight matrix of two convolutional networks is reversely successively adjusted respectively by the method for minimization error.Here by EpDefinition
For the error of p-th of sample, then the error of whole sample set can use formula (2) to represent：
So far 1 iteration is completed, and preserves the weights after adjustment, repeats the first stage, until reaching the iterations of setting
Afterwards, training terminates, and preserves and this weights is exported to checking collection to verify.
3rd, accuracy rate Qualify Phase
Using 1/5th data of training set verify the weights of input, will input and the result of mapping is carried out pair
Than that output error rate, if error rate reaches satisfied requirement, test identification can be carried out, otherwise, adjust input picture
Resolution ratio or increase iterations re-start training.
4th, the image recognition stage
By the Traffic Sign Images detected after pretreatment, it is identified using the network model for training weights
And export its corresponding classification.
A kind of multiresolution convolutional neural networks traffic sign recognition method of the present invention, for solving to use convolutional Neural
Network carries out the problem of speed is slower during Traffic Sign Recognition, using the two dimensional image of different resolution as input, parallel fortune
Calculate two mutually isostructural convolutional neural networks and carry out Feature Mapping and extraction, then the weight threshold based on network training is carried out
Precise classification and identification.By the present invention in that basic CNN structures are instead of with two CNN with different resolution branch,
Full resolution pricture input can map out global and profile feature, and the image of low resolution can map out part and minutia,
The resolution ratio of identification is ensure that, improves model training speed.
Brief description of the drawings
Fig. 1 is pretreated picture example.
Fig. 2 is multiresolution convolutional neural networks schematic diagram.
Fig. 3 is the Traffic Sign Recognition schematic diagram based on multiresolution convolutional neural networks.
Fig. 4 is applied to the convolutional network schematic diagram of Traffic Sign Recognition.
Fig. 5 a are influence schematic diagram of the training samples number to error.
Accompanying drawing 5b is the influence schematic diagram for training iterations to error.
Embodiment
1st, training set is determined, that the present invention selects is GTSRB (German Traffic Sign Recognition benchmark, German traffic
Sign recognition benchmark) in training set, comprising training 39,209, picture, test pictures 12630 are opened.
2nd, the picture in training set is pre-processed, step is to be treated as gray-scale map first, then normalized
For the picture that unified resolution ratio is 48*48 sizes, then these picture backup portions are cut into central area, the resolution ratio cut
For 36*36 picture, accompanying drawing 1 is two kinds of different resolution input samples of example, using this two parts picture as two inputs
Start to train.
3rd, training group is selected；50 samples are randomly chosen respectively every time from sample set as training group.
4th, by each weights vij, wjkAnd threshold valueθk, be arranged to it is small close to 0 random value, and initialize precision control
Parameter ε and learning rate α processed and convolution kernel size.
5th, an input (x is taken from training grouph, yh)、(xL,yl), network is added to, and give their target output vector
Og、Ol。
6th, network carries out first time convolution operation, and convolution is carried out to two input pictures using 5 × 5 convolution kernel, and together
The width convolution characteristic patterns of Shi Shengcheng 6, as shown in figure 4, the size of every width characteristic pattern of two networks is changed into 44*44 and 32*32.
7th, network performs sample size, using the method for mean sample, uses 2 × 2 sample size, generation 18 width sampling
Characteristic pattern, as shown in figure 4, the size of every width characteristic pattern of two networks is changed into 22*22 and 18*18.
8th, continued executing with according to the method for the 6th step and the 7th step, until the network number of plies as defined in reaching, present invention experiment is used
The number of plies be 6 layers, including an input layer and an output layer and four hidden layers.
6th, by the element (y in output vectorh、yl) with target vector in element (oh、ol) be compared, and calculate M
The error of output item, use formula (3) and (4)：
δh=(oh-yh)yh(1-yh) (3)
δl=(ol-yl)yl(1-yl) (4)
7th, the adjustment amount of each weights is calculated successively and the adjustment amount of threshold value uses formula (5) and formula (6)
ΔWjk(n)=(α/(1+L)) * (Δ Wjk(n-1)+1)*δk*hj
ΔVij(n)=(α/(1+N)) * (Δ Vij(n-1)+1)*δi*hj (5)
Δθk(n)=(α/(1+L)) * (Δ θk(n-1)+1)*δk
8th, weights and threshold value are adjusted according to evaluation.
9th, after k often undergoes 1 to M, whether required precision is met using checking collection judge index：E≤ε, wherein E are total miss
Difference function, ε are the Accuracy Controlling Parameters of setting.
10th, training terminates, and weights and threshold value are preserved hereof.At this moment it is considered that each weights have reached steady
Fixed, grader is formed.
11st, in training, the quantity and iterations of training sample often produce a very large impact to final accuracy rate, such as
Shown in Fig. 5 a and Fig. 5 b, therefore big-sample data should be chosen as far as possible and as training set and reach certain repetitive exercise number.
Claims (1)
- A kind of 1. traffic sign recognition method based on multiresolution convolutional neural networks, it is characterised in that step：(1) image pre-processing phaseAdvance processing is carried out to the image in all training sets, is required according to network training, training picture is returned first One changes, and handles as gray level image, and the resolution ratio for being unified for A*A inputs as one, and intercepts this image resolution ratio as in B*B Center portion is allocated as inputting for another；What the multiresolution convolutional neural networks that use performed is the training for having supervision, its sample set By vector one by one to forming, corresponding one preferable output of an input；Before training is started, all weights are all with one A little different small random numbers are initialized；(2) the MRCNN network trainings stage1) the propagated forward stage1. two samples are taken from sample set to (xh,yh)、(xl,yl), by xh、xlRespectively as the defeated of two convolutional neural networks Enter；2. two convolutional neural networks carry out first time convolution respectively；Convolution process includes：With a trainable wave filter fxGo The image of convolution input, then plus one biases bx, obtain convolutional layer C1Feature map；3. by convolutional layer C1Obtained feature map is sampled as input；Sampling process includes：Per four pixel summations of neighborhood It is changed into a pixel, then passes through scalar Wx+1Weighting, it is further added by biasing bx, then pass through a sigmoid activation primitive, production The raw one Feature Mapping figure S for probably reducing four timesx+1, obtain the layer S that sample2；4. repeating the, 2., 3. step respectively obtains the C of each convolutional neural networks3Layer and S4Layer；5. calculate corresponding reality output Oh、Ol；In the propagated forward stage, information by input layer, hidden layer convolution and take out Sample, it is ultimately delivered to output layer；It is exported is represented with formula (1)：O=Fn(……(F2(F1(x*w1)w2)……)wn) (1)2) the back-propagation stage1. calculate reality output Oh、OlWith corresponding preferable output Yh、YlDifference；2. the weight matrix of two convolutional networks is reversely successively adjusted respectively by the method for minimization error；Here by EpIt is defined as pth The error of individual sample, the then error of whole sample set formula (2) expression：<mrow> <msub> <mi>E</mi> <mi>p</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </msubsup> <msup> <mrow> <mo>(</mo> <msub> <mi>y</mi> <mrow> <mi>p</mi> <mi>j</mi> </mrow> </msub> <mo>-</mo> <msub> <mi>o</mi> <mrow> <mi>p</mi> <mi>j</mi> </mrow> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>So far 1 iteration is completed, and preserves the weights after adjustment, repeats the first stage, after reaching the iterations of setting, instruction White silk terminates, and preserves and this weights is exported to checking collection to verify；(3) accuracy rate Qualify PhaseUsing 1/5th data of training set verify the weights of input, will input and the result of mapping is contrasted, Output error rate, if error rate reaches satisfied requirement, test identification is carried out, otherwise, adjust the resolution ratio of input picture Or increase iterations re-starts training；(4) the image recognition stageBy the Traffic Sign Images detected after pretreatment, it is identified using the network model for training weights and defeated Go out its corresponding classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510002850.XA CN104537393B (en) | 2015-01-04 | 2015-01-04 | A kind of traffic sign recognition method based on multiresolution convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510002850.XA CN104537393B (en) | 2015-01-04 | 2015-01-04 | A kind of traffic sign recognition method based on multiresolution convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104537393A CN104537393A (en) | 2015-04-22 |
| CN104537393B true CN104537393B (en) | 2018-01-16 |
Family
ID=52852913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510002850.XA Expired - Fee Related CN104537393B (en) | 2015-01-04 | 2015-01-04 | A kind of traffic sign recognition method based on multiresolution convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104537393B (en) |
Families Citing this family (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850845B (en) * | 2015-05-30 | 2017-12-26 | 大连理工大学 | A kind of traffic sign recognition method based on asymmetric convolutional neural networks |
| CN104978586A (en) * | 2015-06-29 | 2015-10-14 | 百度在线网络技术（北京）有限公司 | Trademark identifying method and apparatus |
| CN104992165A (en) * | 2015-07-24 | 2015-10-21 | 天津大学 | Extreme learning machine based traffic sign recognition method |
| US10373019B2 (en) | 2016-01-13 | 2019-08-06 | Ford Global Technologies, Llc | Low- and high-fidelity classifiers applied to road-scene images |
| CN105512646B (en) * | 2016-01-19 | 2019-03-01 | 腾讯科技（深圳）有限公司 | A kind of data processing method, device and terminal |
| CN111310904B (en) * | 2016-04-29 | 2024-03-08 | 中科寒武纪科技股份有限公司 | Apparatus and method for performing convolutional neural network training |
| CN106339719A (en) * | 2016-08-22 | 2017-01-18 | 微梦创科网络科技（中国）有限公司 | Image identification method and image identification device |
| CN106897682A (en) * | 2017-02-15 | 2017-06-27 | 电子科技大学 | Leucocyte automatic identifying method in a kind of leukorrhea based on convolutional neural networks |
| US11620482B2 (en) | 2017-02-23 | 2023-04-04 | Nokia Technologies Oy | Collaborative activation for deep learning field |
| CN107066586B (en) * | 2017-04-17 | 2019-07-05 | 清华大学深圳研究生院 | Shoes model index management method and system |
| US10922556B2 (en) | 2017-04-28 | 2021-02-16 | Intel Corporation | Storage system of DNN outputs for black box |
| CN109118459B (en) | 2017-06-23 | 2022-07-19 | 南开大学 | Image salient object detection method and device |
| CN107316066B (en) * | 2017-07-28 | 2021-01-01 | 北京工商大学 | Image classification method and system based on multi-channel convolutional neural network |
| US10474925B2 (en) | 2017-07-31 | 2019-11-12 | Industrial Technology Research Institute | Deep neural network with side branches for recognizing and classifying media data and method for using the same |
| CN107463927A (en) * | 2017-09-21 | 2017-12-12 | 广东工业大学 | A kind of deceleration driven detection method and device based on convolutional neural networks |
| CN107679577A (en) * | 2017-10-12 | 2018-02-09 | 理光图像技术（上海）有限公司 | image detecting method based on deep learning |
| CN107766932B (en) * | 2017-10-19 | 2021-06-04 | 珠海格力电器股份有限公司 | Image processing method and device based on neural network |
| CN107644221A (en) * | 2017-10-31 | 2018-01-30 | 南京航空航天大学 | Convolutional neural networks traffic sign recognition method based on compression of parameters |
| CN107704859A (en) * | 2017-11-01 | 2018-02-16 | 哈尔滨工业大学深圳研究生院 | A kind of character recognition method based on deep learning training framework |
| CN108154102B (en) * | 2017-12-21 | 2021-12-10 | 安徽师范大学 | Road traffic sign identification method |
| CN108416348A (en) * | 2018-01-29 | 2018-08-17 | 重庆邮电大学 | Plate location recognition method based on support vector machines and convolutional neural networks |
| JP6833750B2 (en) * | 2018-03-19 | 2021-02-24 | 株式会社東芝 | Recognition devices, vehicle systems and programs |
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN110909578A (en) * | 2018-09-18 | 2020-03-24 | 中国移动通信有限公司研究院 | Low-resolution image recognition method and device and storage medium |
| CN109460974B (en) * | 2018-10-29 | 2021-09-07 | 广州皓云原智信息科技有限公司 | Attendance system based on gesture recognition |
| CN109635141B (en) | 2019-01-29 | 2021-04-27 | 京东方科技集团股份有限公司 | Method, electronic device, and computer-readable storage medium for retrieving an image |
| US10976412B2 (en) * | 2019-02-01 | 2021-04-13 | GM Global Technology Operations LLC | Deep learning for super resolution in a radar system |
| CN110018804B (en) * | 2019-04-12 | 2021-02-02 | 京东方科技集团股份有限公司 | Display device, image display method and electronic equipment |
| CN110084203B (en) * | 2019-04-29 | 2021-03-05 | 北京航空航天大学 | Full convolution network airplane level detection method based on context correlation |
| CN110188705B (en) * | 2019-06-02 | 2022-05-06 | 东北石油大学 | Remote traffic sign detection and identification method suitable for vehicle-mounted system |
| CN110321803B (en) * | 2019-06-10 | 2022-08-30 | 南京邮电大学 | Traffic sign identification method based on SRCNN |
| CN112116060B (en) * | 2019-06-21 | 2023-07-25 | 杭州海康威视数字技术股份有限公司 | Network configuration implementation method and device |
| CN111222646B (en) * | 2019-12-11 | 2021-07-30 | 深圳逻辑汇科技有限公司 | Design method and device of federal learning mechanism and storage medium |
| CN111368909B (en) * | 2020-03-03 | 2021-05-11 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
| CN111325414A (en) * | 2020-03-23 | 2020-06-23 | 哈尔滨理工大学 | Parallel traffic prediction method based on data driving |
| WO2021253148A1 (en) * | 2020-06-15 | 2021-12-23 | Intel Corporation | Input image size switchable network for adaptive runtime efficient image classification |
| CN111914797B (en) * | 2020-08-17 | 2022-08-12 | 四川大学 | Traffic sign identification method based on multi-scale lightweight convolutional neural network |
| CN112259218A (en) * | 2020-09-29 | 2021-01-22 | 垒途智能教科技术研究院江苏有限公司 | Training method for auditory stimulation of infantile autism based on VR interaction technology |
| CN113536942B (en) * | 2021-06-21 | 2024-04-12 | 上海赫千电子科技有限公司 | Road traffic sign recognition method based on neural network |
| CN113436708B (en) * | 2021-07-22 | 2022-10-25 | 杭州电子科技大学 | Delayed CT image generation method based on deep learning algorithm |
| CN116404760B (en) * | 2023-06-09 | 2023-10-20 | 西安新视空间信息科技有限公司 | Distributed power grid transient steady-state operation method and device based on digital twin map |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102262728A (en) * | 2011-07-28 | 2011-11-30 | 电子科技大学 | Road traffic sign identification method |
| US8582807B2 (en) * | 2010-03-15 | 2013-11-12 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
-
2015
- 2015-01-04 CN CN201510002850.XA patent/CN104537393B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8582807B2 (en) * | 2010-03-15 | 2013-11-12 | Nec Laboratories America, Inc. | Systems and methods for determining personal characteristics |
| CN102262728A (en) * | 2011-07-28 | 2011-11-30 | 电子科技大学 | Road traffic sign identification method |
Non-Patent Citations (2)
| Title |
|---|
| Large-scale Video Classification with Convolutional Neural Networks;Andrej Karpathy etc;《IEEE Conference on Computer Vision and Pattern Recognition》;20141231;第1725-1732页 * |
| 基于CNN的字符识别方法研究;王强;《中国优秀硕士学位论文全文数据库 信息科技辑》;20141115(第11期);第8-22页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104537393A (en) | 2015-04-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104537393B (en) | A kind of traffic sign recognition method based on multiresolution convolutional neural networks | |
| CN108615010B (en) | Facial expression recognition method based on parallel convolution neural network feature map fusion | |
| CN105975931B (en) | A kind of convolutional neural networks face identification method based on multiple dimensioned pond | |
| CN110298396A (en) | Hyperspectral image classification method based on deep learning multiple features fusion | |
| CN104392463B (en) | Image salient region detection method based on joint sparse multi-scale fusion | |
| CN103605972B (en) | Non-restricted environment face verification method based on block depth neural network | |
| CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
| CN108388896A (en) | A kind of licence plate recognition method based on dynamic time sequence convolutional neural networks | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| CN113011357B (en) | Depth fake face video positioning method based on space-time fusion | |
| CN107016405A (en) | A kind of insect image classification method based on classification prediction convolutional neural networks | |
| CN110097145A (en) | One kind being based on CNN and the pyramidal traffic contraband recognition methods of feature | |
| CN105956626A (en) | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method | |
| CN106407986A (en) | Synthetic aperture radar image target identification method based on depth model | |
| CN105138993A (en) | Method and device for building face recognition model | |
| CN110399821B (en) | Customer satisfaction acquisition method based on facial expression recognition | |
| CN109635784A (en) | Traffic sign recognition method based on improved convolutional neural networks | |
| CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
| CN105335716A (en) | Improved UDN joint-feature extraction-based pedestrian detection method | |
| CN107657233A (en) | Static sign language real-time identification method based on modified single multi-target detection device | |
| CN107145830A (en) | Hyperspectral image classification method with depth belief network is strengthened based on spatial information | |
| CN106156765A (en) | safety detection method based on computer vision | |
| CN107145893A (en) | A kind of image recognition algorithm and system based on convolution depth network | |
| CN108520212A (en) | Method for traffic sign detection based on improved convolutional neural networks | |
| CN107633229A (en) | Method for detecting human face and device based on convolutional neural networks |
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: 20180116 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |