CN110378167B - Bar code image correction method based on deep learning - Google Patents
Bar code image correction method based on deep learning Download PDFInfo
- Publication number
- CN110378167B CN110378167B CN201910614176.9A CN201910614176A CN110378167B CN 110378167 B CN110378167 B CN 110378167B CN 201910614176 A CN201910614176 A CN 201910614176A CN 110378167 B CN110378167 B CN 110378167B
- Authority
- CN
- China
- Prior art keywords
- bar code
- fuzzy
- image
- code image
- incomplete
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K7/00—Methods or arrangements for sensing record carriers, e.g. for reading patterns
- G06K7/10—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation
- G06K7/14—Methods or arrangements for sensing record carriers, e.g. for reading patterns by electromagnetic radiation, e.g. optical sensing; by corpuscular radiation using light without selection of wavelength, e.g. sensing reflected white light
- G06K7/1404—Methods for optical code recognition
- G06K7/146—Methods for optical code recognition the method including quality enhancement steps
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Toxicology (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- General Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a bar code image correction algorithm based on deep learning, which comprises the following steps: the image is to be parsed into sample points in a probability distribution; the method comprises the steps of analyzing how to generate a pseudo image by deep learning, modeling a fuzzy incomplete bar code image by using a semantic segmentation related model, realizing correction of the fuzzy incomplete bar code image, correcting the fuzzy or incomplete image in a complex scene, then finding out the optimal pseudo image required by complete restoration, and correcting and restoring the fuzzy incomplete bar code image based on U-net, wherein the neural network is of an Encoder-Decoder structure, the Encoder is a contraction path and consists of a convolution layer and a pooling layer, the model is optimized from three angles in order to realize accurate positioning, the correction of the fuzzy incomplete bar code image is realized, and the fuzzy incomplete bar code image is enabled to be complete and clear through an algorithm, so that the identification accuracy of the bar code image is improved, and the protection of an electric power warehousing full-automatic unmanned operation system is guaranteed.
Description
Technical Field
The invention relates to the technical field of barcode image processing, in particular to a barcode image correction method based on deep learning.
Background
At present, in a full-automatic unmanned operation system of power storage, a bar code on a material nameplate needs to be automatically identified. The cabinet type material nameplate material has aluminium and stainless steel, has problems such as reflection of light, surface corrosion, surface mar. The image recognition camera is sensitive to light, the larger the light reflection degree is, the larger the error probability is, the removing algorithm is self, the image quality is the largest factor influencing the bar code recognition accuracy, and the quality of the image is measured from three aspects: the original bar code image of the material nameplate needs to be corrected due to the defects of inclination, definition and blur.
The current correction algorithms for blurred and incomplete images mainly include the following three types:
in the method, special hardware equipment is generally used for scanning three-dimensional shape information of a material nameplate bar code. For example, the structured light source is used to scan an image to obtain three-dimensional information, i.e., depth information, of the image, and then the image is corrected according to the depth information.
The image correction algorithm based on 3D model reconstruction is mainly based on factors causing bar code image blurring and deformity, and comprises bar codes and placing angles thereof, light source directions, image acquisition equipment characteristics and the like. 3D modeling is carried out on the bar code image, and the existing mathematical knowledge is utilized to correct the fuzzy deformity.
The algorithm discards geometric simulation and 3D modeling of a fuzzy incomplete image, directly analyzes the barcode image, and then designs the fuzzy incomplete image correction algorithm which is not influenced by factors except the barcode image.
The three algorithms have respective limitations, which can be summarized as: most of the traditional methods are used for modeling specific scenes, and the models cannot function in a plurality of large scenes.
Disclosure of Invention
The invention aims to solve the technical problem of providing a bar code image correction method based on deep learning, which is used for improving the identification accuracy of bar code images and is a full-automatic unmanned operation system for power warehousing and is used for enabling fuzzy and incomplete bar code images to be complete and clear through an algorithm.
In order to solve the technical problem, the invention provides a bar code image correction algorithm based on deep learning, which comprises the following steps:
s1: analyzing the image into sample points in probability distribution, wherein the sample points are small sample images;
s2: analyzing a pseudo image generated by deep learning, modeling a fuzzy incomplete bar code image by using a semantic segmentation related model, converting a pixel-level classification problem into a pixel-level regression problem, correcting the fuzzy incomplete bar code image, correcting a fuzzy or incomplete image in a complex scene, further deeply learning related contents in the semantic segmentation field, correcting the fuzzy incomplete bar code image, automatically generating a data set of the fuzzy incomplete bar code image by the data set, and finally constructing and optimizing the model;
s3: and then finding an optimal pseudo image required by complete restoration, and correcting and restoring the fuzzy incomplete bar code image based on U-net, wherein the neural network is an Encoder-Decoder structure, the Encoder is a contraction path which consists of a convolution layer and a Pooling layer, so that the extraction of characteristics or the capture of semantics is realized, the Decoder is an expansion path and is realized by transposition convolution and jump connection, in order to realize upsampling, the dimensionality of the image is reduced due to downsampling performed by Pooling operation, and the dimensionality of a feature map is increased by transposition convolution, so that the size of the original image is restored, pixel level regression is realized, and the model is optimized from three angles in order to realize accurate positioning:
changing the structure of the model: changing from U-net to Stacked U-net to improve resolution;
modifying the loss function: in the optimization process of the model, the difference between the original distance between adjacent pixel points and the prediction result is not too large, so that the phenomenon of fuzzy deformation of the bar code image is improved;
post-processing the prediction result to improve the noise phenomenon;
in the final model evaluation, the structural similarity indexes are adopted to evaluate the image similarity before and after correction, and the multi-level structural similarity is summarized.
In S3, in the final model evaluation, the structural similarity index is used to evaluate the similarity of the images before and after correction, and the structural similarity is summarized in multiple scales, and the calculation formula is:
after adopting the structure, the invention has the advantages that:
the fuzzy incomplete bar code image is corrected, and the fuzzy incomplete bar code image becomes complete and clear through an algorithm, so that the identification accuracy of the bar code image is improved, and the driving protection navigation of the electric power warehousing full-automatic unmanned operation system is realized.
Drawings
FIG. 1 is a schematic diagram of the present invention for self-generating a data set of a blurred and incomplete bar code image.
Detailed Description
The invention is described in detail below with reference to the drawings and the detailed description, which should not be construed as limiting the invention.
As shown in fig. 1, an algorithm of a barcode image correction method based on deep learning includes the following steps:
s1: the image is to be parsed into sample points in a probability distribution;
s2: analyzing a pseudo image generated by deep learning, modeling a fuzzy incomplete bar code image by using a semantic segmentation related model, converting a pixel-level classification problem into a pixel-level regression problem, correcting the fuzzy incomplete bar code image, correcting a fuzzy or incomplete image in a complex scene, further deeply learning related contents in the semantic segmentation field, correcting the fuzzy incomplete bar code image, automatically generating a data set of the fuzzy incomplete bar code image by the data set, and finally constructing and optimizing the model;
s3: and then finding an optimal pseudo image required by complete restoration, and correcting and restoring the fuzzy incomplete bar code image based on U-net, wherein the neural network is an Encoder-Decoder structure, the Encoder is a contraction path which consists of a convolution layer and a Pooling layer, so that the extraction of characteristics or the capture of semantics is realized, the Decoder is an expansion path and is realized by transposition convolution and jump connection, in order to realize upsampling, the dimensionality of the image is reduced due to downsampling performed by Pooling operation, and the dimensionality of a feature map is increased by transposition convolution, so that the size of the original image is restored, pixel level regression is realized, and the model is optimized from three angles in order to realize accurate positioning:
changing the structure of the model: changing from U-net to Stacked U-net to improve resolution;
modifying the loss function: in the optimization process of the model, the difference between the original distance between adjacent pixel points and the prediction result is not too large, so that the phenomenon of fuzzy deformation of the bar code image is improved;
post-processing the prediction result to improve the noise phenomenon;
in the final model evaluation, the structural similarity index is adopted to evaluate the similarity of the images before and after correction, the multi-level structural similarity summarization builds a model for the fuzzy incomplete bar code image by using a semantic segmentation related model along with the rise of deep learning in recent years, the pixel-level classification problem is converted into the pixel-level regression problem, the correction of the fuzzy incomplete bar code image is realized, the model has certain generalization capability, and the correction can be carried out on the fuzzy or incomplete image in a complex scene.
In consideration of the complexity of actual services, the traditional method cannot be sufficient, so that the application is combined with relevant knowledge in the deep learning semantic segmentation field, an optimization scheme is provided aiming at the defects of the existing method, and the correction of the fuzzy incomplete bar code image is realized.
And (4) automatically generating a data set of the fuzzy incomplete bar code image by referring to a deep learning method.
And (4) model construction and optimization, and correction and restoration of the fuzzy incomplete bar code image based on U-net. The neural network can be understood as an Encoder-Decoder structure, wherein the Encoder is a contraction path and mainly comprises a convolution layer and a Pooling layer, the main purpose is to realize extraction of features or capture of semantics, the Decoder is an expansion path and is mainly realized by transposition convolution and jump connection, the main purpose is to realize up-sampling, the dimensionality of an image is reduced due to down-sampling performed by Pooling operation, and the transposition convolution can enlarge the dimensionality of a feature map so as to recover the size of an original image, thereby realizing pixel-level regression. The results obtained are however very crude, and the optimization of the model is carried out from three angles in order to achieve a precise positioning:
changing the structure of the model: changing from U-net to Stacked U-net to improve resolution;
modifying the loss function: the model enables the difference between the distance between the original adjacent pixel points and the prediction result not to be too large in the optimization process, so that the phenomenon of fuzzy deformation of the bar code image is improved;
and post-processing the prediction result to improve the noise phenomenon.
And (3) evaluating the model, namely evaluating the Similarity of the images before and after correction by adopting an MS-SSIM index in the final model evaluation, wherein the MS-SSIM is called Multi-Scale structured Similarity and is the summary of SSIM (structural Similarity) under Multi-scale as the name suggests. The calculation formula is as follows:
Claims (1)
1. a bar code image correction method based on deep learning is characterized by comprising the following steps:
s1: analyzing the image into sample points in the probability distribution, wherein the sample points are small sample images;
s2: analyzing a pseudo image generated by deep learning, modeling a fuzzy incomplete bar code image by using a semantic segmentation related model, converting a pixel-level classification problem into a pixel-level regression problem, correcting the fuzzy incomplete bar code image, correcting a fuzzy or incomplete image in a complex scene, further deeply learning related contents in the semantic segmentation field, correcting the fuzzy incomplete bar code image, automatically generating a data set of the fuzzy incomplete bar code image by the data set, and finally constructing and optimizing the model;
s3: finding an optimal pseudo image required by complete restoration, and restoring and correcting a fuzzy incomplete bar code image based on U-net, wherein the neural network is of an Encoder-Decoder structure, the Encoder is a contraction path and consists of a convolution layer and a pooling layer, so that the extraction of characteristics or the capture of semantics is realized, the Decoder is realized by transposition convolution and jump connection, and the transposition convolution enables the dimensionality of a feature map to be increased, so that the size of an original image is restored, and pixel-level regression is realized; the optimization of the model is performed from three perspectives:
changing the structure of the model: changing from U-net to Stacked U-net to improve resolution;
modifying the loss function: the distance between the original adjacent pixel points is different from the prediction result in the optimization process of the model, so that the phenomenon of fuzzy deformation of the bar code image is improved;
post-processing the prediction result to improve the noise phenomenon;
in S3, in the final model evaluation, the structural similarity index is used to evaluate the similarity of the images before and after correction, and the multi-level structural similarity is summarized by the following calculation formula:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614176.9A CN110378167B (en) | 2019-07-09 | 2019-07-09 | Bar code image correction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614176.9A CN110378167B (en) | 2019-07-09 | 2019-07-09 | Bar code image correction method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110378167A CN110378167A (en) | 2019-10-25 |
| CN110378167B true CN110378167B (en) | 2022-09-16 |
Family
ID=68252506
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910614176.9A Active CN110378167B (en) | 2019-07-09 | 2019-07-09 | Bar code image correction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378167B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111123183B (en) * | 2019-12-27 | 2022-04-15 | 杭州电子科技大学 | Rapid magnetic resonance imaging method based on complex R2U _ Net network |
| CN111292255B (en) * | 2020-01-10 | 2023-01-17 | 电子科技大学 | Filling and correcting technology based on RGB image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833186A (en) * | 2017-10-26 | 2018-03-23 | 长沙全度影像科技有限公司 | A kind of simple lens spatial variations image recovery method based on Encoder Decoder deep learning models |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108734677A (en) * | 2018-05-21 | 2018-11-02 | 南京大学 | A kind of blind deblurring method and system based on deep learning |
-
2019
- 2019-07-09 CN CN201910614176.9A patent/CN110378167B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833186A (en) * | 2017-10-26 | 2018-03-23 | 长沙全度影像科技有限公司 | A kind of simple lens spatial variations image recovery method based on Encoder Decoder deep learning models |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108734677A (en) * | 2018-05-21 | 2018-11-02 | 南京大学 | A kind of blind deblurring method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110378167A (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10353271B2 (en) | Depth estimation method for monocular image based on multi-scale CNN and continuous CRF | |
| CN110322495B (en) | Scene text segmentation method based on weak supervised deep learning | |
| CN116664605B (en) | Medical image tumor segmentation method based on diffusion model and multi-mode fusion | |
| CN110766708B (en) | Image comparison method based on contour similarity | |
| US11651581B2 (en) | System and method for correspondence map determination | |
| CN114724155A (en) | Scene text detection method, system and equipment based on deep convolutional neural network | |
| CN113744153B (en) | Double-branch image restoration forgery detection method, system, equipment and storage medium | |
| CN115345866B (en) | Building extraction method in remote sensing image, electronic equipment and storage medium | |
| CN112949338A (en) | Two-dimensional bar code accurate positioning method combining deep learning and Hough transformation | |
| KR20200063368A (en) | Unsupervised stereo matching apparatus and method using confidential correspondence consistency | |
| US20210304364A1 (en) | Method and system for removing noise in documents for image processing | |
| CN110378167B (en) | Bar code image correction method based on deep learning | |
| CN112597998A (en) | Deep learning-based distorted image correction method and device and storage medium | |
| CN115578262A (en) | Polarization image super-resolution reconstruction method based on AFAN model | |
| CN117593275A (en) | Medical image segmentation system | |
| CN110390336B (en) | Method for improving feature point matching precision | |
| CN116778164A (en) | Semantic segmentation method for improving deep V < 3+ > network based on multi-scale structure | |
| CN116129417A (en) | Digital instrument reading detection method based on low-quality image | |
| CN115578574A (en) | Three-dimensional point cloud completion method based on deep learning and topology perception | |
| CN113657225B (en) | Target detection method | |
| CN111582275B (en) | Serial number identification method and device | |
| Jia et al. | One‐Shot M‐Array Pattern Based on Coded Structured Light for Three‐Dimensional Object Reconstruction | |
| CN111985535A (en) | Method and device for optimizing human body depth map through neural network | |
| CN115984104B (en) | Multi-focus image fusion method and device based on self-supervision learning | |
| CN117078564B (en) | Intelligent generation method and system for video conference picture |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |