CN107016521A - A kind of warehouse nameplate recognition methods based on image convolution nerual network technique - Google Patents
A kind of warehouse nameplate recognition methods based on image convolution nerual network technique Download PDFInfo
- Publication number
- CN107016521A CN107016521A CN201710283801.7A CN201710283801A CN107016521A CN 107016521 A CN107016521 A CN 107016521A CN 201710283801 A CN201710283801 A CN 201710283801A CN 107016521 A CN107016521 A CN 107016521A
- Authority
- CN
- China
- Prior art keywords
- picture
- warehouse
- recognition methods
- sign board
- methods 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/087—Inventory or stock management, e.g. order filling, procurement or balancing against orders
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Multimedia (AREA)
- Economics (AREA)
- Marketing (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Accounting & Taxation (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of warehouse nameplate recognition methods based on image convolution nerual network technique, mainly include the following steps that:The interesting part to photo is extracted from picture stream first, processing is zoomed in and out to picture afterwards, an equal amount of picture is converted into;Detected based on optical flow approach for Sign Board picture, when detecting picture and changing, extract the picture of change moment picture and former frame and a later frame;Using this three width picture as input, picture calculating is carried out using convolutional neural networks, it is final to judge that Sign Board occurs how to change.The inventive method is by extracting area-of-interest in Sign Board scene picture, picture is zoomed into fixed size, detect Sign Board change, and carry out piecemeal processing using convolutional Neural algorithm, avoid large warehoused middle Sign Board quantity many, scene complexity caused by be difficult to detection the problem of, greatly improve measuring accuracy.It the composite can be widely applied to warehousing management field.
Description
Technical field
The present invention relates to storehouse management, Sign Board identification field of especially storing in a warehouse, and in particular to one kind is based on image convolution
The warehouse nameplate recognition methods of nerual network technique.
Background technology
With the constantly improve of China market economic structure, the high speed development of productivity is used as the core of business administration:Storehouse
Library management turns into the key factor of enterprise development competitiveness.But in management, bulk storage plant often has the wrong material of hair, sends out flushing
Situation occur, wherein there is the negligent subjective composition of custodial staff unavoidably, and most important reason should be goods yard and mark
Knowledge is unclear, and this goods yard and mark are asymmetric, are on the one hand the opposing party because there occurs mistake when placing Sign Board
Face is moved by mistake because Sign Board there occurs in use.But, store in a warehouse article Sign Board using be storage pipe
Essential one side in reason, therefore, enterprise to storage article Sign Board in the urgent need to realizing management and monitoring.
The management of conventional storage article Sign Board is focused primarily on goods yard and Sign Board is symmetrical, the modernization hand of Sign Board
Above section etc., but the maloperation of Sign Board, as the more mistake of a probability of occurrence, but never effective means are come
Evade, because these links can produce a large amount of and not easy-operating information, this brings great inconvenience to the management of enterprise.With
The development of image recognition technology, conventional images identification technology means can be passed through, it is possible to achieve the accurate identification of Sign Board.Volume
Product neutral net is developed recently, and causes a kind of efficient identification method paid attention to extensively, particularly in pattern classification neck
Domain, because the network avoids the complicated early stage pretreatment to image, can directly input original image, thus obtain more
It is widely applied.
Sign Board of storing in a warehouse manages difficult point:General bulk storage plant storage Sign Board species is more and quantity is big, warehouse keeper
Number energy is limited, it is impossible to which the moment observes the change of goods yard Sign Board, is moved by mistake once Sign Board there occurs, it is likely that give storage pipe
Reason brings irreversible consequence.Therefore using the storage Sign Board observation procedure under image detection means, storehouse is monitored in real time
Store up Sign Board change.
The content of the invention
The invention aims to overcome the shortcomings of the above, there is provided a kind of storage mark based under convolutional neural networks
Board high accuracy recognition methods.
The purpose of the present invention is achieved through the following technical solutions:A kind of warehouse based on image convolution nerual network technique
Nameplate recognition methods, is comprised the following steps that:
A. the interesting part to photo is extracted from picture stream first, processing is zoomed in and out to picture afterwards and is converted into equally
The picture of size;
B. detected based on optical flow approach for Sign Board picture, when detecting picture and changing, extract and become
Change moment picture and the picture of former frame and a later frame;
C. this three width picture is subjected to picture calculating as input using convolutional neural networks, it is final to judge that Sign Board occurs such as
What changes.
Further improvement of the present invention is:Step A is specially:
A1. extract region interested in picture and be stored as new picture;
A2. new picture is uniformly zoomed into 32mm × 32mm sizes;
A3. A1 is repeated, new picture is obtained;
A4. A2 is repeated, new picture is converted into formed objects.
Further improvement of the present invention is:Step B is specially:Using optical flow method to the new picture that is extracted in step A
Detected, when detecting that profile changes in new picture, the image of significant change former frame figure therewith is occurred into for profile
As simultaneously latter two field picture is preserved therewith.
Further improvement of the present invention is:Step C is specially:Convolutional neural networks recognition methods, convolutional neural networks
Including 7 layers, be followed successively by S1 convolutional layers, S1 down-samplings layer, S2 convolutional layers, S2 down-samplings layer, S3 convolutional layers, characteristic vector layer with it is defeated
Go out layer.
Further improvement of the present invention is:Optical flow method is to be associated by two-dimension speed with gradation of image, introduces light
Constraint equation is flowed, the basic skills of obtained calculating light stream is mainly used in tracking objective contour.
Further improvement of the present invention is:The picture of one secondary 32mm × 32mm sizes, 6 are changed into via S1 volumes of basic unit
28mm × 28mm characteristic pattern, S1 sample levels include 6 14mm × 14mm characteristic patterns, via S2 volumes of basic unit be changed into 16 10mm ×
10mm characteristic pattern, S1 sample levels include 16 5mm × 5mm characteristic patterns, and the 3rd Ge Juan basic units include 300 neurons.
The present invention has advantages below compared with prior art:The inventive method is felt by extracting in Sign Board scene picture
Interest region, fixed size, detection Sign Board change are zoomed to by picture, and carry out piecemeal processing using convolutional Neural algorithm,
Avoid that large warehoused middle Sign Board quantity is more, caused by scene complexity the problem of being difficult to detection, greatly improve test
Precision.
Brief description of the drawings:
Fig. 1 is the step flow chart of the inventive method;
Fig. 2 be inventive network in convolutional neural networks schematic flow sheet;
Fig. 3 is the convolution process schematic diagram in inventive network.
Embodiment:
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below by the skill in the embodiment of the present invention
Art scheme is clearly and completely described, it is clear that described embodiment is a part of embodiment of the invention, rather than all
Embodiment.The element and feature described in one embodiment of the invention can be with one or more other embodiment party
The element and feature shown in formula is combined.It should be noted that eliminating for purposes of clarity, in explanation unrelated to the invention
, the expression and description of part known to persons of ordinary skill in the art and processing.Based on the embodiment in the present invention, this area
The every other embodiment that those of ordinary skill is obtained on the premise of creative work is not paid, belongs to guarantor of the present invention
The scope of shield.
As shown in figure 1, a kind of warehouse nameplate recognition methods based on image convolution nerual network technique, including following step
Suddenly:
A. the interesting part to photo is extracted from picture stream first, processing is zoomed in and out to picture afterwards and is converted into equally
The picture of size;
B. detected based on optical flow approach for Sign Board picture, when detecting picture and changing, extract and become
Change moment picture and the picture of former frame and a later frame;
C. this three width picture is subjected to picture calculating as input using convolutional neural networks, it is final to judge that Sign Board occurs such as
What changes.
Convolutional neural networks are developed recentlies, and cause a kind of efficient identification method paid attention to extensively.It is in research
It is used to find that its unique network structure can be effectively reduced during the neuron of local sensitivity and set direction in cat cortex
The complexity of Feedback Neural Network, then proposes convolutional neural networks.Now, convolutional neural networks have become numerous science
One of the study hotspot in field, particularly in pattern classification field, locates in advance because the network is avoided to the complicated early stage of image
Reason, can directly input original image, thus obtain more being widely applied.
Step B is detects using optical flow method to the new picture extracted in step A, when detecting profile in new picture
When changing, by profile occur significant change image therewith previous frame image latter two field picture is preserved simultaneously therewith.
Step C is specially that convolutional neural networks include 7 layers, is followed successively by C1 convolutional layers, S1 down-samplings layer, C2 convolutional layers, S2
Down-sampling layer, C3 convolutional layers, characteristic vector layer and output layer.
Neutral net is as shown in Fig. 2 each neural network model is made up of neutral net unit, shown in Fig. 2
For the neutral net with hidden layer.
As shown in figure 3, convolution process can be described as:With wave filter f convolution input pictures, a biasing b is added afterwards,
Obtain convolutional layer Cx.It can be regarded as doing convolution algorithm from the mapping of plane to a next plane, down-sampling layer can be regarded as
It is fuzzy filter, plays a part of Further Feature Extraction.Spatial resolution is successively decreased between hidden layer and hidden layer, and contained by every layer
Number of planes is incremented by, and so can be used for detecting more characteristic informations.Gone to perceive input picture with the convolution kernel of a fixed size
In each neuron(I.e. each pixel), characteristic patterns are produced at C1 layers after convolution, then every group of four pictures in characteristic pattern
Element is summed again, and weighted value, biasing is put, and S2 layers of characteristic pattern is obtained by a function, and these characteristic patterns pass through convolution again
Obtain C3 layers.This hierarchical structure produces S4 as S2 again, by S4 layers each characteristic pattern with it is each in convolutional layer C5
Individual neuron connection, can so prevent the generation of over-fitting.Finally, these pixel values are rasterized in characteristic vector layer, and
Connect into a vector and be input to traditional neutral net, exported.Usually, C layers are characterized extract layer, i.e. convolutional layer,
Remove to perceive above one layer of each characteristic pattern weights with a convolution kernel being made up of, this has just extracted the feature of image, and
And generate the characteristic pattern of the convolutional layer;S layers are down-sampling layers, and each computation layer of network is made up of multiple Feature Mappings, each
Feature Mapping is that the weights of all neurons in a plane, plane are equal.
Finally it should be noted that:Although the present invention and its advantage have been described in detail above it should be appreciated that not
Various changes can be carried out in the case of the spirit and scope of the present invention being defined by the claims appended hereto, substitute and
Conversion.Moreover, the scope of the present invention is not limited only to process, equipment, means, the specific reality of method and steps described by specification
Apply example.One of ordinary skilled in the art will readily appreciate that from the disclosure, can be used and held according to the present invention
The row function essentially identical to corresponding embodiment described herein obtains result, the existing and future essentially identical with it
Process, equipment, means, method or step to be developed.Therefore, appended claim is wrapped in the range of being directed at them
Include such process, equipment, means, method or step.
Claims (6)
1. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique, it is characterised in that:Specific steps are such as
Under:
A. the interesting part to photo is extracted from picture stream first, processing is zoomed in and out to picture afterwards and is converted into equally
The picture of size;
B. detected based on optical flow method for Sign Board picture, when detecting picture and changing, extract and change
The picture of moment picture and former frame and a later frame;
C. this three width picture is subjected to picture calculating as input using convolutional neural networks, it is final to judge that Sign Board occurs such as
What changes.
2. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique according to claim 1, it is special
Levy and be:Step A is specially:
A1. extract region interested in picture and be stored as new picture;
A2. new picture is uniformly zoomed into 32mm × 32mm sizes;
A3. step A1 is repeated, new picture is obtained;
A4. step A2 is repeated, new picture is converted into formed objects.
3. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique according to claim 1, it is special
Levy and be:Step B is specially:The new picture extracted in step A is detected using optical flow method, when detecting in new picture
When profile changes, by profile occur significant change image therewith previous frame image therewith latter two field picture simultaneously preserve under
Come.
4. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique according to claim 1, it is special
Levy and be:Step C is specially:Described convolutional neural networks recognition methods, is characterised by, convolutional neural networks include 7 layers, according to
Secondary is S1 convolutional layers, S1 down-samplings layer, S2 convolutional layers, S2 down-samplings layer, S3 convolutional layers, characteristic vector layer and output layer.
5. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique according to claim 3, it is special
Levy and be:Optical flow method is to be associated by two-dimension speed with gradation of image, introduces optical flow constraint equation, obtained calculating light stream
Basic skills, be mainly used in track objective contour.
6. a kind of warehouse nameplate recognition methods based on image convolution nerual network technique according to claim 4, it is special
Levy and be:The picture of one secondary 32mm × 32mm sizes, is changed into six 28mm × 28mm characteristic pattern, S1 samplings via S1 volumes of basic unit
Layer includes six 14mm × 14mm characteristic patterns, is changed into 16 10mm × 10mm characteristic pattern, S1 sample levels via S2 volumes of basic unit
Comprising 16 5mm × 5mm characteristic patterns, the 3rd Ge Juan basic units include 300 neurons.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710283801.7A CN107016521A (en) | 2017-04-26 | 2017-04-26 | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710283801.7A CN107016521A (en) | 2017-04-26 | 2017-04-26 | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107016521A true CN107016521A (en) | 2017-08-04 |
Family
ID=59447115
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710283801.7A Pending CN107016521A (en) | 2017-04-26 | 2017-04-26 | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107016521A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107742121A (en) * | 2017-10-23 | 2018-02-27 | 国网江苏省电力公司南通供电公司 | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique |
CN109426803A (en) * | 2017-09-04 | 2019-03-05 | 三星电子株式会社 | The method and apparatus of object for identification |
CN110197276A (en) * | 2018-02-27 | 2019-09-03 | 意法半导体国际有限公司 | The data volume bruin accelerated for deep learning |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN105160310A (en) * | 2015-08-25 | 2015-12-16 | 西安电子科技大学 | 3D (three-dimensional) convolutional neural network based human body behavior recognition method |
-
2017
- 2017-04-26 CN CN201710283801.7A patent/CN107016521A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN105160310A (en) * | 2015-08-25 | 2015-12-16 | 西安电子科技大学 | 3D (three-dimensional) convolutional neural network based human body behavior recognition method |
Non-Patent Citations (1)
Title |
---|
刘煜 等著: "《稀疏表示基础理论与典型应用》", 31 October 2014 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109426803A (en) * | 2017-09-04 | 2019-03-05 | 三星电子株式会社 | The method and apparatus of object for identification |
US11804047B2 (en) | 2017-09-04 | 2023-10-31 | Samsung Electronics Co., Ltd. | Method and apparatus for recognizing object |
CN109426803B (en) * | 2017-09-04 | 2023-11-03 | 三星电子株式会社 | Method and device for identifying objects |
CN107742121A (en) * | 2017-10-23 | 2018-02-27 | 国网江苏省电力公司南通供电公司 | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique |
CN110197276A (en) * | 2018-02-27 | 2019-09-03 | 意法半导体国际有限公司 | The data volume bruin accelerated for deep learning |
CN110197276B (en) * | 2018-02-27 | 2024-03-22 | 意法半导体国际有限公司 | Data volume engraving device for deep learning acceleration |
US11977971B2 (en) | 2018-02-27 | 2024-05-07 | Stmicroelectronics International N.V. | Data volume sculptor for deep learning acceleration |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107169956B (en) | Color woven fabric defect detection method based on convolutional neural network | |
CN108446716B (en) | The PolSAR image classification method merged is indicated with sparse-low-rank subspace based on FCN | |
CN108765506A (en) | Compression method based on successively network binaryzation | |
CN111091101B (en) | High-precision pedestrian detection method, system and device based on one-step method | |
Liu et al. | A classification method of glass defect based on multiresolution and information fusion | |
Lv et al. | A novel pixel-wise defect inspection method based on stable background reconstruction | |
CN107016521A (en) | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique | |
CN103268484A (en) | Design method of classifier for high-precision face recognitio | |
CN114419413A (en) | Method for constructing sensing field self-adaptive transformer substation insulator defect detection neural network | |
CN114463759A (en) | Lightweight character detection method and device based on anchor-frame-free algorithm | |
CN110163294A (en) | Remote Sensing Imagery Change method for detecting area based on dimensionality reduction operation and convolutional network | |
CN111046213B (en) | Knowledge base construction method based on image recognition | |
CN113850284B (en) | Multi-operation detection method based on multi-scale feature fusion and multi-branch prediction | |
Roslan et al. | Real-time plastic surface defect detection using deep learning | |
Shit et al. | An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection | |
Liang et al. | YOLOD: a task decoupled network based on YOLOv5 | |
CN109284752A (en) | A kind of rapid detection method of vehicle | |
CN117037290A (en) | Human face fake detection method based on global context structure difference | |
CN107180419A (en) | A kind of medium filtering detection method based on PCA networks | |
CN107742121A (en) | A kind of warehouse nameplate recognition methods based on image convolution nerual network technique | |
CN117011219A (en) | Method, apparatus, device, storage medium and program product for detecting quality of article | |
CN115375966A (en) | Image countermeasure sample generation method and system based on joint loss function | |
Tang et al. | Industrial anomaly detection with multiscale autoencoder and deep feature extractor‐based neural network | |
CN114973122A (en) | Helmet wearing detection method based on improved YOLOv5 | |
Zhang et al. | A generative adversarial network with dual discriminators for infrared and visible image fusion based on saliency detection |
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: 20170804 |