CN107220643A - The Traffic Sign Recognition System of deep learning model based on neurological network - Google Patents
The Traffic Sign Recognition System of deep learning model based on neurological network Download PDFInfo
- Publication number
- CN107220643A CN107220643A CN201710236154.4A CN201710236154A CN107220643A CN 107220643 A CN107220643 A CN 107220643A CN 201710236154 A CN201710236154 A CN 201710236154A CN 107220643 A CN107220643 A CN 107220643A
- Authority
- CN
- China
- Prior art keywords
- layer
- traffic sign
- picture
- sign recognition
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/09—Arrangements for giving variable traffic instructions
- G08G1/0962—Arrangements for giving variable traffic instructions having an indicator mounted inside the vehicle, e.g. giving voice messages
-
- 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
- G06V2201/09—Recognition of logos
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of traffic sign recognition method of deep learning model based on neurological network and system, it is adaptable to the detection and identification of the traffic sign in image or video.System includes IMAQ, picture pretreatment, Traffic Sign Recognition, four modules of voice reminder.Picture pretreatment module scales three steps comprising color positioning, SHAPE DETECTION, picture, obtains the picture of same size.Traffic Sign Recognition module carries out Classification and Identification to picture after pretreatment, by the deep neural network model based on neurological network trained, obtains the result of traffic sign Classification and Identification, finally transmits to voice reminder module and carries out voice reminder.The core of the present invention is the deep neural network model based on neurological network, and it has, and model is small, precision is high, computing consumes small, portable to advantages such as cell phone platforms.Whole system has the advantages such as the species of identification traffic sign is more, precision is high, real-time is good.
Description
Technical field
The present invention relates to a kind of computer vision and machine learning techniques, belong to the method for Target detection and identification, specifically
Be related to the traffic sign recognition method and system of a kind of deep learning model based on neurological network, it is adaptable to image or
The detection and identification of traffic sign in video.
Background technology
In recent years, unmanned development is more ripe, and aids in driving to come into practical stage, the identification of traffic sign
It is one of current intelligence auxiliary most important module of DAS (Driver Assistant System), and is the important component of unmanned technology.
Traffic Sign Recognition module generally comprises two aspects of detection and localization and Classification and Identification.
In terms of the positioning of traffic sign, the region that there may be traffic sign can be oriented.It has been fruitful and has used
Method based on color realizes that image is split, and being suitable for the color space of Traffic Sign Images segmentation includes rgb space, HSI skies
Between etc., and use is rgb space to the present invention.
Convenient in the identification of traffic sign, most scholars are using traditional convolutional neural networks identification classification traffic
Mark, but have model big, calculation cost is high, is not suitable for migrating to the defect of mobile platform.
Therefore, the Traffic Sign Recognition module that calculation cost is low, model portable, model small volume, accuracy rate are high is in nothing
People drives and aided in play an important role in driving.
The content of the invention
The invention aims to overcome the existing complicated high, ginseng of Traffic Sign Recognition System learnt based on conventional depth
Number is more, be difficult to migrate to the defects such as mobile platform.
The improved technical problem of the present invention is the computationally intensive of traditional convolutional neural networks, causes to recognize on a mobile platform
A kind of slow-footed problem, it is proposed that traffic sign recognition method of the depth network model based on neurological network.
Technical solution of the present invention includes IMAQ, picture pretreatment, Traffic Sign Recognition, four modules of voice reminder,
Such as Fig. 1.
1. IMAQ part
The system transplantation is obtained to mobile platform (Android platform) by mobile phone camera or vehicle-mounted traveling recorder
Take every two field picture input picture pretreatment module.
2. picture preprocessing part
Picture pretreatment is divided into three modules：Color positioning, SHAPE DETECTION, picture scaling.
Color is positioned：Tentatively traffic sign can be extracted using the color characteristic (red, yellow, Lan Sanse) of traffic sign
Come.
SHAPE DETECTION：On the basis of color positioning, using the shape facility (triangle, circle, rectangle) of traffic sign,
Detect the region comprising traffic sign and interception comes out.
Picture is scaled：For the picture that specification is intercepted, unified the size for 32*32, and input Traffic Sign Recognition
Module, after processing as shown in Figure 2.
3. Traffic Sign Recognition part
The technical scheme of traffic identification module is the deep learning network based on neurological network and learned using migration
Habit adapts it to Traffic Sign Recognition.The design principle of neurological network is to replace 3x3 convolution using 1x1 convolution kernel
Core, reduces by 9 times of parameter inputs, its core component is compact layer, i.e., replace one layer of convolutional layer with compression layer and extension layer, is pressed
Contracting layer is 1x1 convolutional layers, and extension layer is that 1x1 combines obtained combination layer with 3x3.In order to adapt to Traffic Sign Recognition, by Fig. 3
It is improved to the network architecture as shown in Figure 4.
4. voice reminder part
The network has speed fast, small volume, the characteristics of accuracy rate meets basic application standard, it is adaptable to mobile platform,
And among embedded APP, the mode of the traffic sign that Classification and Identification is gone out voice reminder reminds driver.
Advantages and positive effects of the present invention are：
The present invention provide a kind of highway traffic sign of the deep learning model based on neurological network know automatically and
System for prompting, this system carries out transfer learning with neurological network, there is very high real-time and very low operand, can
Cell phone platform is migrated to, just need to can only realize that basic highway traffic sign is recognized and voice reminder using mobile phone, it is to avoid
Accident caused by neglecting in driver's driving conditions.
The early stage image preprocessing of the present invention can reduce the input dimension and image volume of deep learning, and prominent traffic
The feature of mark.
The present invention deep neural network model using training set be GTSRB (German Traffic Sign Recognition benchmark,
German trafficsign recognition benchmark) in training set, include training 39,209, picture, test
12630, picture) when, parameter is than traditional few hundreds of times of model parameter, and the weight parameter file that training is completed only has 4MB left
The right side but reaches 93.5% measuring accuracy.
The present invention has the advantage such as species is more, precision height, real-time are good of identification traffic sign, reduce illumination variation,
The influences of the factor to image recognition such as color fading, motion blur, complicated background, improve antijamming capability, and identification is accurate
Rate is high, and false recognition rate is low.
Brief description of the drawings
The present invention is further described with reference to the accompanying drawings and examples.
The module diagram of Fig. 1 Traffic Sign Recognition Systems of the present invention.
Comparison diagram before and after the picture pretreatment of Fig. 2 present invention.
The compact Rotating fields schematic diagram of Fig. 3 present invention.
The deep neural network Organization Chart based on neurological network of Fig. 4 present invention.
Embodiment
Technical scheme is described in detail with reference to the accompanying drawings and examples.
As illustrated, the present invention the deep learning model based on neurological network traffic sign recognition method and
System includes IMAQ, picture pretreatment, Traffic Sign Recognition, four modules of voice reminder.Wherein, IMAQ is mainly born
Duty collection includes the image of traffic sign；Image pre-processing module is mainly responsible for traffic sign and handle in the image that detection is obtained
Its extracted region comes out, and is carrying out the scaling of uniform sizes；Traffic Sign Recognition module is nucleus module, after transfer learning
Neurological network carry out Traffic Sign Recognition classification；Voice reminder module is responsible for the traffic sign prompting department that will identify that
Machine.
Module one：Image capture module.
The Android phone APP of independent research, the APP is used to use the camera of mobile phone in the embodiment of the present invention
Video recording operation is carried out with the speed of 20-30 frames per second and video file is preserved, and by the picture real-time Transmission of preservation extremely
Pretreatment module.
Module two：Picture pretreatment module.
Pretreatment module of the present invention is divided into three parts, and respectively color positioning, SHAPE DETECTION, picture scales three submodules
Block, and this step carries out picture pretreatment.
Step one：The every frame picture got is subjected to Gaussian Blur processing first, is calculated and schemed using two-dimensional Gaussian function
Piece matrix weight (x, y be periphery coordinate for the relative coordinate of center pixel, σ is blur radius)：
Calculate the Gaussian Blur value of picture：Weight matrix and original colour matrix multiple will be obtained, Gaussian Blur is obtained
Center pixel colour afterwards.
The threshold values of blue three colors of reddish yellow is set, mask is built according to threshold values, and picture is carried out with the picture after Gaussian Blur processing
The bit arithmetic that element is added, binary conversion treatment is carried out with big law, is obtained the body position of blue (red or Huang) color, i.e. color and is determined
Position.
Step 2：The structural element (rectangle, triangle, circle) being consistent with traffic sign shape size is defined first, is obtained
To morphology kernel, closing operation of mathematical morphology (first expanding post-etching), (expansion of computation of morphology gradient are carried out in conjunction with this kernel profit
The difference of figure and etch figures) retain profile.
Finally extract resulting profile and obtain maximum square using polygon (rectangle, triangle is circular) approximate algorithm
The profile of shape, rectangular image is intercepted out using it as standard, and this image includes the traffic mark needed for identification in the case of excluding error
Will.Next step 3：Image is zoomed in and out using bilinear interpolation, unified size is obtained (by four adjacent pixel meters
Calculate)
(Dst is output image, and Src is input picture, for a purpose pixel, sets coordinate to be obtained by reciprocal transformation
The floating-point coordinate arrived is (i+u, j+v), and wherein i, j are the integer part of floating-point coordinate, and u, v are the fractional part of floating-point coordinate
Point, then this pixel obtain value Dst (i+u, j+v) can in input picture coordinate be (i, j), (i+1, j), (i, j+1), (i+1, j
+ 1) value of four pixels is determined around corresponding to)
Dst (i+u, j+v)=(1-u) * (1-v) * Src (i, j)+(1-u) * v*Src (i, j+ 1)+u* (1-v) * Src (i+
1, j)+u*v*Src (i+1, j+1)
Module three：Traffic Sign Recognition module.
Step one：Deep neural network model based on neurological network is built.
The core layer of this model is compact layer, and the convolutional layer obtained by 1x1 convolution kernels (compression layer) connects 1x1 convolution respectively
The convolutional layer that the convolutional layer of core is obtained with 3x3 convolution kernels, and combine this two convolutional layer and be expanded layer, such as scheme.
Because the image that picture pretreatment module is obtained is 32x32 pixels, the picture element matrix for 32x32x3 is inputted.
First layer is convolutional layer, and convolution kernel 3x3, step-length is 1, effectively filling, and obtained convolutional layer is 28x28x36, is used in combination
RELU functions are activated.
The second layer is pond layer, and 2x2 maximums pond obtains 14x14x36 matrixes.
Third layer is compact layer, first with compression layer, 14x14x36 is inputted to 1x1 convolution kernel, depth is 16, step-length
For 1, effectively filling obtains 14x14x16 compression layer, and it is 32 secondly to carry out depth with 1x1 convolution kernels and 3x3 convolution kernels respectively
Extension, be combined and obtain 14x14, x64 Fire layers, and with RELU functions activate.
4th layer is compact layer, and the convolution kernel that third layer is inputted into 1x1 carries out depth and is 64 compression, then passes through respectively
1x1x72,3x3x72 convolution kernel are extended and combined the Fire layers for obtaining 14x14x144, the activation of RELU functions.
Layer 5 is pond layer, and 7x7x144 pond layer is obtained by 2x2 maximum pond.
Layer 6 is convolutional layer, by 1x1 convolution kernel, and depth is 43, and step-length is one, effectively filling, obtains 7x7x43
Compression convolutional layer.
Layer 7 is average pond layer, and 1x1x43 average pond layer is obtained by 7x7x43 average pondization operation.
By tiling operation be converted into 43 1 dimension matrix and be added with biasing and obtain the output of 43 types.
Model framework Fig. 3, step 2：Training pattern.
Training set is GTSRB (German Traffic Sign Recognition benchmark, German trafficsign recognition
Benchmark the training set in), includes 39,209, picture of training, test pictures 12630.
Training set does not simultaneously need picture pretreatment.
Initiation parameter, by random initializtion of the weight of each layer by normal distribution, average is 0, and standard deviation is 0.1,
And stochastic gradient descent learning rate is set as 0.0009, cycle-index is 25.
Training group is 128 samples randomly selected every time.
Training sample (x, y) is respectively the comparison standard of input and result.
Training sample is inputted to the depth network model based on neurological network of above-mentioned structure, obtains final
43 classification results.
Preserve the model that training is completed.
Step 3：Test model.
Test set is used to test the model preserved, the accuracy rate of detection model, and compared with individual conventional model.
Module four：Voice reminder module.
First, connection IMAQ, picture pretreatment, voice reminder module.Secondly, by obtained pretreated image
The model trained is inputted, is obtained a result, and driver is reminded by the voice prompting function of APP softwares.
Finally, driver can in historical record feedback result, for changing Optimized model.
The specific embodiment of the present invention is described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring the substantive content of the present invention.
Claims (6)
1. a kind of Traffic Sign Recognition System of the deep learning model based on neurological network, includes IMAQ, figure
Piece pretreatment, Traffic Sign Recognition, four modules of voice reminder；Picture pretreatment is divided into three modules：Color positioning, shape inspection
Survey, picture is scaled；Traffic identification module is the deep learning network based on neurological network and makes it using transfer learning
Adapt to Traffic Sign Recognition.
2. Traffic Sign Recognition System according to claim 1, it is characterised in that：The system transplantation leads to mobile platform
Cross mobile phone camera or vehicle-mounted traveling recorder obtains and picture pretreatment module is inputted per two field picture.
3. Traffic Sign Recognition System according to claim 2, it is characterised in that：The color positioning：Utilize traffic mark
The color characteristic of will is red, yellow, Lan Sanse tentatively can extract traffic sign；SHAPE DETECTION：The basis positioned in color
On, using the shape facility of traffic sign, detect the region comprising traffic sign and interception comes out；Picture is scaled：In order to advise
The picture of model interception, is unified the size for 32*32, and input the Traffic Sign Recognition module.
4. Traffic Sign Recognition System according to claim 3, it is characterised in that：The Traffic Sign Recognition module bag
Include：Deep neural network model structure, training pattern and test model based on neurological network.
5. Traffic Sign Recognition System according to claim 4, it is characterised in that：The core layer of this model is compact layer,
The convolutional layer obtained by 1x1 convolution kernels connects the convolutional layer that the convolutional layer of 1x1 convolution kernels and 3x3 convolution kernels are obtained, and group respectively
This two convolutional layer is closed to be expanded layer；
First layer is convolutional layer, and convolution kernel 3x3, step-length is 1, effectively filling, and obtained convolutional layer is 28x28x36, and uses RELU
Function is activated；
The second layer is pond layer, and 2x2 maximums pond obtains 14x14x36 matrixes；
Third layer is compact layer, first with compression layer, 14x14x36 is inputted to 1x1 convolution kernel, depth is 16, and step-length is 1,
Effectively filling, obtains 14x14x16 compression layer, secondly carries out the expansion that depth is 32 with 1x1 convolution kernels and 3x3 convolution kernels respectively
Exhibition, is combined and obtains 14x14, x64 Fire layers, and is activated with RELU functions；
4th layer is compact layer, and the convolution kernel that third layer is inputted into 1x1 carries out depth and is 64 compression, then passes through respectively
1x1x72,3x3x72 convolution kernel are extended and combined the Fire layers for obtaining 14x14x144, the activation of RELU functions；
Layer 5 is pond layer, and 7x7x144 pond layer is obtained by 2x2 maximum pond；
Layer 6 is convolutional layer, by 1x1 convolution kernel, and depth is 43, and step-length is one, effectively filling, obtains 7x7x43 pressure
Crinkle lamination；
Layer 7 is average pond layer, and 1x1x43 average pond layer is obtained by 7x7x43 average pondization operation；
By tiling operation be converted into 43 1 dimension matrix and be added with biasing and obtain the output of 43 types.
6. Traffic Sign Recognition System according to claim 5, it is characterised in that：Obtained pretreated image is defeated
Enter the model trained, obtain a result, and driver is reminded by the voice prompting function of APP softwares.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710236154.4A CN107220643A (en) | 2017-04-12 | 2017-04-12 | The Traffic Sign Recognition System of deep learning model based on neurological network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710236154.4A CN107220643A (en) | 2017-04-12 | 2017-04-12 | The Traffic Sign Recognition System of deep learning model based on neurological network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107220643A true CN107220643A (en) | 2017-09-29 |
Family
ID=59927553
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710236154.4A Pending CN107220643A (en) | 2017-04-12 | 2017-04-12 | The Traffic Sign Recognition System of deep learning model based on neurological network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107220643A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108545021A (en) * | 2018-04-17 | 2018-09-18 | 济南浪潮高新科技投资发展有限公司 | A kind of auxiliary driving method and system of identification special objective |
| CN108564048A (en) * | 2018-04-20 | 2018-09-21 | 天津工业大学 | A kind of depth convolutional neural networks method applied to Traffic Sign Recognition |
| CN108647588A (en) * | 2018-04-24 | 2018-10-12 | 广州绿怡信息科技有限公司 | Goods categories recognition methods, device, computer equipment and storage medium |
| CN108681727A (en) * | 2018-07-11 | 2018-10-19 | 天津天瞳威势电子科技有限公司 | A kind of traffic marking recognition methods of view-based access control model and device |
| CN108918532A (en) * | 2018-06-15 | 2018-11-30 | 长安大学 | A kind of through street traffic sign breakage detection system and its detection method |
| CN109002863A (en) * | 2018-06-27 | 2018-12-14 | 佛山市顺德区中山大学研究院 | A kind of compact convolutional neural networks of resourceoriented constrained system |
| CN109300128A (en) * | 2018-09-29 | 2019-02-01 | 聚时科技(上海)有限公司 | The transfer learning image processing method of structure is implied based on convolutional Neural net |
| CN109508635A (en) * | 2018-10-08 | 2019-03-22 | 哈尔滨理工大学 | A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network |
| CN109741484A (en) * | 2018-12-24 | 2019-05-10 | 南京理工大学 | Automobile data recorder and its working method with image detection and voice alarm function |
| CN110175561A (en) * | 2019-05-24 | 2019-08-27 | 上海电机学院 | A kind of detection of road signs and recognition methods |
| CN110647977A (en) * | 2019-08-26 | 2020-01-03 | 北京空间机电研究所 | Method for optimizing Tiny-YOLO network for detecting ship target on satellite |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104134364A (en) * | 2014-07-29 | 2014-11-05 | 电子科技大学 | Real-time traffic signal identification method and system with self-learning capacity |
| US20150067625A1 (en) * | 2012-05-30 | 2015-03-05 | International Business Machines Corporation | Machine-learning based datapath extraction |
| CN104517103A (en) * | 2014-12-26 | 2015-04-15 | 广州中国科学院先进技术研究所 | Traffic sign classification method based on deep neural network |
| CN106504187A (en) * | 2016-11-17 | 2017-03-15 | 乐视控股（北京）有限公司 | Video frequency identifying method and device |
-
2017
- 2017-04-12 CN CN201710236154.4A patent/CN107220643A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150067625A1 (en) * | 2012-05-30 | 2015-03-05 | International Business Machines Corporation | Machine-learning based datapath extraction |
| CN104134364A (en) * | 2014-07-29 | 2014-11-05 | 电子科技大学 | Real-time traffic signal identification method and system with self-learning capacity |
| CN104517103A (en) * | 2014-12-26 | 2015-04-15 | 广州中国科学院先进技术研究所 | Traffic sign classification method based on deep neural network |
| CN106504187A (en) * | 2016-11-17 | 2017-03-15 | 乐视控股（北京）有限公司 | Video frequency identifying method and device |
Non-Patent Citations (2)
| Title |
|---|
| FORREST N. IANDOLA: "SqueezeNet:AlexNet-level accuracy with 50x fewer parameter and <0.5MB model size", 《ARXIV E-PRINTS》 * |
| 贾永红: "《数字图像处理技巧》", 31 January 2017, 武汉：武汉大学出版社 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108545021A (en) * | 2018-04-17 | 2018-09-18 | 济南浪潮高新科技投资发展有限公司 | A kind of auxiliary driving method and system of identification special objective |
| CN108564048A (en) * | 2018-04-20 | 2018-09-21 | 天津工业大学 | A kind of depth convolutional neural networks method applied to Traffic Sign Recognition |
| CN108647588A (en) * | 2018-04-24 | 2018-10-12 | 广州绿怡信息科技有限公司 | Goods categories recognition methods, device, computer equipment and storage medium |
| CN108918532B (en) * | 2018-06-15 | 2021-06-11 | 长安大学 | System and method for detecting damage of expressway traffic sign |
| CN108918532A (en) * | 2018-06-15 | 2018-11-30 | 长安大学 | A kind of through street traffic sign breakage detection system and its detection method |
| CN109002863A (en) * | 2018-06-27 | 2018-12-14 | 佛山市顺德区中山大学研究院 | A kind of compact convolutional neural networks of resourceoriented constrained system |
| CN108681727A (en) * | 2018-07-11 | 2018-10-19 | 天津天瞳威势电子科技有限公司 | A kind of traffic marking recognition methods of view-based access control model and device |
| CN108681727B (en) * | 2018-07-11 | 2024-06-11 | 天津天瞳威势电子科技有限公司 | Traffic sign recognition method and device based on vision |
| CN109300128A (en) * | 2018-09-29 | 2019-02-01 | 聚时科技(上海)有限公司 | The transfer learning image processing method of structure is implied based on convolutional Neural net |
| CN109300128B (en) * | 2018-09-29 | 2022-08-26 | 聚时科技(上海)有限公司 | Transfer learning image processing method based on convolution neural network hidden structure |
| CN109508635A (en) * | 2018-10-08 | 2019-03-22 | 哈尔滨理工大学 | A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network |
| CN109508635B (en) * | 2018-10-08 | 2022-01-07 | 海南师范大学 | Traffic light identification method based on TensorFlow combined with multilayer CNN network |
| CN109741484A (en) * | 2018-12-24 | 2019-05-10 | 南京理工大学 | Automobile data recorder and its working method with image detection and voice alarm function |
| CN110175561A (en) * | 2019-05-24 | 2019-08-27 | 上海电机学院 | A kind of detection of road signs and recognition methods |
| CN110647977A (en) * | 2019-08-26 | 2020-01-03 | 北京空间机电研究所 | Method for optimizing Tiny-YOLO network for detecting ship target on satellite |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107220643A (en) | The Traffic Sign Recognition System of deep learning model based on neurological network | |
| CN110348376B (en) | Pedestrian real-time detection method based on neural network | |
| CN112465748B (en) | Crack identification method, device, equipment and storage medium based on neural network | |
| WO2021238019A1 (en) | Real-time traffic flow detection system and method based on ghost convolutional feature fusion neural network | |
| CN110570396A (en) | industrial product defect detection method based on deep learning | |
| CN112070158B (en) | Facial flaw detection method based on convolutional neural network and bilateral filtering | |
| CN108171701A (en) | Conspicuousness detection method based on U networks and confrontation study | |
| CN104299006A (en) | Vehicle license plate recognition method based on deep neural network | |
| CN111862143B (en) | Automatic monitoring method for river dike collapse | |
| CN112132196B (en) | Cigarette case defect identification method combining deep learning and image processing | |
| CN113807464A (en) | Unmanned aerial vehicle aerial image target detection method based on improved YOLO V5 | |
| CN107506792B (en) | Semi-supervised salient object detection method | |
| CN110991444A (en) | Complex scene-oriented license plate recognition method and device | |
| CN114022408A (en) | Remote sensing image cloud detection method based on multi-scale convolution neural network | |
| CN111062381B (en) | License plate position detection method based on deep learning | |
| CN105405138A (en) | Water surface target tracking method based on saliency detection | |
| CN111209858A (en) | Real-time license plate detection method based on deep convolutional neural network | |
| CN109753996A (en) | Hyperspectral image classification method based on D light quantisation depth network | |
| CN111507275A (en) | Video data time sequence information extraction method and device based on deep learning | |
| CN110827304A (en) | Traditional Chinese medicine tongue image positioning method and system based on deep convolutional network and level set method | |
| CN116071315A (en) | Product visual defect detection method and system based on machine vision | |
| CN116485885A (en) | Method for removing dynamic feature points at front end of visual SLAM based on deep learning | |
| CN113409355A (en) | Moving target identification system and method based on FPGA | |
| CN112308087A (en) | Integrated imaging identification system and method based on dynamic vision sensor | |
| CN103295238B (en) | Video real-time location method based on ROI motion detection on Android platform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170929 |