CN110458096A - A kind of extensive commodity recognition method based on deep learning - Google Patents
A kind of extensive commodity recognition method based on deep learning Download PDFInfo
- Publication number
- CN110458096A CN110458096A CN201910736313.6A CN201910736313A CN110458096A CN 110458096 A CN110458096 A CN 110458096A CN 201910736313 A CN201910736313 A CN 201910736313A CN 110458096 A CN110458096 A CN 110458096A
- Authority
- CN
- China
- Prior art keywords
- commodity
- data
- picture
- detection
- network
- 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 34
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 66
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000013145 classification model Methods 0.000 claims abstract description 23
- 238000012360 testing method Methods 0.000 claims description 26
- 238000013480 data collection Methods 0.000 claims description 9
- 230000002708 enhancing effect Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 6
- 230000001965 increasing effect Effects 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to image identification technical fields, disclose a kind of extensive commodity recognition method based on deep learning, include the following steps: S1: establishing commodity detection model;S2: being input to commodity detection model for picture to be detected, obtains commodity attribute data all in picture to be detected;S3: commodity classification model is established;S4: being input to commodity classification model for picture to be detected, according to all commodity attribute data, obtains corresponding merchandise classification data.The present invention solve the problems, such as precision of the existing technology be difficult to meet large-scale project demand, training sample demand greatly, can not iteratively faster updates, sample reusability is low, equipment requirement is high, feature representation ability is limited, workload is huge and arithmetic speed is slow.
Description
Technical field
The invention belongs to image identification technical fields, and in particular to a kind of extensive commodity identification side based on deep learning
Method.
Background technique
Commodity Automatic-settlement is mainly according to the information in picture, using object detection method, information in abstract image,
Detect the items list contained in image.Under normal circumstances, requiring can be to more commodity image (containing a plurality of in image
Different commodity) detection identification is carried out, finally obtain items list in image.Object detection task is necessary not only for identification static map
There is any object as in, is what classification, it is also necessary to predict the position where object.In object detecting areas, it is common to use
Target detection or Target Segmentation method realize positioning and classification to object space in a module, finally identify static map
All target objects as in.
Disadvantage of the existing technology:
1) the general objectives detection of the prior art or dividing method, this for small-scale (target category is smaller) commodity from
Dynamic clearing project, can achieve preferable effect, but the extensive commodity Automatic-settlement item more huge for merchandise classification
Mesh but seems gradually out of strength;
2) precision is difficult to meet large-scale project demand;For the prior art when identifying that number is thousands of, model training difficulty is big
Big to increase, precision also declines therewith, it is difficult to guarantee stability, be unable to reach project commercialization requirement;
3) training sample demand is very big, and acquisition cost increases severely;The sample number acquired for increasing commodity training stage needs newly
Mesh is very big, greatly increases the cost manually marked;
It 4) can not iteratively faster update.With the increase of identification number, training speed can die-off therewith, while frequency of training
It is required that be consequently increased, eventually lead to that the model modification period is longer, the iteratively faster characteristic in market can not be adapted to;
5) sample reusability is low;Training sample coupling used in the prior art is strong, i.e., multiple objects are schemed at same,
Leading to data between object, there are certain relationships, cannot achieve complete decoupling, eventually lead to the data that high cost collects,
It can only serve certain particular demands, all scenes can not be suitable for;
6) equipment requirement is high;The prior art will be positioned and is integrated into a module with classification, to equipment requirement (in such as equipment
Deposit, video memory) as identification target category increases and very fast growth, and then equipment cost is caused to greatly improve;Simultaneously particular for height
The application demand of hair, fast response time requires cluster device higher;
7) for traditional detection and classification method such as SVM, relative depth study is since structure is simple, feature representation ability
It is limited and large-scale target identification can not be coped with;
8) it needs to be split each commodity when making training sample, this will bring huge workload, and operation
Speed is slow.
Summary of the invention
In order to solve the above problems existing in the present technology, it is an object of that present invention to provide a kind of based on the big of deep learning
Scale commodity recognition method is difficult to meet large-scale project demand, training sample need for solving precision of the existing technology
Ask greatly, can not iteratively faster update, sample reusability is low, equipment requirement is high, feature representation ability is limited, workload it is huge with
And the problem that arithmetic speed is slow.
The technical scheme adopted by the invention is as follows:
A kind of extensive commodity recognition method based on deep learning, includes the following steps:
S1: commodity detection model is established;
S2: being input to commodity detection model for picture to be detected, obtains commodity attribute data all in picture to be detected;
S3: commodity classification model is established;
S4: being input to commodity classification model for picture to be detected, according to all commodity attribute data, obtains corresponding quotient
Product categorical data.
Further, in step S1, the specific method for establishing commodity detection model includes the following steps:
S1-1: existing more more commodity data collection of scene are subjected to data enhancing processing;
S1-2: data set input detection network after enhancing is iterated training;
S1-3: judging whether to meet detection network iteration termination condition, no if then exporting optimal commodity detection model
Then return step S1-2;
It currently hands over and changes than reaching default detection network friendship and reaching default detection network than threshold value or current iteration number
For frequency threshold value, meet detection network iteration termination condition.
Further, in step S1-1, the more commodity data collection of more scenes are divided into detection network training collection and detection network
Test set, detection network training collection include no less than 90000 trained pictures, include no less than 1300 class commodity in training picture
And contextual data;
Detecting network test collection includes no less than 20000 test pictures, includes no less than 1000 class quotient in test picture
Product and contextual data.
Further, in step S1-1, the method for data enhancing processing includes that rotation, cutting, translation, mirror image and brightness change
Become.
Further, in step S1-2, training is iterated using SoftNMS algorithm.
Further, in step S3, the specific method for establishing commodity classification model includes the following steps:
S3-1: existing commodity list class data set is added in negative sample;
S3-2: commodity list class data set is inputted into sorter network, is iterated training;
S3-3: judging whether to meet sorter network iteration termination condition, no if then exporting optimal commodity classification model
Then return step S3-2;
When friendship and than reaching default sorter network friendship and reaching default sorter network iteration than threshold value or current iteration number
Frequency threshold value meets sorter network iteration termination condition.
Further, in step S3-1, commodity list class data set is divided into sorter network training set and sorter network is tested
Collection, sorter network training set include no less than 90000 trained pictures, include being no less than 1300 class commodity and field in training picture
Scape data, and the training picture number of every class commodity belongs to 60-100;
Sorter network test set includes no less than 20000 test pictures, includes no less than 1000 class quotient in test picture
Product and contextual data, and the test picture number of every class commodity belongs to 60-100.
Further, the extensive commodity recognition method based on deep learning is further comprising the steps of:
New commodity data set is periodically acquired, and updates the more commodity data collection of more scenes and commodity list class data set, wherein
The new commodity picture number of the new commodity data set is no less than 60.
Further, in step S4, the merchandise classification and the merchandise classification in commodity list class data set of merchandise classification data
It is consistent.
The invention has the benefit that
(1) precision improves, and reaches commercial standard (CS);Identification module is divided into two, the disparate modules net optimal using the direction
Network structure is trained;Training difficulty substantially reduces, and model accuracy is promoted, and recognition capability is more stable;
(2) commodity occlusion issue is solved;When commodity are blocked less than 50%, detection network still is able to effectively
It detected;
(3) training sample needed for is reduced, acquisition cost decline;For increasing commodity newly, it is only necessary to every class acquire 60 or with
On single-item data for classify;Since commodity detection module can effectively detect all commodity (no matter new and old)
Positioning, so there is no need to increase detection module training data newly;Comparison with general objectives detection method, new samples collecting quantity significantly under
Drop, acquisition cost also decline therewith;
(4) iteratively faster updates;Since commodity detection module can effectively detect all commodity (no matter new and old)
Positioning so there is no need to increase detection module training data newly, therefore only needs to acquire a certain amount of data, and then updates object classification module
, market renewal speed can be kept up with;
(5) data are uncorrelated between classification, can reuse;Since detection module is indifferent to object category, classify simultaneously
Data needed for module are single-item data, even if the interim undercarriage of certain commodity, it is only necessary to such commodity be picked out training set, update classification
Modular model will not influence other commodity.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the flow chart of the extensive commodity recognition method based on deep learning.
Specific embodiment
With reference to the accompanying drawing and specific embodiment come the present invention is further elaborated.It should be noted that for
Although the explanation of these way of example is to be used to help understand the present invention, but and do not constitute a limitation of the invention.The present invention
Disclosed function detail is only used for description example embodiments of the present invention.However, this hair can be embodied with many alternative forms
It is bright, and be not construed as limiting the invention in the embodiment that the present invention illustrates.
It should be appreciated that terminology used in the present invention is only used for description specific embodiment, it is not intended to limit of the invention show
Example embodiment.If term " includes ", " including ", "comprising" and/or " containing " are used in the present invention, institute's sound is specified
Bright feature, integer, step, operation, unit and/or component existence, and be not excluded for one or more other features, number
Amount, step, operation, unit, component and/or their combination existence or increase.
It should be appreciated that it will be further noted that the function action occurred may go out with attached drawing in some alternative embodiments
Existing sequence is different.Such as related function action is depended on, it can actually substantially be executed concurrently, or sometimes
Two figures continuously shown can be executed in reverse order.
It should be appreciated that providing specific details, in the following description in order to which example embodiment is understood completely.
However those of ordinary skill in the art are it is to be understood that implementation example embodiment without these specific details.
Such as system can be shown in block diagrams, to avoid with unnecessary details come so that example is unclear.In other instances, may be used
Or not show well-known process, structure and technology unnecessary details, to avoid making example embodiment unclear.
Embodiment 1:
As shown in Figure 1, a kind of extensive commodity recognition method based on deep learning, includes the following steps:
S1: establishing commodity detection model, and specific method includes the following steps:
S1-1: existing more more commodity data collection of scene are subjected to data enhancing processing, more more commodity data Ji Bao of scene
Include the picture of a variety of difference commodity under multiple and different backgrounds and light environment;
The more commodity data collection of more scenes are divided into detection network training collection and detection network test collection, detect network training collection
It include being no less than 1300 class commodity and contextual data in training picture including being no less than 90000 trained pictures;
Detecting network test collection includes no less than 20000 test pictures, includes no less than 1000 class quotient in test picture
Product and contextual data;
The method of data enhancing processing includes that rotation, cutting, translation, mirror image and brightness change;
S1-2: by data set input detection network after enhancing, training is iterated using SoftNMS algorithm;
It detects Web vector graphic Faster Rcnn inception resnet and SoftNMS and detects network structure model, essence
Exactness and recall rate are more than 96%;
Traditional detection frame suppressing method (non-maxima suppression NMS) will directly be greater than with the IOU of the maximum detection block of score
The score zero setting of other frames of some threshold value, this makes two lean on relatively close or has the target blocked a little that can only detect one
A, accuracy is low;Detection Web vector graphic SoftNMS algorithm of the invention replaces traditional NMS algorithm, increases the standard of detection block
True property;SoftNMS target detection frame updates score according to the IOU of maximum score detection block, and the frame score for making IOU bigger is lower,
So that overlapping frame " development " is not completely inhibited, to increase the detection recall rate for blocking commodity;
S1-3: judging whether to meet detection network iteration termination condition, no if then exporting optimal commodity detection model
Then return step S1-2;
It currently hands over and changes than reaching default detection network friendship and reaching default detection network than threshold value or current iteration number
For frequency threshold value, meet detection network iteration termination condition;
S2: being input to commodity detection model for picture to be detected, obtains commodity attribute data all in picture to be detected;
Preliminary experiment is carried out to merchandise classification number respectively herein, has selected 300,600,1000,1300,2000 class quotient respectively
Product train sorter network, experiments have shown that keep accuracy that must increase a of classifier sample while merchandise classification increases
Number, the number of samples for increasing similar commodity when existing simultaneously similar commodity are conducive to improve the discrimination of similar commodity;Through
It crosses the preliminary experiment present invention and obtains empirical value such as the following table 1 merchandise classification and sample size experience of merchandise classification and every class number of samples
It is worth shown in result table, solves the problems, such as that model is difficult to restrain when traditional detection method is greater than 1000 for type of merchandize;
Table 1
Class number N | N≤600 | 600≤N≤1300 | 1300≤N≤2000 | N≥2000 |
Required number of pictures n | n>60 | n>80 | n>150 | N>200 |
S3: establishing commodity classification model, and different commodity classification models can be trained for different application scenarios, is increased
The flexibility of commodity classification model;Commodity classification model is instructed using inception_v4 as the basic network of sorter network
Practice, accuracy is more than 98%, and specific method includes the following steps:
S3-1: being added existing commodity list class data set for negative sample, commodity list class data set be included in different background and
The data of particular commodity are shot under light environment, commodity put posture, shooting angle stochastic transformation in commodity list class data set;
Commodity list class data set is divided into sorter network training set and sorter network test set, sorter network training set includes
No less than 90000 trained pictures, interior training picture includes no less than 1300 class commodity and contextual data, and the instruction of every class commodity
Practice picture number and belongs to 60-100;
Sorter network test set includes no less than 20000 test pictures, includes no less than 1000 class quotient in test picture
Product and contextual data, and the test picture number of every class commodity belongs to 60-100;
S3-2: commodity list class data set is inputted into sorter network, is iterated training;
S3-3: judging whether to meet sorter network iteration termination condition, no if then exporting optimal commodity classification model
Then return step S3-2;
When friendship and than reaching default sorter network friendship and reaching default sorter network iteration than threshold value or current iteration number
Frequency threshold value meets sorter network iteration termination condition;
The degree of automation of commodity classification model is improved, need to only collect the figure of the good every class commodity for needing to identify as required
Piece can start that sorter network is trained and the publication of commodity classification model, and work simplification when use, any non-technical personnel is ok
Commodity classification model is manipulated, so that model is more flexible;
S4: being input to commodity classification model for picture to be detected, according to all commodity attribute data, obtains corresponding quotient
Product categorical data;
The merchandise classification of merchandise classification data is consistent with the merchandise classification in commodity list class data set.
Preferably, the extensive commodity recognition method based on deep learning is further comprising the steps of:
New commodity data set is periodically acquired, and updates the more commodity data collection of more scenes and commodity list class data set, wherein
The new commodity picture number of the new commodity data set is no less than 60;
Since exterior of commodity updates, iteration is fast, and commodity classification model quickly learns new commodity to seem especially heavy
It wants, the present invention need to only collect the sample and training commodity classification model of new commodity for increasing commodity newly, allow commodity classification mould
Type has good scalability, flexibility and controllability to newly-increased commodity, reduces workload and time cost.
Present invention uses commodity detection model and commodity classification model, recognition effect and stability, and can quickly change
In generation, updates, and provides strong technical support for commodity Automatic-settlement task, and it is distinctive effectively to solve commodity Automatic-settlement scene
Technological difficulties are only responsible for carrying out the target object in still image wherein detection network is trained using large-scale data
Detection and positioning, but do not need to pay close attention to the target object that is, even if the detection network based on mass data training is not in face of
The scene known all has good anti-interference ability;Cascade network of the sorter network as detection network is only responsible for detection net
The target object that network detects carries out classification judgement, and commodity Detection task is divided into two, and is guaranteeing higher recognition correct rate
Under the premise of, greatly improve the extended capability of model;Detect network can be suitable for different application scenarios and it is some not
The commodity known, therefore only need to optimize commodity classification model for different application scenarios, which simplify the works of model modification
It measures;Even if the detection network based on mass data training all has good anti-interference ability in face of unknown scene.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are each under the inspiration of the present invention
The product of kind form.Above-mentioned specific embodiment should not be understood the limitation of pairs of protection scope of the present invention, protection of the invention
Range should be subject to be defined in claims, and specification can be used for interpreting the claims.
Claims (9)
1. a kind of extensive commodity recognition method based on deep learning, characterized by the following steps:
S1: commodity detection model is established;
S2: being input to commodity detection model for picture to be detected, obtains commodity attribute data all in picture to be detected;
S3: commodity classification model is established;
S4: being input to commodity classification model for picture to be detected, according to all commodity attribute data, obtains corresponding commodity class
Other data.
2. the extensive commodity recognition method according to claim 1 based on deep learning, it is characterised in that: the step
In rapid S1, the specific method for establishing commodity detection model includes the following steps:
S1-1: existing more more commodity data collection of scene are subjected to data enhancing processing;
S1-2: data set input detection network after enhancing is iterated training;
S1-3: judge whether otherwise meet detection network iteration termination condition returns if then exporting optimal commodity detection model
Return step S1-2;
It currently hands over and than reaching default detection network friendship and reaching default detection network iteration time than threshold value or current iteration number
Number threshold value meets detection network iteration termination condition.
3. the extensive commodity recognition method according to claim 2 based on deep learning, it is characterised in that: the step
In rapid S1-1, the more commodity data collection of more scenes are divided into detection network training collection and detection network test collection, the detection net
Network training set includes no less than 90000 trained pictures, includes being no less than 1300 class commodity and scene in the training picture
Data;
The detection network test collection includes no less than 20000 test pictures, includes no less than in the test picture
1000 class commodity and contextual data.
4. the extensive commodity recognition method according to claim 2 based on deep learning, it is characterised in that: the step
Suddenly in S1-1, the method for data enhancing processing includes that rotation, cutting, translation, mirror image and brightness change.
5. the extensive commodity recognition method according to claim 2 based on deep learning, it is characterised in that: the step
In rapid S1-2, training is iterated using SoftNMS algorithm.
6. the extensive commodity recognition method according to claim 1 based on deep learning, it is characterised in that: the step
In rapid S3, the specific method for establishing commodity classification model includes the following steps:
S3-1: existing commodity list class data set is added in negative sample;
S3-2: commodity list class data set is inputted into sorter network, is iterated training;
S3-3: judge whether to meet sorter network iteration termination condition, if then exporting optimal commodity classification model, otherwise return
Return step S3-2;
When friendship and than reaching default sorter network friendship and reaching default sorter network the number of iterations than threshold value or current iteration number
Threshold value meets sorter network iteration termination condition.
7. the extensive commodity recognition method according to claim 6 based on deep learning, it is characterised in that: the step
In rapid S3-1, commodity list class data set is divided into sorter network training set and sorter network test set, the sorter network instruction
To practice and collects including no less than 90000 trained pictures, the interior training picture includes being no less than 1300 class commodity and contextual data,
And the training picture number of every class commodity belongs to 60-100;
The sorter network test set includes no less than 20000 test pictures, includes no less than in the test picture
1000 class commodity and contextual data, and the test picture number of every class commodity belongs to 60-100.
8. the extensive commodity recognition method according to claim 1 based on deep learning, it is characterised in that: further include with
Lower step:
New commodity data set is periodically acquired, and updates the more commodity data collection of more scenes and commodity list class data set, wherein is described
The new commodity picture number of new commodity data set be no less than 60.
9. the extensive commodity recognition method according to claim 1 based on deep learning, it is characterised in that: the step
In rapid S4, merchandise classification and the merchandise classification in commodity list class data set of merchandise classification data are consistent.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910736313.6A CN110458096A (en) | 2019-08-09 | 2019-08-09 | A kind of extensive commodity recognition method based on deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910736313.6A CN110458096A (en) | 2019-08-09 | 2019-08-09 | A kind of extensive commodity recognition method based on deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110458096A true CN110458096A (en) | 2019-11-15 |
Family
ID=68485750
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910736313.6A Pending CN110458096A (en) | 2019-08-09 | 2019-08-09 | A kind of extensive commodity recognition method based on deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110458096A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111062786A (en) * | 2019-12-25 | 2020-04-24 | 创新奇智(青岛)科技有限公司 | Model updating method based on establishment of commodity appearance characteristic mapping table |
CN111104924A (en) * | 2019-12-31 | 2020-05-05 | 上海品览数据科技有限公司 | Processing algorithm for effectively identifying low-resolution commodity image |
CN111553726A (en) * | 2020-04-22 | 2020-08-18 | 上海海事大学 | HMM-based (hidden Markov model) -based system and method for predicting bill swiping |
CN111797896A (en) * | 2020-06-01 | 2020-10-20 | 锐捷网络股份有限公司 | Commodity identification method and device based on intelligent baking |
CN113095383A (en) * | 2021-03-30 | 2021-07-09 | 广州图匠数据科技有限公司 | Auxiliary sale material identification method and device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110309911A1 (en) * | 2009-02-10 | 2011-12-22 | Martin Kemper | Apparatus for detecting and processing data in cash desk |
CN107045641A (en) * | 2017-04-26 | 2017-08-15 | 广州图匠数据科技有限公司 | A kind of identification of pallets method based on image recognition technology |
US20180240089A1 (en) * | 2017-02-21 | 2018-08-23 | Toshiba Tec Kabushiki Kaisha | Inventory management computer system |
CN108734162A (en) * | 2018-04-12 | 2018-11-02 | 上海扩博智能技术有限公司 | Target identification method, system, equipment and storage medium in commodity image |
CN108921198A (en) * | 2018-06-08 | 2018-11-30 | 山东师范大学 | commodity image classification method, server and system based on deep learning |
CN109522967A (en) * | 2018-11-28 | 2019-03-26 | 广州逗号智能零售有限公司 | A kind of commodity attribute recognition methods, device, equipment and storage medium |
-
2019
- 2019-08-09 CN CN201910736313.6A patent/CN110458096A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110309911A1 (en) * | 2009-02-10 | 2011-12-22 | Martin Kemper | Apparatus for detecting and processing data in cash desk |
US20180240089A1 (en) * | 2017-02-21 | 2018-08-23 | Toshiba Tec Kabushiki Kaisha | Inventory management computer system |
CN107045641A (en) * | 2017-04-26 | 2017-08-15 | 广州图匠数据科技有限公司 | A kind of identification of pallets method based on image recognition technology |
CN108734162A (en) * | 2018-04-12 | 2018-11-02 | 上海扩博智能技术有限公司 | Target identification method, system, equipment and storage medium in commodity image |
CN108921198A (en) * | 2018-06-08 | 2018-11-30 | 山东师范大学 | commodity image classification method, server and system based on deep learning |
CN109522967A (en) * | 2018-11-28 | 2019-03-26 | 广州逗号智能零售有限公司 | A kind of commodity attribute recognition methods, device, equipment and storage medium |
Non-Patent Citations (3)
Title |
---|
NAVANEETH BODLA 等: "Improving Object Detection With One Line of Code", 《ARXIV》 * |
潘婷 等: "基于卷积神经网络的车辆和行人检测算法", 《浙江科技学院学报》 * |
胡正委 等: "基于改进Faster RCNN与Grabcut的商品图像检测", 《计算机系统应用》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111062786A (en) * | 2019-12-25 | 2020-04-24 | 创新奇智(青岛)科技有限公司 | Model updating method based on establishment of commodity appearance characteristic mapping table |
CN111104924A (en) * | 2019-12-31 | 2020-05-05 | 上海品览数据科技有限公司 | Processing algorithm for effectively identifying low-resolution commodity image |
CN111104924B (en) * | 2019-12-31 | 2023-09-01 | 上海品览数据科技有限公司 | Processing algorithm for identifying low-resolution commodity image |
CN111553726A (en) * | 2020-04-22 | 2020-08-18 | 上海海事大学 | HMM-based (hidden Markov model) -based system and method for predicting bill swiping |
CN111553726B (en) * | 2020-04-22 | 2023-04-28 | 上海海事大学 | HMM-based bill-of-brush prediction system and method |
CN111797896A (en) * | 2020-06-01 | 2020-10-20 | 锐捷网络股份有限公司 | Commodity identification method and device based on intelligent baking |
CN113095383A (en) * | 2021-03-30 | 2021-07-09 | 广州图匠数据科技有限公司 | Auxiliary sale material identification method and device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110458096A (en) | A kind of extensive commodity recognition method based on deep learning | |
CN109740588B (en) | X-ray picture contraband positioning method based on weak supervision and deep response redistribution | |
CN109359515A (en) | A kind of method and device that the attributive character for target object is identified | |
CN110232379A (en) | A kind of vehicle attitude detection method and system | |
CN111461164B (en) | Sample data set capacity expansion method and model training method | |
WO2023284465A1 (en) | Image detection method and apparatus, computer-readable storage medium, and computer device | |
Shuai et al. | Object detection system based on SSD algorithm | |
EP4209959A1 (en) | Target identification method and apparatus, and electronic device | |
CN112613569A (en) | Image recognition method, and training method and device of image classification model | |
CN102314591B (en) | Method and equipment for detecting static foreground object | |
CN108664986A (en) | Based on lpThe multi-task learning image classification method and system of norm regularization | |
CN112087316B (en) | Network anomaly root cause positioning method based on anomaly data analysis | |
CN113780342A (en) | Intelligent detection method and device based on self-supervision pre-training and robot | |
CN111144215A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
CN104573701B (en) | A kind of automatic testing method of Tassel of Corn | |
CN104966109A (en) | Medical laboratory report image classification method and apparatus | |
Cheng et al. | Water quality monitoring method based on TLD 3D fish tracking and XGBoost | |
CN112200862A (en) | Training method of target detection model, target detection method and device | |
CN113065447A (en) | Method and equipment for automatically identifying commodities in image set | |
CN112784102A (en) | Video retrieval method and device and electronic equipment | |
CN112668365A (en) | Material warehousing identification method, device, equipment and storage medium | |
CN111160170A (en) | Self-learning human behavior identification and anomaly detection method | |
CN116311190A (en) | Clothing type detection and identification method based on YOLOv5s | |
CN110428012A (en) | Brain method for establishing network model, brain image classification method, device and electronic equipment | |
Altundogan et al. | Multiple object tracking with dynamic fuzzy cognitive maps using deep learning |
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: 20191115 |
|
RJ01 | Rejection of invention patent application after publication |