CN109376738A - Altering detecting method based on vehicle VIN code start-stop symbol type - Google Patents
Altering detecting method based on vehicle VIN code start-stop symbol type Download PDFInfo
- Publication number
- CN109376738A CN109376738A CN201811105641.8A CN201811105641A CN109376738A CN 109376738 A CN109376738 A CN 109376738A CN 201811105641 A CN201811105641 A CN 201811105641A CN 109376738 A CN109376738 A CN 109376738A
- Authority
- CN
- China
- Prior art keywords
- vin code
- stop
- stop symbol
- model
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000001514 detection method Methods 0.000 claims abstract description 47
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 28
- 238000013135 deep learning Methods 0.000 claims description 18
- 238000005286 illumination Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000007619 statistical method Methods 0.000 claims description 7
- 238000007689 inspection Methods 0.000 claims description 6
- 238000012986 modification Methods 0.000 claims description 6
- 230000004048 modification Effects 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 claims description 2
- 238000000151 deposition Methods 0.000 claims 1
- 238000012360 testing method Methods 0.000 abstract description 9
- 238000012550 audit Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 206010016256 fatigue Diseases 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000011895 specific detection Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses the altering detecting methods based on vehicle VIN code start-stop symbol type, comprising the following steps: obtains the VIN code archival image in the vehicle VIN code image and server to be detected in vehicle annual test；VIN code region is detected, judges that target whether there is；Two pieces of regions of VIN code head and the tail are obtained, detects in the two regions and is accorded with the presence or absence of start-stop, start-stop accords with if it exists, then judges the type of start-stop symbol, and if it does not exist, then type is set as empty；Judge whether the start-stop symbol of vehicle VIN code to be detected is consistent with the start-stop symbol of server archival image；On the contrary if it exists, unanimously, then recording this mark is 0 for the above judgement, then to record this mark be 1, and saves picture concerned；If record mark is 0, vehicle VIN code tampering detection passes through, and shows that VIN code is not distorted, if record mark is that 1, VIN code tampering detection does not pass through, shows that VIN code is tampered.Meanwhile unacceptable reason and problem picture are verified according to the position acquisition that mark 1 occurs.The present invention realizes the automatic Verification of VIN code in vehicle annual test, and existing manual examination and verification mode is substituted, has saved manpower, accelerates audit speed, and can be effectively detected whether VIN code is tampered, and ensure that the disclosure of examination, just.
Description
Technical field
The present invention relates to the artificial intelligence judgment technology fields of automotive vehicle annual test, in particular to a kind of to be based on vehicle
The altering detecting method of VIN code start-stop symbol type.
Background technique
Constantly improve with living standards of the people with the continuous social and economic development, Urban vehicles poputation rapidly increases
It is long.The workload of automotive vehicle annual test also increases rapidly therewith.Vehicle VIN code tampering detection is main in traditional vehicle annual test
It is by artificial detection, the case where distorting for vehicle VIN code, general survey personnel are difficult with the naked eye to go to differentiate, and influence to verify
Accuracy rate.
How accurately and rapidly vehicle VIN code to be verified, while avoiding desk checking at high cost, fatiguability is easily dredged
The drawbacks such as suddenly, are technical problems urgently to be solved.
Summary of the invention
The purpose of the present invention is: propose a kind of altering detecting method based on vehicle VIN code start-stop symbol type, it is automatic to audit
Whether vehicle VIN code is consistent with server archive content, to meet nowadays the needs of to vehicle annual test working efficiency, accuracy rate.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of altering detecting method based on vehicle VIN code start-stop symbol type, comprising the following steps:
Archive VIN code image in S1, acquisition vehicle VIN code image to be detected and server；
S2, using the target detection model inspection VIN code image based on deep learning, judge that VIN code region whether there is,
Then recording this mark if it exists is 0, extracts VIN code region；Then recording this mark if it does not exist is 1, and saves related figure
Piece, into statistical analysis process；
S3, two regions of VIN code head and the tail are obtained, using the two regions of target detection model inspection based on deep learning
It is accorded with the presence or absence of start-stop；
S4, if it does not exist start-stop accord with, then output type is sky；Start-stop accords with if it exists, then using point based on deep learning
Class model judges that start-stop accords with type, judges whether image to be detected and the start-stop of archival image symbol are consistent, record this if consistent
Item mark is 0；It is 1 that this mark is recorded if inconsistent, and saves picture concerned, into statistical analysis process；
It is S5, for statistical analysis to the result of the action of whole process, flag bit all 0 is recorded, then VIN code tampering detection
Pass through, if it exists mark 1, then VIN code tampering detection does not pass through；Meanwhile the position acquisition verification occurred according to mark 1 does not pass through
The reason of and problem picture.
Judge that VIN code region whether there is in the step S2 and use following method:
The classification information of vehicle VIN code image object to be detected is obtained using Softmax, uses bounding box
Regression obtains the location information of vehicle VIN code image object to be detected.
Vehicle VIN code target detection model obtaining step in the step S2 based on deep learning is as follows:
S21, training data prepare: obtaining the vehicle VIN code image of different shooting conditions (such as illumination, angle)；
S22, data mark: vehicle VIN code region is marked in the picture using rectangle frame, rectangle frame region domestic demand is complete
Include vehicle VIN code；
S23, model training: it utilizes and has used the trained VGG basic model of ImageNet, the vehicle that will have been marked
VIN code image inputs in SSD frame, is finely adjusted on basic model, training vehicle VIN code target detection model.
Described be finely adjusted on basic model includes the following steps:
S231. the mean value file of vehicle VIN code data set is calculated；
S232. the last output of modification SSD frame；
Basic studies rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005, learned by S233. regularized learning algorithm rate
Habit rate strategy is set as " multistep ", and gamma is set as 0.1, momentum and is set as 0.9；
S234. load grounding model is finely adjusted.
Start-stop symbol target detection model obtaining step in the step S3 based on deep learning is as follows:
S31, training data prepare: obtaining the start-stop symbol image of different shooting conditions (such as illumination, angle)；
S32, data mark: start-stop symbol region is marked in the picture using rectangle frame；
S33, model training: it utilizes and has used the trained VGG basic model of ImageNet, the vehicle that will have been marked
VIN code image inputs in SSD frame, is finely adjusted on basic model, to preferably start-stop be trained to accord with target detection model.
Described be finely adjusted on basic model includes the following steps:
S231. the mean value file of vehicle VIN code data set is calculated；
S232. the last output of modification SSD frame；
Basic studies rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005, learned by S233. regularized learning algorithm rate
Habit rate strategy is set as " multistep ", and gamma is set as 0.1, momentum and is set as 0.9；
S234. load grounding model is finely adjusted.
Start-stop symbol classification of type model obtaining step in the step S4 based on deep learning is as follows:
S41, the vehicle VIN code start-stop for obtaining different angle, illumination, type and picture quality accord with image；
S42, classification of type deep learning network model is accorded with using start-stop symbol image training start-stop, obtains start-stop symbol type point
Class model.
Start-stop accords with classification of type model and uses this kind of convolutional neural networks of SSD, and uses VGG network as feature extraction
Device,
Classification of type model is accorded with the start-stop symbol image training start-stop in image to be detected and server images first, will be risen
Only the character in symbol is divided into 14 kinds, is indicated respectively with 0~13,
Then classified using the model to start-stop to be detected symbol image,
Finally classification results are obtained with Softmax.
It includes 3 convolutional layers, 2 pond layers and 2 full articulamentums that start-stop, which accords with classification of type model,.
The beneficial effects of the present invention are: present invention is mainly applied to vehicle VIN code tampering detection in automotive vehicle annual test,
It realizes the automatic Verification that VIN code is distorted, while unsanctioned verification image and reason can be passed back to server preservation and stayed
Wait collect evidence.Both manpower has been saved, has in turn ensured the just, openly of verifying work.
Detailed description of the invention
Fig. 1 is vehicle VIN code tampering detection flow chart of the invention.
Fig. 2 is structural schematic diagram of the invention.
Fig. 3 is the structural schematic diagram of object detection unit of the present invention.
Fig. 4 is the structural schematic diagram of start-stop symbol classification of type unit of the present invention.
Specific embodiment
Below in conjunction with attached drawing, the present invention will be further described.
Present invention is primarily based on module of target detection, classification of type module and determination modules.
As shown in Fig. 2, module of target detection accords with object detection unit group by vehicle VIN code object detection unit and start-stop
At.Firstly, image is passed to vehicle VIN code object detection unit, vehicle VIN code area image is obtained.Mesh is accorded with using start-stop again
Detection unit is marked, start-stop is obtained and accords with area image.Module of target detection detects vehicle VIN code region first, then in vehicle VIN
Start-stop is obtained in code region and accords with region, and this substep detection means can be effectively avoided because vehicle VIN code areas case is complicated
Bring erroneous detection influences, and improves the accuracy rate of vehicle VIN code zone location and start-stop symbol zone location, further improves vehicle
The accuracy that VIN code is detected and compared.
Vehicle VIN code object detection unit is based on SSD (Single Shot MultiBox Detector) frame come real
Existing.The frame uses VGG network as feature extractor.However, the effect of detection is also different for different characteristic patterns,
Therefore SSD uses feature pyramid structure, is classified and is returned simultaneously on multiple characteristic patterns.It is obtained using Softmax
The classification information of target obtains the location information of target using bounding box regression.As shown in figure 3, detection mould
Block first by vehicle VIN code area image input area detection model, obtain first N number of one-dimension array [class, x, y,
Width, height], first element of array represents object type, is obtained by Softmax, if target is vehicle VIN code
It is 1, if not being then 0.Rectangular area where four element characterization target objects, passes through bounding box after array
Regression is obtained, x, and y represents rectangle upper left angular coordinate, and width represents rectangle width, and height represents rectangular elevation.
Each array corresponds to a region, constructs region distance information using region rectangle frame size, most with rectangle frame area
Big array is exported as detection module, and vehicle VIN code region is then extracted from image by rectangle frame location information.This side
Method can effectively pick out other interference regions in background.
One new network of re -training is more complicated, and needs very big data volume, and parameter regulation also compares
Therefore difficulty is a good selection using fine tuning.So-called fine tuning is exactly that oneself is added on trained model
Data, the suitable model of training.Fine tuning is advantageous in that without re -training model, to greatly improve efficiency.Meanwhile
In the case that data volume itself is little, the feature that model learning can be made to arrive is finely tuned with more robustness.
Vehicle VIN code target detection model acquisition methods are as follows:
S21, training data prepare: obtaining the vehicle VIN code image of different shooting conditions (such as illumination, angle)；
S22, data mark: vehicle VIN code region is marked in the picture using rectangle frame, rectangle frame region domestic demand is complete
Include vehicle VIN code；
S23, model training: it utilizes and has used the trained VGG basic model of ImageNet, the vehicle that will have been marked
VIN code image inputs in SSD frame, is finely adjusted on basic model, to preferably train vehicle VIN code target detection mould
Type.Specifically: firstly, the mean value file of vehicle VIN code data set is calculated, because of the mean value file of vehicle VIN code data set
It is not quite alike with the mean value file of ImageNet data set.Then, the last output of modification SSD frame, ImageNet is one
The classification task of 1000 classes, and this model only has 2 classes, is background and vehicle VIN code respectively.It is followed by regularized learning algorithm rate, is learnt
Rate is very big corresponding to the performance influence of neural network, but by experience and can only test to obtain suitable value, by testing,
Basic studies rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005 by this patent, and learning rate strategy is set as
" multistep ", gamma is set as 0.1, momentum and is set as 0.9.Finally, load grounding model is finely adjusted.It is logical
Above-mentioned setting is crossed, model can restrain quickly and loss drops to 2 or so, improves the efficiency and precision of model training.
Start-stop symbol object detection unit is also based on SSD (Single Shot MultiBox Detector) frame and comes in fact
Existing.As shown in figure 3, the image input start-stop in vehicle VIN code image head and the tail region is accorded with detection model first, N number of one is obtained
Dimension group [class, x, y, width, height], first element of array represent object type, are obtained by Softmax, if
Target is that start-stop symbol is then 1, if not being then 0.Rectangular area where four element characterization target objects, passes through after array
Bounding box regression is obtained, and x, y represent rectangle upper left angular coordinate, and width represents rectangle width, height
Represent rectangular elevation.Each array corresponds to a region, extracts the position of start-stop symbol from image by rectangle frame location information
It sets.The method can effectively pick out other interference regions in background.
It is as follows that start-stop accords with target detection model acquisition methods:
S31, training data prepare: obtaining the start-stop symbol image of different shooting conditions (such as illumination, angle)；
S32, data mark: start-stop symbol region is marked in the picture using rectangle frame；
S33, model training: it utilizes and has used the trained VGG basic model of ImageNet, the vehicle that will have been marked
VIN code image inputs in SSD frame, is finely adjusted on basic model, to preferably start-stop be trained to accord with target detection model.
Specifically: firstly, the mean value file of start-stop symbol data set is calculated, because of the mean value file and ImageNet of start-stop symbol data set
The mean value file of data set is not quite alike.Then, the last output of modification SSD frame, ImageNet are points of 1000 classes
Generic task, and this model only has 2 classes, is background and start-stop symbol respectively.It is followed by regularized learning algorithm rate, learning rate corresponds to nerve net
The performance influence of network is very big, but can only be by experience and experiment to obtain suitable value, and by experiment, this patent will be learned substantially
Habit rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005, and learning rate strategy is set as " multistep ", and gamma is set
It is set to 0.1, momentum and is set as 0.9.Finally, load grounding model is finely adjusted.By above-mentioned setting, model can be very
Rapid convergence and loss drops to 2 or so, improves the efficiency and precision of model training.
The specific detection method of start-stop symbol classification of type module includes: as shown in figure 4, first with image to be detected and service
Start-stop symbol image training start-stop in device image accords with classification of type model, we accord with picture sample by observing a large amount of start-stops, greatly
It causes the character in according with start-stop to be divided into 14 kinds (including the case where that no start-stop accords with), is indicated respectively with 0~13.Then the mould is used
Type classifies to start-stop to be detected symbol image, finally obtains classification results with Softmax.Finally compare the type of start-stop symbol
It is whether consistent.
It is as follows that start-stop accords with classification of type model acquisition methods:
S1, training data prepare: obtaining the start-stop symbol image of different shooting conditions (such as illumination, angle)；
S2, data mark: different labels is stamped to different types of start-stop symbol；
S3, model training: using the training data marked, training start-stop accords with classification of type model.The model includes 3
Convolutional layer, 2 pond layers and 2 full articulamentums.A Softmax, the knot of output category are connect behind second full articulamentum
Fruit.
Tampering detection standard based on vehicle VIN code start-stop symbol type of the invention is as follows: vehicle VIN in image to be detected
Code region whether there is；Whether VIN digital content is correct；The type of start-stop symbol and rising for server archival image of image to be detected
Whether the type only accorded with is consistent；The present invention indicates verification state using one-dimension array [x1, x2, x3], initial value be [0,0,
0], flag bit x1 represents vehicle VIN code region and whether there is, and then x1 is 0 if it exists, and then x1 is 1 if it does not exist；Flag bit x2 generation
Whether Table VI N digital content is correct, if then x2 is 0, if not being 1 in then x2；Flag bit x3 represents the start-stop of image to be detected
Whether the type of the start-stop symbol of the type and archival image of symbol is consistent, and x3 is 0 if consistent, if inconsistent x3 is 1；Finally,
Statistical mark position state, if mark is is 0, verification passes through, if it exists 1, then it verifies and does not pass through.Occurred according to state 1
The available unsanctioned reason of verification in position.If x1 is 1, vehicle VIN code region or shooting angle may be not present in image
It spends against regulation；If x2 is 1, possible reason is that image taking is imperfect；If x3 is 1, the start-stop of image to be detected
It accords with not corresponding with the start-stop symbol for achieving photo, it is understood that there may be the case where distorting.
Determination module judges whether vehicle VIN code tampering detection passes through according to verification standard, and school is directly returned to if passing through
Success flag is tested, it is careful to remain the later period for the position back-checking failure cause and corresponding picture for being 1 according to flag bit if not passing through
Verify card.
Implementation detailed process of the invention as shown in Figure 1, the altering detecting method based on vehicle VIN code start-stop symbol type,
Include the following steps:
Archive VIN code image in S1, acquisition vehicle VIN code image to be detected and server；
S2, using the target detection model inspection VIN code image based on deep learning, judge that VIN code region whether there is,
Then recording this mark if it exists is 0, extracts VIN code region；Then recording this mark if it does not exist is 1, and saves related figure
Piece, into statistical analysis process；
S3, two regions of VIN code head and the tail are obtained, using the two regions of target detection model inspection based on deep learning
It is accorded with the presence or absence of start-stop；
S4, if it does not exist start-stop accord with, then output type is sky；Start-stop accords with if it exists, then using point based on deep learning
Class method judges that start-stop accords with type.Judge whether image to be detected and the start-stop of archival image symbol type are consistent, remember if consistent
Recording this mark is 0；It is 1 that this mark is recorded if inconsistent, and saves picture concerned, into statistical analysis process；
The advantages of basic principles and main features and this programme of this programme have been shown and described above.The technology of the industry
Personnel are it should be appreciated that this programme is not restricted to the described embodiments, and the above embodiments and description only describe this
The principle of scheme, under the premise of not departing from this programme spirit and scope, this programme be will also have various changes and improvements, these changes
Change and improvement is both fallen within the scope of claimed this programme.This programme be claimed range by appended claims and its
Equivalent thereof.
Claims (9)
1. a kind of altering detecting method based on vehicle VIN code start-stop symbol type, which comprises the following steps:
Archive VIN code image in S1, acquisition vehicle VIN code image to be detected and server；
S2, using the target detection model inspection VIN code image based on deep learning, judge that VIN code region whether there is, if depositing
It is 0 then recording this mark, extracts VIN code region；Then recording this mark if it does not exist is 1, and saves picture concerned, into
Enter to statistically analyze process；
S3, obtain VIN code from beginning to end two regions, using the two regions of target detection model inspection based on deep learning whether
There are start-stop symbols；
S4, if it does not exist start-stop accord with, then output type is sky；Start-stop accords with if it exists, then uses the classification mould based on deep learning
Type judges that start-stop accords with type, judges whether image to be detected and the start-stop of archival image symbol are consistent, this mark is recorded if consistent
Will is 0；It is 1 that this mark is recorded if inconsistent, and saves picture concerned, into statistical analysis process；
It is S5, for statistical analysis to the result of the action of whole process, flag bit all 0 is recorded, then VIN code tampering detection leads to
It crosses, if it exists mark 1, then VIN code tampering detection does not pass through；Meanwhile it is unacceptable according to the position acquisition verification that mark 1 occurs
Reason and problem picture.
2. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as described in claim 1, which is characterized in that
Judge that VIN code region whether there is in the step S2 and use following method:
The classification information of vehicle VIN code image object to be detected is obtained using Softmax, uses bounding box
Regression obtains the location information of vehicle VIN code image object to be detected.
3. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as described in claim 1, which is characterized in that
Vehicle VIN code target detection model obtaining step in the step S2 based on deep learning is as follows:
S21, training data prepare: obtaining the vehicle VIN code image of different shooting conditions (such as illumination, angle)；
S22, data mark: vehicle VIN code region is marked in the picture using rectangle frame, rectangle frame region domestic demand completely includes
Vehicle VIN code；
S23, model training: utilizing and used the trained VGG basic model of ImageNet, the vehicle VIN code that will have been marked
Image inputs in SSD frame, is finely adjusted on basic model, training vehicle VIN code target detection model.
4. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as claimed in claim 3, which is characterized in that
Described be finely adjusted on basic model includes the following steps:
S231. the mean value file of vehicle VIN code data set is calculated；
S232. the last output of modification SSD frame；
Basic studies rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005 by S233. regularized learning algorithm rate, learning rate
Strategy is set as " multistep ", and gamma is set as 0.1, momentum and is set as 0.9；
S234. load grounding model is finely adjusted.
5. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as claimed in claim 1 or 2 or 3 or 4,
It is characterized in that, the start-stop symbol target detection model obtaining step in the step S3 based on deep learning is as follows:
S31, training data prepare: obtaining the start-stop symbol image of different shooting conditions (such as illumination, angle)；
S32, data mark: start-stop symbol region is marked in the picture using rectangle frame；
S33, model training: utilizing and used the trained VGG basic model of ImageNet, the vehicle VIN code that will have been marked
Image inputs in SSD frame, is finely adjusted on basic model, to preferably start-stop be trained to accord with target detection model.
6. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as claimed in claim 5, which is characterized in that
Described be finely adjusted on basic model includes the following steps:
S231. the mean value file of vehicle VIN code data set is calculated；
S232. the last output of modification SSD frame；
Basic studies rate is adjusted to 0.0001, weight_decay and is adjusted to 0.0005 by S233. regularized learning algorithm rate, learning rate
Strategy is set as " multistep ", and gamma is set as 0.1, momentum and is set as 0.9；
S234. load grounding model is finely adjusted.
7. a kind of start-stop based on deep learning as described in claims 1 or 2 or 3 or 4 or 6 accords with genre classification methods, special
Sign is that the start-stop symbol classification of type model obtaining step in the step S4 based on deep learning is as follows:
S41, the vehicle VIN code start-stop for obtaining different angle, illumination, type and picture quality accord with image；
S42, classification of type deep learning network model is accorded with using start-stop symbol image training start-stop, obtains start-stop and accords with classification of type mould
Type.
8. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as claimed in claim 7, it is characterised in that:
Start-stop accords with classification of type model and uses this kind of convolutional neural networks of SSD, and uses VGG network as feature extractor,
Classification of type model is accorded with the start-stop symbol image training start-stop in image to be detected and server images first, start-stop is accorded with
In character be divided into 14 kinds, indicated respectively with 0~13,
Then classified using the model to start-stop to be detected symbol image,
Finally classification results are obtained with Softmax.
9. a kind of altering detecting method based on vehicle VIN code start-stop symbol type as claimed in claim 8, it is characterised in that:
It includes 3 convolutional layers, 2 pond layers and 2 full articulamentums that start-stop, which accords with classification of type model,.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811105641.8A CN109376738A (en) | 2018-09-20 | 2018-09-20 | Altering detecting method based on vehicle VIN code start-stop symbol type |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811105641.8A CN109376738A (en) | 2018-09-20 | 2018-09-20 | Altering detecting method based on vehicle VIN code start-stop symbol type |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109376738A true CN109376738A (en) | 2019-02-22 |
Family
ID=65401544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811105641.8A Pending CN109376738A (en) | 2018-09-20 | 2018-09-20 | Altering detecting method based on vehicle VIN code start-stop symbol type |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109376738A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245583A (en) * | 2019-05-27 | 2019-09-17 | 上海眼控科技股份有限公司 | A kind of intelligent identification Method of Vehicular exhaust survey report |
| CN110363761A (en) * | 2019-07-22 | 2019-10-22 | 上海眼控科技股份有限公司 | A kind of start-stop Mark Detection system and method for vehicle chassis dynamic detection |
| CN111984881A (en) * | 2019-05-21 | 2020-11-24 | 北京沃东天骏信息技术有限公司 | Prompting method, prompting device, terminal, server and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107818322A (en) * | 2017-10-13 | 2018-03-20 | 上海眼控科技股份有限公司 | A kind of vehicle VIN code tampering detection system and methods for vehicle annual test |
| CN108304787A (en) * | 2018-01-17 | 2018-07-20 | 河南工业大学 | Road target detection method based on convolutional neural networks |
-
2018
- 2018-09-20 CN CN201811105641.8A patent/CN109376738A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107818322A (en) * | 2017-10-13 | 2018-03-20 | 上海眼控科技股份有限公司 | A kind of vehicle VIN code tampering detection system and methods for vehicle annual test |
| CN108304787A (en) * | 2018-01-17 | 2018-07-20 | 河南工业大学 | Road target detection method based on convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| 吴亚熙;岑峰;: "基于卷积神经网络的多层级目标检测方法", 软件, no. 04 * |
| 吴晓烽;吴丽君;吴振辉;陈志聪;林培杰;文吉成;: "用于盲人视觉辅助的多目标快速识别并同步测距方法", 福州大学学报(自然科学版), no. 04 * |
| 李丹;牛中彬;汪鑫耘;: "卷积神经网络在车辆识别系统中的应用", 安徽工业大学学报(自然科学版), no. 02 * |
| 李超凡;陈庆奎;: "基于学习算法SSD的实时道路拥堵检测", 软件导刊, no. 06 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111984881A (en) * | 2019-05-21 | 2020-11-24 | 北京沃东天骏信息技术有限公司 | Prompting method, prompting device, terminal, server and storage medium |
| CN110245583A (en) * | 2019-05-27 | 2019-09-17 | 上海眼控科技股份有限公司 | A kind of intelligent identification Method of Vehicular exhaust survey report |
| CN110363761A (en) * | 2019-07-22 | 2019-10-22 | 上海眼控科技股份有限公司 | A kind of start-stop Mark Detection system and method for vehicle chassis dynamic detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109447076A (en) | A kind of vehicle VIN code recognition detection method for vehicle annual test | |
| CN109344835A (en) | Altering detecting method based on vehicle VIN code character position | |
| CN109784326A (en) | A kind of vehicle chassis detection method based on deep learning | |
| CN103761531B (en) | The sparse coding license plate character recognition method of Shape-based interpolation contour feature | |
| CN103324937B (en) | The method and apparatus of label target | |
| CN109344753A (en) | A kind of tiny fitting recognition methods of Aerial Images transmission line of electricity based on deep learning | |
| CN107818322A (en) | A kind of vehicle VIN code tampering detection system and methods for vehicle annual test | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN111368690A (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
| CN108520114A (en) | A kind of textile cloth defect detection model and its training method and application | |
| CN108537269A (en) | A kind of the object detection deep learning method and its system of weak interactive mode | |
| CN111080691A (en) | Infrared hot spot detection method and device for photovoltaic module | |
| CN109376738A (en) | Altering detecting method based on vehicle VIN code start-stop symbol type | |
| CN110288006A (en) | A kind of license plate number automatic Verification method and system | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN107958200A (en) | A kind of intelligent checking system and method for car repacking exhaust fan | |
| CN104777176A (en) | PCB detection method and device | |
| CN113158891B (en) | Cross-camera pedestrian re-identification method based on global feature matching | |
| Zhang et al. | Oriented feature augmentation for few-shot object detection in remote sensing images | |
| CN104484679B (en) | Non- standard rifle shooting warhead mark image automatic identifying method | |
| CN105989600A (en) | Characteristic point distribution statistics-based power distribution network device appearance detection method and system | |
| CN107798302A (en) | A kind of intelligent checking system and method for car mounting luggage frame | |
| CN111078565A (en) | Analysis method of software test result based on HOG feature extraction and SVM multi-classifier | |
| Peng et al. | Application of deep residual neural network to water meter reading recognition | |
| CN110705535A (en) | Method for automatically detecting test paper layout character line |
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 |
Application publication date: 20190222 |
|
| RJ01 | Rejection of invention patent application after publication |