CN109615358A - A kind of dining room automatic settlement method and system based on deep learning image recognition - Google Patents
A kind of dining room automatic settlement method and system based on deep learning image recognition Download PDFInfo
- Publication number
- CN109615358A CN109615358A CN201811295435.8A CN201811295435A CN109615358A CN 109615358 A CN109615358 A CN 109615358A CN 201811295435 A CN201811295435 A CN 201811295435A CN 109615358 A CN109615358 A CN 109615358A
- Authority
- CN
- China
- Prior art keywords
- posting
- module
- service plate
- image
- class label
- 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.)
- Granted
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
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/08—Payment architectures
- G06Q20/14—Payment architectures specially adapted for billing systems
-
- 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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/12—Hotels or restaurants
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Finance (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Multimedia (AREA)
- Development Economics (AREA)
- Image Analysis (AREA)
Abstract
A kind of dining room automatic settlement method and system based on deep learning image recognition, this system are divided into monocular cam, image processing module, pricing module and payment system.Service plate cog region complete image is acquired by monocular cam, the operation through ENet Object Detection algorithm model obtains the posting and class label data of service plate after being transmitted to image processing module, data are exported to pricing module, pricing is according to vegetable type and quantity in pricing module, total price is calculated and be shown, is paid the bill convenient for customer by payment system.The present invention carries out dining room Automatic-settlement using the image recognition technology based on deep learning, while the advantages such as, deployment flexible, strong robustness low with Normal visual technology settlement cost, overcome Normal visual technology clearing identification inadequate problem of accuracy under the influence of angle, light, the unfavorable factors such as blocking, recognizable content is also more, and practicability is stronger.
Description
Technical field
The present invention relates to food and drink clearing technical field more particularly to a kind of dining room based on deep learning image recognition are automatic
Settlement method and system.
Background technique
Food and drink is essential link in people's life, with the continuous quickening of urban life rhythm, more and more
People solves the problems, such as diet by dining room.The mainstream way in dining room is autonomous selection then unified checkout charging at present.In general,
Current dining room clearing form mainly has:
(1) traditional artificial settlement method:
Traditional artificial clearing by manually distinguishing vegetable classification, expense and amount to, have dinner peak period when the personnel that have dinner it is more,
Artificial Clearing House distinguishes computational efficiency lowly and is easy to cause mistake.
(2) settlement method based on bar code identification:
It posts bar-code label on service plate to be associated with corresponding dish valence, in the infrared or laser bar code scanner of billing terminal
It is read out to obtain phase dutiable value, then calculates total amount in a manner of manual or automatic and carry out clearing.It is artificial compared to traditional
Clearing, it is as a result more accurate reliable, but due to needing to scan service plate one by one, settlement efficiency promotion is limited, and bar-code label is easy
Damage, the effect is unsatisfactory in actual use.
(3) based on the settlement method of RFID identification:
RFID chip is embedded in service plate in advance, vegetable price is associated with respective chip, in billing terminal to chip content
It is read out to obtain corresponding dish price.The shortcomings that the method is must to use special tableware, and settlement system is at high cost, deployment
Complexity, embedded chip are easy to damage.
(4) based on the settlement method of image recognition:
The existing settlement method based on image recognition, the pixel characteristic of vegetable or the color of service plate, shape, decorative pattern is special
Sign is associated with dish valence, obtains corresponding vegetable price by image-recognizing method in billing terminal.Wherein: vegetable identification model
Complexity, discrimination is low, error-prone, and releases new dish and need to update complex model again;Knot based on service plate color, shape recognition
Calculation system, identification feature is single, and the influence of factor is blocked vulnerable to light, angle, service plate, the scene Shandong poor in face of environment-identification
Stick is very poor;For printing the identification settlement method of special pattern on service plate, need using special special tableware, the later period increases
Adding new tableware will not be general.
Summary of the invention
For the settlement method model complexity currently based on image recognition, discrimination is low, the robust under certain usage scenarios
Property is poor, actually uses defect at high cost, the present invention proposes a kind of dining room Automatic-settlement side based on deep learning image recognition
Method and system.
To achieve the above object, the technical solution used in the present invention is:
A kind of dining room automatic settlement method based on deep learning image recognition, which comprises the following steps:
(1) complete image of service plate cog region is acquired by monocular cam;
(2) complete image that will acquire is input to ENet Object Detection model system, first passes around backbone network
Network ENet extracts essential characteristic, selects the quality features convenient for positioning and identification using FFWA filter screen, then exports credible
Spend (conf) and preliminary posting, class label;
(3) it is post-processed by adaptive NMS algorithm, screens out confidence level (conf) lower posting;
(4) final posting and class label data are exported;
(5) final posting and class label data are transmitted to pricing;
(6) pricing by final posting and class label data conversion at vegetable type and quantity, according to preparatory
Vegetable total price is calculated and be shown in the dish valence of typing;
(7) client completes to pay the bill by payment system, returns to step (1).
The generation method of " posting " and " class label " and effect are respectively: input service plate cog region image
Afterwards, algorithm model identifies service plate from whole image and is marked with block diagram, and convenient for record service plate quantity, this is generated
" posting ";At the same time, algorithm identifies different types of service plate, to apply to service plate and marking convenient for distinguishing variety classes
Label, this generates " class label ".
The posting and class label data belongs to a kind of characteristic, these characteristics and specific vegetable
Type, quantity are corresponded to, and when generating specific " posting " and " class label " data, will generate corresponding vegetable kind
Class and incremental data.
A kind of system for implementing the method characterized by comprising
(1) for acquiring the monocular cam of the complete image of service plate cog region;
(2) image processing module, the complete image for acquiring monocular cam carry out image procossing, export posting
And class label data;
(3) pricing module, for by final posting and class label data conversion at vegetable type and quantity,
Vegetable total price is calculated and be shown according to the dish valence of preparatory typing;
(4) payment system completes payment according to the vegetable total price that pricing module provides for client.
The image processing module is ENet Object Detection model.
The ENet Object Detection model includes:
(a) for extracting the backbone network ENet module of complete image essential characteristic;
(b) the FFWA module of the quality features convenient for positioning and identification is selected for filter screen;
The complete image exports confidence level (conf) and preliminary positioning after the processing of ENet module and FFWA module
Frame, class label;
(c) adaptive NMS algoritic module, for after the data of ENet module and the output of FFWA module carry out
Reason, screens out confidence level (conf) lower posting, and export final posting and class label data.
Image processing module, pricing module and the payment system installation is run in computer systems.
The beneficial effects of the present invention are: using vision having using the vision technique settlement system based on deep learning
While the advantages such as technology, deployment flexible, strong robustness low compared to the cost of other clearing forms, specific depth is utilized
Habit technology makes service plate generate the feature more with expressiveness and extensive power, also has well in the poor scene of environment-identification
Robustness, service plate recognition accuracy is higher, and identifiable service plate type is also more.In addition, passing through specific adaptive N MS algorithm
The recall rate that object is blocked in detection is improved, service plate occlusion issue can be effectively solved.Furthermore for algorithm system of the invention
System, devises specific loss function, and the problems such as blocking between difficult case, object can be solved with training pattern.
Detailed description of the invention
Fig. 1 is operational flow diagram when dining room settlement system of the invention uses
Fig. 2 is the ENet Object Detection algorithm model operational flow diagram in dining room settlement system of the invention;
Fig. 3 is the principle explanatory diagram of Shuffle (Double Channel Shuffle) twice;
Fig. 4 is the principle explanatory diagram of Channel Attention mechanism;
Fig. 5 is the principle explanatory diagram of FFWA (Fast FPN with Attention).
Specific embodiment
Referring to Fig. 1-Fig. 5, a kind of dining room automatic settlement method based on deep learning image recognition of the present invention, basic ideas
It is: service plate image is obtained using monocular cam, ENet Object Detection algorithm model (system) is input to, is passing through
After trained (allowing system to do the repetition training of identification service plate, to be the general training process of deep learning) ENet Object
After Detection model calculation, " posting " and " class label " data of service plate are exported, pricing handles these data and obtains
To service plate quantity and classification, vegetable total price is calculated and be shown according to the dish valence of preparatory typing, facilitates customer payment.Described
" posting " and " class label " is respectively: after input service plate cog region image, algorithm model can be identified from whole image
Service plate is simultaneously marked with block diagram, and convenient for record service plate quantity, this generates " posting ";At the same time, algorithm identifies not
Congener service plate, to apply label to service plate convenient for distinguishing variety classes, this generates " class label ".
Specific method of the invention the following steps are included:
(1) complete image of service plate cog region is acquired by monocular cam;Service plate cog region refers to that monocular cam can be clapped
The region taken the photograph can take complete service plate image in this region.
(2) complete image that will acquire is input to ENet Object Detection model system, first passes around backbone network
Network ENet extracts essential characteristic, selects the quality features convenient for positioning and identification using FFWA filter screen, then exports credible
Spend (conf) and preliminary posting, class label.
(3) it is post-processed by adaptive NMS algorithm, screens out confidence level (conf) lower posting.Specifically
The method of screening is exactly to run adaptive NMS algorithm, and the adaptive NMS algorithm that the present invention selects is soft-NMS algorithm,
It is a kind of algorithm dedicated for post-processing.
(4) final posting and class label data are exported.
(5) final posting and class label data are transmitted to pricing.
(6) pricing by final posting and class label data conversion at vegetable type and quantity, according to preparatory
Vegetable total price is calculated and be shown in the dish valence of typing.
It is corresponding with specific vegetable type, quantity that posting and class label data are stored in the database of system
Data will generate corresponding vegetable type and quantity number when system generates specific posting and class label data
According to.
(7) client completes to pay the bill by payment system, returns to step (1).
EfficientNet backbone network (the ENet backbone network i.e. in figure) algorithm model of autonomous Design of the invention
By the efficient combination of DepthWise Conv and Group Conv, reduces parameter amount, improve arithmetic speed;To prevent table
Weaken up to power, improves the probability of quality features generation by Shuffle twice first, then pass through Channel Attention
Mechanism enhances advantageous characteristic, reduces disadvantage feature.
Shuffle (Double Channel Shuffle) twice:
As shown in Figure 3, it is assumed that have 8 convolutional channels, be then divided into 4 groups (adjacent two are classified as one group), every group respectively into
Row convolution can reduce internal storage access rate as shown in the second row in this way, improve operation, but reduce generation quality features simultaneously
A possibility that, therefore we upset 8 putting in order for channel before convolution, to produce more combinations, also
Improve a possibility that quality features generate;We upset after the completion of convolution, and by the sequence in 8 channels simultaneously, then tie
It closes Channel Attention to weight 8 channels again, be exported as final convolution feature.
Channel Attention mechanism:
Referring to fig. 4, evaluation is carried out with Sigmoid function respectively to one group of channel, obtained numerical value can be regarded as each
The weight in channel recycles this weight to be multiplied with each channel, this step is equivalent to weighted average, to improve high-quality channel
Specific gravity.
Two-way attention mechanism is the operation of Channel Attention mechanism in two directions, as Fig. 2, Fig. 5 are equal
There is embodiment, also there is the specific paraphrase for two-way attention mechanism in lower two sections of texts.
Referring to Fig. 5, feature pyramid network (FPN) can promote the performance of wisp detection, we set on this basis
FFWA (Fast FPN with Attention) has been counted, i.e., has supervised Down Layer's using the semantic information of Up Layer
External appearance characteristic screening, while every layer of feature is screened using Channel Attention, to extract quality features;But FPN increases
Model calculation amount, we utilize EfficientNet backbone network, reduce parameter amount, ensure that speed advantage.It is every in figure
One square refers to the feature of pilot process.
FFWA (Fast FPN with Attention):
Upward calculating process, i.e., normal convolution algorithm;Downward calculating process first carries out deconvolution behaviour to this layer of convolution
Make, generate the channel of size identical as lower layer, then carry out evaluation using Sigmoid, as the weight of lower channel, finally will
Lower layer's convolution and this multiplied by weight, obtain the output of this layer.
Lower layer's convolution has more low-level image features, and upper layer convolution has more semantic features, can effectively improve in this way
The detection of obvious object, while the feature of bottom are also beneficial to detect more wisps, entirely detect to improve
Recall rate in journey, avoids missing inspection.
Since the food in service plate is changing always, identify that service plate increases very big difficulty for model, so we set
It is difficult to service plate identification bring to solve cosmetic variation to have counted adaptive N ormalization, mainly in conjunction with Batch
Normalization and Instance Normalization, then respectively with multiplied by weight, then two kinds of results addeds, by
There is invariance to interior container in Batch Normalization, Instance Normalization has invariance to appearance,
So combining the two and finding a balance weight, the generalization ability of model can be largely improved.
Adaptive N ormalization (adaptive normalization):
BN (batch normalization): internal container has invariance, is conducive to service plate classification and judges;IN (example normalization): to appearance
With invariance, conducive to the appearance migration in the changeable situation of the external appearance characteristics such as color, shape, size.For service plate identification process
Middle food is changeable and characteristic that service plate is constant, becomes feasible in conjunction with both technologies.In conjunction with method: out=alpha*BN+
beta*IN.The two parameters of alpha and beta obtain a final normalizing every time according to data auto scaling itself
Change as a result, the advantages of thus effectively combining the two.
It is centainly blocked due to being had between service plate, will lead to subsequent service plate Primary Location frame quilt using common NMS algorithm
It deletes, therefore is post-processed using adaptive NMS algorithm, i.e. soft-NMS, line is used on the basis of former NMS algorithmic function
Property weighting:
Wherein M is the maximum posting of present score, and Nt is to inhibit threshold value, Si sore.
Have benefited from soft-NMS, we retain ineligible candidate frame, and give smaller score, to improve
The recall rate of object is blocked in detection, can effectively solve service plate occlusion issue.
, can be bad to part service plate recognition effect during model training, we improve loss function, allow model can
Preferably to learn the feature of these difficult cases:
Y=- (1-p) * Log (X)
Wherein: X, that is, model output, the corresponding probability of p, that is, service plate, Y, that is, loss function
Classical loss function does not have (1-p) this factor, and this factor that we add can increase difficult case
Loss, so that model be allowed to be more concerned about these difficult cases.
During automatic settlement method of the invention, the complete of required service plate is acquired by monocular cam in step (1)
After image, (filtering, noise reduction, white balance, distortion processing, radiation variation etc.) can be simply pre-processed, to image to be promoted
Image recognition accuracy rate.
The present invention can acquire user's face using other cameras while above-mentioned monocular cam obtains service plate image
The face-image of acquisition and pre-stored user images are compared the account information to confirm active user by image.
Claims (7)
1. a kind of dining room automatic settlement method based on deep learning image recognition, which comprises the following steps:
(1) complete image of service plate cog region is acquired by monocular cam;
(2) complete image that will acquire is input to ENet Object Detection model system, first passes around backbone network
ENet extracts essential characteristic, selects the quality features convenient for positioning and identification using FFWA filter screen, then exports confidence level
(conf) and preliminary posting, class label;
(3) it is post-processed by adaptive NMS algorithm, screens out confidence level (conf) lower posting;
(4) final posting and class label data are exported;
(5) final posting and class label data are transmitted to pricing;
(6) pricing by final posting and class label data conversion at vegetable type and quantity, according to preparatory typing
Dish valence vegetable total price is calculated and be shown;
(7) client completes to pay the bill by payment system, returns to step (1).
2. the method as described in claim 1, which is characterized in that the generation method of " posting " and " class label " and
Effect is respectively: after input service plate cog region image, algorithm model identifies service plate from whole image and is marked with block diagram
Come, convenient for record service plate quantity, this generates " posting ";At the same time, algorithm identifies different types of service plate, for convenient for
It distinguishes variety classes and label is applied to service plate, this generates " class label ".
3. the method as described in claim 1, which is characterized in that the posting and class label data belongs to a kind of feature
Data, these characteristics are carried out with specific vegetable type, quantity it is corresponding, when generating specific " posting " and " classification
When label " data, corresponding vegetable type and incremental data will be generated.
4. a kind of system for implementing claim 1 the method characterized by comprising
(1) for acquiring the monocular cam of the complete image of service plate cog region;
(2) image processing module, the complete image for acquiring monocular cam carry out image procossing, export posting and class
Distinguishing label data;
(3) pricing module, for by final posting and class label data conversion at vegetable type and quantity, according to
Vegetable total price is calculated and be shown in the dish valence of preparatory typing;
(4) payment system completes payment according to the vegetable total price that pricing module provides for client.
5. system as claimed in claim 4, which is characterized in that the image processing module is ENet Object
Detection model.
6. system as claimed in claim 5, which is characterized in that the ENet Object Detection model includes:
(a) for extracting the backbone network ENet module of complete image essential characteristic;
(b) the FFWA module of the quality features convenient for positioning and identification is selected for filter screen;
The complete image exports confidence level (conf) and preliminary posting, class after the processing of ENet module and FFWA module
Distinguishing label;
(c) adaptive NMS algoritic module, for being post-processed to the data exported by ENet module and FFWA module,
Confidence level (conf) lower posting is screened out, and exports final posting and class label data.
7. system as claimed in claim 4, which is characterized in that image processing module, pricing module and the payment
System installation is run in computer systems.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811295435.8A CN109615358B (en) | 2018-11-01 | 2018-11-01 | Deep learning image recognition-based restaurant automatic settlement method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811295435.8A CN109615358B (en) | 2018-11-01 | 2018-11-01 | Deep learning image recognition-based restaurant automatic settlement method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109615358A true CN109615358A (en) | 2019-04-12 |
CN109615358B CN109615358B (en) | 2020-11-03 |
Family
ID=66002230
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811295435.8A Active CN109615358B (en) | 2018-11-01 | 2018-11-01 | Deep learning image recognition-based restaurant automatic settlement method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109615358B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110580505A (en) * | 2019-08-29 | 2019-12-17 | 杭州火小二科技有限公司 | Intelligent cash registering method based on service plate identification |
CN110889859A (en) * | 2019-11-11 | 2020-03-17 | 珠海上工医信科技有限公司 | U-shaped network for fundus image blood vessel segmentation |
CN111047799A (en) * | 2019-11-29 | 2020-04-21 | 北京三快在线科技有限公司 | Cash register device, article identification method and storage medium |
CN111640268A (en) * | 2020-04-22 | 2020-09-08 | 深圳拓邦股份有限公司 | Intelligent settlement method and system based on dinner plate shape and color |
CN113455660A (en) * | 2021-05-28 | 2021-10-01 | 天津博诺智创机器人技术有限公司 | Intelligent food cooperation method and breakfast preparation system applying same |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140003668A1 (en) * | 2000-11-06 | 2014-01-02 | Nant Holdings Ip, Llc | Image Capture and Identification System and Process |
CN107064042A (en) * | 2017-03-02 | 2017-08-18 | 中国科学院合肥物质科学研究院 | The method for qualitative analysis of infrared spectrum |
CN107122730A (en) * | 2017-04-24 | 2017-09-01 | 乐金伟 | Free dining room automatic price method |
CN107622183A (en) * | 2017-08-15 | 2018-01-23 | 上海派森诺生物科技股份有限公司 | A kind of fetal chromosomal Ploidy detection analysis method based on multiple indexes |
CN107992871A (en) * | 2017-12-21 | 2018-05-04 | 陕西伟景机器人科技有限公司 | The automatic accounting method in dining room based on image recognition |
CN108268869A (en) * | 2018-02-13 | 2018-07-10 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
CN108647591A (en) * | 2018-04-25 | 2018-10-12 | 长沙学院 | Activity recognition method and system in a kind of video of view-based access control model-semantic feature |
-
2018
- 2018-11-01 CN CN201811295435.8A patent/CN109615358B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140003668A1 (en) * | 2000-11-06 | 2014-01-02 | Nant Holdings Ip, Llc | Image Capture and Identification System and Process |
CN107064042A (en) * | 2017-03-02 | 2017-08-18 | 中国科学院合肥物质科学研究院 | The method for qualitative analysis of infrared spectrum |
CN107122730A (en) * | 2017-04-24 | 2017-09-01 | 乐金伟 | Free dining room automatic price method |
CN107622183A (en) * | 2017-08-15 | 2018-01-23 | 上海派森诺生物科技股份有限公司 | A kind of fetal chromosomal Ploidy detection analysis method based on multiple indexes |
CN107992871A (en) * | 2017-12-21 | 2018-05-04 | 陕西伟景机器人科技有限公司 | The automatic accounting method in dining room based on image recognition |
CN108268869A (en) * | 2018-02-13 | 2018-07-10 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
CN108647591A (en) * | 2018-04-25 | 2018-10-12 | 长沙学院 | Activity recognition method and system in a kind of video of view-based access control model-semantic feature |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110580505A (en) * | 2019-08-29 | 2019-12-17 | 杭州火小二科技有限公司 | Intelligent cash registering method based on service plate identification |
CN110889859A (en) * | 2019-11-11 | 2020-03-17 | 珠海上工医信科技有限公司 | U-shaped network for fundus image blood vessel segmentation |
CN111047799A (en) * | 2019-11-29 | 2020-04-21 | 北京三快在线科技有限公司 | Cash register device, article identification method and storage medium |
CN111640268A (en) * | 2020-04-22 | 2020-09-08 | 深圳拓邦股份有限公司 | Intelligent settlement method and system based on dinner plate shape and color |
CN113455660A (en) * | 2021-05-28 | 2021-10-01 | 天津博诺智创机器人技术有限公司 | Intelligent food cooperation method and breakfast preparation system applying same |
Also Published As
Publication number | Publication date |
---|---|
CN109615358B (en) | 2020-11-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109615358A (en) | A kind of dining room automatic settlement method and system based on deep learning image recognition | |
CN103208156B (en) | A kind of food and drink Automated Clearing House system and method | |
CN103902961B (en) | Face recognition method and device | |
CN107122730A (en) | Free dining room automatic price method | |
CN106599869A (en) | Vehicle attribute identification method based on multi-task convolutional neural network | |
Zhou et al. | Convnets vs. transformers: Whose visual representations are more transferable? | |
CN103839084B (en) | Multi-kernel support vector machine multi-instance learning algorithm applied to pedestrian re-identification | |
CN109840530A (en) | The method and apparatus of training multi-tag disaggregated model | |
CN109584248A (en) | Infrared surface object instance dividing method based on Fusion Features and dense connection network | |
CN107610123A (en) | A kind of image aesthetic quality evaluation method based on depth convolutional neural networks | |
CN101576951B (en) | Iris external boundary positioning method based on shades of gray and classifier | |
CN108427953A (en) | A kind of character recognition method and device | |
CN102156885B (en) | Image classification method based on cascaded codebook generation | |
CN108388905B (en) | A kind of Illuminant estimation method based on convolutional neural networks and neighbourhood context | |
CN106529380A (en) | Image identification method and device | |
CN104504362A (en) | Face detection method based on convolutional neural network | |
CN109508664A (en) | A kind of vegetable identification pricing method based on deep learning | |
CN108875669A (en) | A kind of commodity identification technology merged based on visible light with infrared image | |
Björklund et al. | Automatic license plate recognition with convolutional neural networks trained on synthetic data | |
CN106898011B (en) | A method of determining convolutional neural networks convolution nuclear volume based on edge detection | |
CN108038504A (en) | A kind of method for parsing property ownership certificate photo content | |
CN104361357A (en) | Photo set classification system and method based on picture content analysis | |
CN104063686A (en) | System and method for performing interactive diagnosis on crop leaf segment disease images | |
CN109102029A (en) | Information, which maximizes, generates confrontation network model synthesis face sample quality appraisal procedure | |
CN110020603A (en) | Self-checkout system and method based on convolutional neural networks |
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 | ||
CB02 | Change of applicant information |
Address after: 100094 603, 7 / F, block A2, No.1 Zhongguancun, No.81 Beiqing Road, Haidian District, Beijing Applicant after: BEIJING VIZUM INTELLIGENT TECHNOLOGY Co.,Ltd. Address before: 100088 Beijing City, Haidian District Zhichun Road Jinqiu International Building No. 6 A block 1311 Applicant before: BEIJING VIZUM INTELLIGENT TECHNOLOGY Co.,Ltd. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CB03 | Change of inventor or designer information |
Inventor after: Fang Lian Inventor after: Qian Chaochao Inventor before: Fang Lian |
|
CB03 | Change of inventor or designer information |