CN109886060A - A kind of bar code area positioning method based on deep learning - Google Patents
A kind of bar code area positioning method based on deep learning Download PDFInfo
- Publication number
- CN109886060A CN109886060A CN201910130810.1A CN201910130810A CN109886060A CN 109886060 A CN109886060 A CN 109886060A CN 201910130810 A CN201910130810 A CN 201910130810A CN 109886060 A CN109886060 A CN 109886060A
- Authority
- CN
- China
- Prior art keywords
- bar code
- training
- data
- module
- method based
- 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 25
- 238000013135 deep learning Methods 0.000 title claims abstract description 17
- 238000012549 training Methods 0.000 claims abstract description 60
- 239000000284 extract Substances 0.000 claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000007418 data mining Methods 0.000 claims abstract description 4
- 238000012805 post-processing Methods 0.000 claims description 10
- 230000000644 propagated effect Effects 0.000 claims description 9
- 230000003252 repetitive effect Effects 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 5
- 230000004807 localization Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 230000000452 restraining effect Effects 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 13
- 238000001514 detection method Methods 0.000 abstract description 7
- 238000011161 development Methods 0.000 abstract description 6
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000012152 algorithmic method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
Abstract
The present invention provides a kind of, and the bar code area positioning method based on deep learning can greatly improve the accuracy that the feature of product area positioning is refined, improve discrimination using depth learning technology automatic sorting extraction product orientation feature；And in the case where facing successive generations of products, without separately developing algorithm, the algorithm development period is greatly shortened, improves the ability of the compatibility multiple product of detection device.It includes training part and predicted portions；The training part acquires a large amount of training data in advance, markup information, forms training set, in the training stage, advanced row Data Mining and processing are trained module later, in training module, convolutional neural networks are built by training set, network is positioned later and extracts feature, then carry out weight study, judge whether to restrain, model file is generated if convergence, if do not restrained, returns and repositions network extraction feature, model file carries out model verifying after generating.
Description
Technical field
The present invention relates to the technical field of bar code zone location, specially a kind of bar code zone location based on deep learning
Method.
Background technique
In existing industrial bar codes identification industry, the industrial bar codes zone location algorithm of mainstream is needed according to product spy
Property, artificially defined various rules carry out characterizing definition and modeling, mainly using based on the artificial machine vision side for extracting feature
Method acquires bar code image by industrial camera, is passed to rule-based localization method, extracts feature, output test result.Tradition
Detection method there are several respects: in detection algorithm development process, need to put into a large amount of energy and carry out algorithm prototype
Design and verifying；After successive generations of products, need to develop algorithm again, algorithmic method generalization ability is poor；Development cycle is long；It is fixed
Position index is difficult to reach target.
Summary of the invention
In view of the above-mentioned problems, the present invention provides a kind of bar code area positioning method based on deep learning, using deep
It spends learning art automatic sorting and extracts product orientation feature, the accurate of the feature refinement of product area positioning can be greatly improved
Property, improve discrimination；And in the case where facing successive generations of products, without separately developing algorithm, algorithm is greatly shortened
Development cycle improves the ability of the compatibility multiple product of detection device.
A kind of bar code area positioning method based on deep learning, it is characterised in that: it includes training part and prediction section
Point；
The training part acquires a large amount of training data in advance, markup information, forms training set, in the training stage, first
Data Mining and processing are carried out, module is trained later, in training module, convolutional neural networks is built by training set, it
Positioning network extracts feature afterwards, then carries out weight study, judges whether to restrain, and model file is generated if convergence, if do not received
It holds back, returns and reposition network extraction feature, model file carries out model verifying after generating, verifying is not up to standard, then returns again
Adjusting training collection, repetitive exercise, until meeting the requirements, verifying is up to standard, then disposes use；
The predicted portions include acquisition data, data prediction, prediction module, post-processing module, prediction result, acquisition
Enter data prediction after data, later by pretreated data after positioning network extracts feature, by training department
Model file obtained is divided to orient the potential region of bar code, module obtains positioning result region after post treatment later, passes through
Output result exports a width and has cut the Bar code positioning image behind a large amount of background areas.
It is further characterized by: when the training data, markup information, acquire a certain number of bar code pictures, picture
Data are intended to comprising barcode data, and labeled data is with the foundation of triple channel graphic form, and the labeled data in bar code region is stored in
In xml document, the starting point comprising tab area and width height, the width height of initial data and locating depth, the tag name information of bar code；
Described image is explored and pretreatment, according to training data and its corresponding markup information, carries out to data set
Enhancing, EDS extended data set guarantee the rich and varied property of input data sample；
The training module includes propagated forward and backpropagation two parts, after training set establishes, builds convolution mind
It is convenient for automatic sorting barcode data training set feature through network, obtains model file；
The training process of the training module it is as follows: a certain note of instruction code picture is input to convolutional neural networks in training set
In, it is operated by convolution sum deconvolution and carries out propagated forward, exported an one-dimensional vector, recycle this bar code picture corresponding
Markup information and the one-dimensional vector, which calculate, loses, and then carries out backpropagation using chain rule to adjust according to the error of the two
Weight, until convergence, generates final model file；
Whether weight file of the authentication module for Self -adaptive reaches use state, and can dispose if reaching makes
With otherwise readjusting training set, repetitive exercise, until meeting the requirements；For verifying index, set according to demand, including but
It is not limited to will test rate as index, when the verification and measurement ratio for reaching demand then stops repetitive exercise；
The acquisition data obtain image by image capturing system online；
The characteristics of described image preprocessing module is distributed according to bar code does ROI operation to the image of acquisition, reduces and calculate
Amount improves accuracy rate；
The prediction module only includes propagated forward part, and image is by convolution sum deconvolution operation in convolutional neural networks
Later, an one-dimensional vector is generated, the origin coordinates of the frame of type, number, localization region including each bar code detected
With the high information of width, thus, the potential region of orienting bar code, the corresponding candidate frame of output to post-processing module processing；
Accounting pair of the post-processing module by the area of the score threshold of each candidate frame and posting in the picture
The potential region generated in prediction module is filtered, final to obtain positioning result region；
Filtering of the post-processing module to candidate frame are as follows: the score of candidate frame is less than filtering out for score threshold, then
It filters out the area of the ROI image of the area and input of remaining candidate frame is relatively small, last remaining candidate frame is then
For positioning result region.
After adopting the above technical scheme, using depth learning technology, without artificially being concluded and being built to product characteristic
Mould, instead by the statistics of magnanimity product, product orientation feature is extracted using depth learning technology automatic sorting；Phase
The prior art in the industry than industrial bar codes identifying rows can greatly improve the accuracy that the feature of product area positioning is refined, improve
Discrimination；In the case where facing successive generations of products, without separately developing algorithm, the algorithm development period is greatly shortened,
Improve the ability of the compatibility multiple product of detection device.
Detailed description of the invention
Fig. 1 is the principle of the present invention block diagram.
Specific embodiment
A kind of bar code area positioning method based on deep learning, is shown in Fig. 1: it includes training part and predicted portions；
Training part acquires a large amount of training data, markup information, formation training set in advance and first carries out in the training stage
Data Mining and processing, are trained module later, in training module, build convolutional neural networks, Zhi Houding by training set
Position network extracts feature, then carries out weight study, judges whether to restrain, and model file is generated if convergence, if not restraining
Return repositions network and extracts feature, and model file carries out model verifying after generating, verifying is not up to standard, then returns to readjustment
Training set, repetitive exercise, until meeting the requirements, verifying is up to standard, then disposes use；
Predicted portions include acquisition data, data prediction, prediction module, post-processing module, prediction result, acquire data
Enter data prediction afterwards, later by pretreated data after positioning network extracts feature, by training part institute
The model file of acquisition orients the potential region of bar code, and module obtains positioning result region after post treatment later, passes through output
As a result it exports a width and has cut the Bar code positioning image behind a large amount of background areas.
When training data, markup information, a certain number of bar code pictures are acquired, image data is intended to comprising barcode data,
Labeled data is with the foundation of triple channel graphic form, and the labeled data in bar code region is stored in xml document, includes tab area
Starting point and width is high, initial data width is high and locating depth, the tag name information of bar code；
Image is explored and pretreatment, according to training data and its corresponding markup information, enhances data set,
EDS extended data set guarantees the rich and varied property of input data sample；
Training module includes propagated forward and backpropagation two parts, after training set establishes, builds convolutional Neural net
Network is convenient for automatic sorting barcode data training set feature, obtains model file；
The training process of training module it is as follows: a certain note of instruction code picture is input in convolutional neural networks in training set,
It is operated by convolution sum deconvolution and carries out propagated forward, exported an one-dimensional vector, recycle this corresponding mark of bar code picture
It infuses information and the one-dimensional vector is calculated and lost, backpropagation is then carried out using chain rule to adjust power according to the error of the two
Value, until convergence, generates final model file；
Whether weight file of the authentication module for Self -adaptive reaches use state, and use can be disposed if reaching,
Otherwise training set, repetitive exercise, until meeting the requirements are readjusted；For verifying index, set according to demand, including but unlimited
In the rate that will test as index, when the verification and measurement ratio for reaching demand then stops repetitive exercise；
Acquisition data obtain image by image capturing system online；
The characteristics of image pre-processing module is distributed according to bar code does ROI operation to the image of acquisition, reduces calculation amount, mentions
High-accuracy；
Prediction module only includes propagated forward part, and image operates it by convolution sum deconvolution in convolutional neural networks
Afterwards, generate an one-dimensional vector, the origin coordinates of the frame of type, number, localization region including each bar code detected and
The high information of width, thus, the potential region of orienting bar code, the corresponding candidate frame of output to post-processing module processing；
Post-processing module is by the area of the score threshold of each candidate frame and posting accounting in the picture to prediction
The potential region generated in module is filtered, final to obtain positioning result region；
Filtering of the post-processing module to candidate frame are as follows: the score of candidate frame is less than filtering out for score threshold, then surplus
Under candidate frame area and input ROI image area it is relatively small filter out, last remaining candidate frame is then fixed
Position results area.
It utilizes depth learning technology, without artificially being concluded and modeled to product characteristic, instead it is logical
The statistics for crossing magnanimity product extracts product orientation feature using depth learning technology automatic sorting；Compared to industrial bar codes identifying rows
The prior art in the industry can greatly improve the accuracy that the feature of product area positioning is refined, improve discrimination；Facing product
In the case where update, without separately developing algorithm, the algorithm development period is greatly shortened, improves the compatibility of detection device
The ability of multiple product.
Its advantages are as follows:
A no longer needs artificially to conclude by depth learning technology programming count bar code zone location feature and modeling bar code
Location feature；
B improves the generalization ability of detection method, i.e., after replacement product, algorithm is developed again without labor intensive again；
C greatly improves the accuracy of product area location feature refinement, improves discrimination.
Specific embodiments of the present invention are described in detail above, but content is only the preferable implementation of the invention
Example, should not be considered as limiting the invention the practical range of creation.It is all to become according to equalization made by the invention application range
Change and improve etc., it shall still fall within the scope of this patent.
Claims (10)
1. a kind of bar code area positioning method based on deep learning, it is characterised in that: it includes training part and predicted portions；
The training part acquires a large amount of training data, markup information, formation training set in advance and first carries out in the training stage
Data Mining and processing, are trained module later, in training module, build convolutional neural networks, Zhi Houding by training set
Position network extracts feature, then carries out weight study, judges whether to restrain, and model file is generated if convergence, if not restraining
Return repositions network and extracts feature, and model file carries out model verifying after generating, verifying is not up to standard, then returns to readjustment
Training set, repetitive exercise, until meeting the requirements, verifying is up to standard, then disposes use；
The predicted portions include acquisition data, data prediction, prediction module, post-processing module, prediction result, acquire data
Enter data prediction afterwards, later by pretreated data after positioning network extracts feature, by training part institute
The model file of acquisition orients the potential region of bar code, and module obtains positioning result region after post treatment later, passes through output
As a result it exports a width and has cut the Bar code positioning image behind a large amount of background areas.
2. a kind of bar code area positioning method based on deep learning as described in claim 1, it is characterised in that: the training
When data, markup information, acquire a certain number of bar code pictures, image data is intended to comprising barcode data, labeled data be with
Triple channel graphic form is established, and the labeled data in bar code region is stored in xml document, starting point and width comprising tab area
High, initial data width height and locating depth, the tag name information of bar code.
3. a kind of bar code area positioning method based on deep learning as claimed in claim 2, it is characterised in that: described image
It explores and pre-processes, according to training data and its corresponding markup information, data set is enhanced, EDS extended data set.
4. a kind of bar code area positioning method based on deep learning as claimed in claim 3, it is characterised in that: the training
Module includes propagated forward and backpropagation two parts, after training set establishes, builds convolutional neural networks convenient for returning automatically
It receives barcode data training set feature, obtains model file.
5. a kind of bar code area positioning method based on deep learning as claimed in claim 4, it is characterised in that: the training
The training process of module it is as follows, a certain note of instruction code picture is input in convolutional neural networks in training set, anti-by convolution sum
Convolution operation carry out propagated forward, export an one-dimensional vector, recycle this corresponding markup information of bar code picture and this one
Dimensional vector calculates loss, then carries out backpropagation using chain rule according to the error of the two to adjust weight, until convergence,
Generate final model file.
6. a kind of bar code area positioning method based on deep learning as described in claim 1, it is characterised in that: the acquisition
Data obtain image by image capturing system online.
7. a kind of bar code area positioning method based on deep learning as claimed in claim 6, it is characterised in that: described image
The characteristics of preprocessing module is distributed according to bar code does ROI operation to the image of acquisition.
8. a kind of bar code area positioning method based on deep learning as claimed in claim 7, it is characterised in that: the prediction
Module only includes propagated forward part, and image generates one one after convolution sum deconvolution operation in convolutional neural networks
Dimensional vector, the origin coordinates and the high information of width of the frame of type, number, localization region including each bar code detected, by
This, the potential region of orienting bar code, the corresponding candidate frame of output to post-processing module processing.
9. a kind of bar code area positioning method based on deep learning as claimed in claim 8, it is characterised in that: place after described
Reason module is by the area of the score threshold of each candidate frame and posting accounting in the picture to generating in prediction module
Potential region is filtered, final to obtain positioning result region.
10. a kind of bar code area positioning method based on deep learning as claimed in claim 9, it is characterised in that: after described
Processing module is filtered into candidate frame, and the score of candidate frame is less than filtering out for score threshold, then in remaining candidate frame
Area and input ROI image area it is relatively small filter out, last remaining candidate frame is then positioning result region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130810.1A CN109886060A (en) | 2019-02-21 | 2019-02-21 | A kind of bar code area positioning method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130810.1A CN109886060A (en) | 2019-02-21 | 2019-02-21 | A kind of bar code area positioning method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109886060A true CN109886060A (en) | 2019-06-14 |
Family
ID=66928854
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910130810.1A Pending CN109886060A (en) | 2019-02-21 | 2019-02-21 | A kind of bar code area positioning method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886060A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287752A (en) * | 2019-06-25 | 2019-09-27 | 北京慧眼智行科技有限公司 | A kind of dot matrix code detection method and device |
| CN110427793A (en) * | 2019-08-01 | 2019-11-08 | 厦门商集网络科技有限责任公司 | A kind of code detection method and its system based on deep learning |
| CN111767754A (en) * | 2020-06-30 | 2020-10-13 | 创新奇智(北京)科技有限公司 | Identification code identification method and device, electronic equipment and storage medium |
| CN113556194A (en) * | 2021-07-20 | 2021-10-26 | 电信科学技术第五研究所有限公司 | Wireless signal region strength detection method based on deep learning |
| CN115630660A (en) * | 2022-12-23 | 2023-01-20 | 湖北凯乐仕通达科技有限公司 | Barcode positioning method and device based on convolutional neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120224743A1 (en) * | 2011-03-04 | 2012-09-06 | Rodriguez Tony F | Smartphone-based methods and systems |
| CN107617573A (en) * | 2017-09-30 | 2018-01-23 | 浙江瀚镪自动化设备股份有限公司 | A kind of logistics code identification and method for sorting based on multitask deep learning |
| CN108427942A (en) * | 2018-04-22 | 2018-08-21 | 广州麦仑信息科技有限公司 | A kind of palm detection based on deep learning and crucial independent positioning method |
| CN108492291A (en) * | 2018-03-12 | 2018-09-04 | 苏州天准科技股份有限公司 | A kind of photovoltaic silicon chip Defect Detection system and method based on CNN segmentations |
| CN108920992A (en) * | 2018-08-08 | 2018-11-30 | 长沙理工大学 | A kind of positioning and recognition methods of the medical label bar code based on deep learning |
-
2019
- 2019-02-21 CN CN201910130810.1A patent/CN109886060A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120224743A1 (en) * | 2011-03-04 | 2012-09-06 | Rodriguez Tony F | Smartphone-based methods and systems |
| CN107617573A (en) * | 2017-09-30 | 2018-01-23 | 浙江瀚镪自动化设备股份有限公司 | A kind of logistics code identification and method for sorting based on multitask deep learning |
| CN108492291A (en) * | 2018-03-12 | 2018-09-04 | 苏州天准科技股份有限公司 | A kind of photovoltaic silicon chip Defect Detection system and method based on CNN segmentations |
| CN108427942A (en) * | 2018-04-22 | 2018-08-21 | 广州麦仑信息科技有限公司 | A kind of palm detection based on deep learning and crucial independent positioning method |
| CN108920992A (en) * | 2018-08-08 | 2018-11-30 | 长沙理工大学 | A kind of positioning and recognition methods of the medical label bar code based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 徐超、闫胜业: "改进的卷积神经网络行人检测方法", 《计算机应用》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287752A (en) * | 2019-06-25 | 2019-09-27 | 北京慧眼智行科技有限公司 | A kind of dot matrix code detection method and device |
| CN110427793A (en) * | 2019-08-01 | 2019-11-08 | 厦门商集网络科技有限责任公司 | A kind of code detection method and its system based on deep learning |
| CN110427793B (en) * | 2019-08-01 | 2022-04-26 | 厦门商集网络科技有限责任公司 | Bar code detection method and system based on deep learning |
| CN111767754A (en) * | 2020-06-30 | 2020-10-13 | 创新奇智(北京)科技有限公司 | Identification code identification method and device, electronic equipment and storage medium |
| CN111767754B (en) * | 2020-06-30 | 2024-05-07 | 创新奇智(北京)科技有限公司 | Identification code identification method and device, electronic equipment and storage medium |
| CN113556194A (en) * | 2021-07-20 | 2021-10-26 | 电信科学技术第五研究所有限公司 | Wireless signal region strength detection method based on deep learning |
| CN115630660A (en) * | 2022-12-23 | 2023-01-20 | 湖北凯乐仕通达科技有限公司 | Barcode positioning method and device based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886060A (en) | A kind of bar code area positioning method based on deep learning | |
| Qiao et al. | LGPMA: complicated table structure recognition with local and global pyramid mask alignment | |
| Azizah et al. | Deep learning implementation using convolutional neural network in mangosteen surface defect detection | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN110088776A (en) | For training the computer equipment of deep neural network | |
| CN110188761A (en) | Recognition methods, device, computer equipment and the storage medium of identifying code | |
| CN105740892A (en) | High-accuracy human body multi-position identification method based on convolutional neural network | |
| CN103984920B (en) | Three-dimensional face identification method based on sparse representation and multiple feature points | |
| CN110031597A (en) | A kind of biological water monitoring method | |
| CN106372666A (en) | Target identification method and device | |
| CN109784130A (en) | Pedestrian recognition methods and its device and equipment again | |
| CN108492291A (en) | A kind of photovoltaic silicon chip Defect Detection system and method based on CNN segmentations | |
| CN107316029A (en) | A kind of live body verification method and equipment | |
| CN109886295A (en) | A kind of butterfly recognition methods neural network based and relevant device | |
| CN108154133A (en) | Human face portrait based on asymmetric combination learning-photo array method | |
| CN110390289A (en) | Based on the video security protection detection method for censuring understanding | |
| CN106022241A (en) | Face recognition method based on wavelet transformation and sparse representation | |
| CN107798308A (en) | A kind of face identification method based on short-sighted frequency coaching method | |
| Paul et al. | Extraction of facial feature points using cumulative histogram | |
| CN110232361A (en) | Human body behavior intension recognizing method and system based on the dense network of three-dimensional residual error | |
| CN110334719A (en) | The method and system of object image are built in a kind of extraction remote sensing image | |
| CN104200202B (en) | A kind of upper half of human body detection method based on cumulative perceptron | |
| Yan et al. | A multi-task learning model for better representation of clothing images | |
| Hudaya et al. | Implementation of verification and matching E-KTP with faster R-CNN and ORB |
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: 20190614 |
|
| RJ01 | Rejection of invention patent application after publication |