CN110490267A - A kind of bill method for sorting based on deep learning - Google Patents
A kind of bill method for sorting based on deep learning Download PDFInfo
- Publication number
- CN110490267A CN110490267A CN201910787567.0A CN201910787567A CN110490267A CN 110490267 A CN110490267 A CN 110490267A CN 201910787567 A CN201910787567 A CN 201910787567A CN 110490267 A CN110490267 A CN 110490267A
- Authority
- CN
- China
- Prior art keywords
- bill
- training
- step1
- step3
- type
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Abstract
The invention discloses a kind of bill method for sorting based on deep learning, the following steps are included: S1. collects business datum, it carries out filing mark: rough sort is carried out plus business picture of the Unsupervised clustering algorithm K-Means to collection using the VGG16 network of ImageNet pre-training, then by manually carrying out accurately sorting out the bill data for obtaining and having marked；S2. it divides data set: TotalDatasct is divided into training set TrainDataset and verifying collection ValidationDataset, based on each specific bill type number of samples number selectively divide trained and validation data set, the case where training data and verify data division level consider imbalanced training sets；S3. it trains STNEfficientNet: verifying network structure STNEfficientNet using TrainDataset training and using ValidationDataset；S4. the model that deployment training is completed carries out automatic sorting to the new bill that client uploads using trained network STNEfficientNet.Training speed of the present invention is fast, and precision is high, can effectively promote the efficiency of manual sort's bill.
Description
Technical field
The present invention relates to pattern-recognition technical field of image processing more particularly to a kind of bill sortings based on deep learning
Method.
Background technique
The classification (sorting) of invoice has critically important status in the processes such as financial accounting.Invoice only carries out correct
Classification, the processing of subsequent accounting process could not be influenced.Traditional invoice classification way is usually to be sent out by manually
The classification and filing of ticket are handled, inefficiency and error-prone.
Summary of the invention
In view of the above-mentioned problems, the invention proposes a kind of bill method for sorting based on deep learning, based on STN network and
General picture classification technology EfficientNet [2] is applied to invoice classification field, and utilized by EfficientNet network
Spatial Transformer Networks [1] (STN) network is improved, to promote the bills of poor quality such as distortion, inclination
Nicety of grading, for the present invention in bill classification field, training speed is fast, and precision is high, can effectively promote the efficiency of manual sort's bill.
To more fully understand relevant issues, now make explanations to following proper noun.
VGG: a kind of typical convolutional neural networks, details see reference document [3]；
Total classification of bill sorting: refer to that the bill of how many type altogether is sorted, this type sum depends on
Specific business demand.For example, if business only needs to distinguish identity card and train ticket, in the bill sorting task, bill point
The total classification number picked is 2；If business needs to distinguish identity card, train ticket, quota invoice, in the bill sorting task, bill
Total classification number of sorting is 3；
A kind of K-Means cluster: simple common clustering algorithm [4]；
Remember that network structure of the invention is STNEfficientNet, i.e., STN is embedded in EfficientNet network header shape
At network structure.
Bibliography:
[1] https: //arxiv.org/pdf/1506.02025.pdf；
[2] https: //arxiv.org/pdf/1905.11946.pdf；
[3] https: //arxiv.org/pdf/1409.1556.pdf；
[4] https: //en.wikipedia.org/wiki/K-means_clustering；
The automatic sorting problem of bill is substantially the classification problem of a bill picture, and it can be considered to use quilt
It proves to solve the problems, such as this with the EfficientNet sorter network that inference speed is fast, precision is high, model is small.
A kind of picture classification algorithm of the EfficientNet network as newest proposition, experiments have shown that its with inference speed fast, model
Feature small, with high accuracy.Usually there is the interference such as various degrees of deformation, distortion, inclination in the bill of practical business, therefore conventional
Disaggregated model practical business bill sorting in fail to obtain preferable result.STN network is proved to have rotational automatic
Correction, perspective stretch the function restored automatically, and STN and EfficientNet can effectively promote bill and be deformed in bill, turned round
Sorting accuracy rate under the interference such as bent, inclination；Method of the present invention using cascade STN module and EfficientNet network, solution
The certainly automatic sorting problem of bill.
The present invention through the following technical solutions to achieve the above objectives:
A kind of bill method for sorting based on deep learning, comprising the following steps:
S1. business datum is collected, filing mark is carried out: using the VGG16 network of ImageNet pre-training plus unsupervised
Clustering algorithm K-Means carries out rough sort to the business picture of collection, then by manually carrying out accurately sorting out to obtain marking
Bill data；
S2. it divides data set: TotalDataset being divided into training set TrainDataset and verifying collects
ValidationDataset, based on each specific bill type number of samples number selectively divide trained and verify data
Collection, the case where training data and verify data division level consider imbalanced training sets；
S3. it trains STNEfficientNet: being tested using TrainDataset training and using ValidationDataset
Demonstrate,prove network structure STNEfficientNet；
S4. the model that deployment training is completed uploads client new using trained network STNEfficientNet
Bill carry out automatic sorting.
Further scheme is that specific step is as follows by S1:
Step1-1: the n bills that client uploads are collected and are achieved, image_000001.png, image_ are labeled as
0000002.png ..., image_n.png；
Step1-2: using the VGG16 network of ImageNet pre-training, 4096 dimensional vectors of its 2nd layer of output reciprocal are obtained
(image_i.png, vec_i_4096), wherein i indicates i-th invoice；
Step1-3: the 4096 dimensional vector Invoice_Vec=(image_ of every bill are obtained using step Step1-2
I.png, vec_i_4096), i ∈ [1, n] is clustered Invoice_Vec for N=200 class using K-Means algorithm；
Step1-4: this n picture is clustered as the picture after N=200 class, artificial nucleus to and filed corresponding
Bill type, concrete operations process are as follows:
Step1-4-1: taking any type N_j in N class, therefrom take 20 pictures at random, if picture in the cluster classification
Sum is less than 20, then takes whole；
Step1-4-2: then actual invoice type belonging to each picture in 20 picture of artificial judgment takes invoice class
The most type of number of pictures is the preset kind of N_j in type；
Step1-4-3: judging whether the preset kind of every one kind N_j in N class cluster result has obtained, if then turning to walk
Rapid Step1-4-4, otherwise turns Step1-4-1；
Step1-4-4: merging the identical N_j1 and N_j2 of preset kind is same type M_j1_j2, for example, N_1 type and
The preset kind of N_20 type is identical, then is merged into same preset kind, as M_1_20, if without type and N_j phase
Together, then its amalgamation result is separately denoted as M_j；
Step1-4-5: all data completed that merge of note are M=[M_1, M_2 ..., M_j ..., M_x]；
Step1-4-6: taking each data subset M_j in M, successively carries out bill classification completion to wherein every invoice and returns
Shelves；
Step1-5: the picture after all filings is denoted as TotalDataset, including two parts: part 1 is institute
There are n bill pictures of collection, part 2 is the corresponding bill classification of every bill picture.
Further scheme is that specific step is as follows by S2:
Step2-1: counting the sample number of all kinds of bills, and training set TrainDataset is arranged as sky, and setting verifying collects
ValidationDataset is sky；
Does Step2-2: the one kind in bill type for taking N category to be poured in judge that its sample number is greater than 500 if then
Step2-3 is gone to step, if it is not, then going to step Step2-4；
Step2-3: randomly choosing 100 from the bill type, is directly placed into verifying collection ValidationDataset,
The bill type residue sample is put into training set TrainDataset；
Step2-4: randomly choosing 100 from the bill type, replicates this 100 and is put into verifying collection
ValidationDataset, the bill type whole sample are put into training set TrainDataset；
Step2-5: whether the sample standard deviation of all bill types has been divided to training set TrainDataset and verifying collection
ValidationDataset, if then turning Step3, if otherwise turning Step2-2.
Further scheme is that specific step is as follows by S3:
Step3-1: set algorithm parameter specifically includes initial learning rate learning rate=1e-4, batch of training
Secondary size batch-size=16, network reference services device optimizer are band momentum momentum=0.9 and weight decays
The stochastic gradient descent method SGD of weight_decay=1e-4, algorithm total cycle of training are EPOACH=50, initial epoach
=0；
Step3-2: establishing network model model=STNEfficientNet (), constructs the specific steps of network model such as
Under:
Step3-1: a spatial alternation network STN is established；
Step3-2: EfficientNet network is established；
Step3-3: by STN cascade EfficientNet network, STNEfficientNet network is formed；
Step3-3: judging whether epoach is less than EPOACH, if then going to step Step3-4, if otherwise turning S4；
Step3-4: taking batch-size invoice from TrainDataset, be trained to model, and utilizes SGD
Carry out model algorithm update；
Step3-5: judge whether that training is completed in all invoices in TrainDataset, if so, going to step
Step3-6, if otherwise going to step Step3-4；
Step3-6: verifying the precision of model model with ValidationDataset, and save trained model, turns
Step3-3。
The beneficial effects of the present invention are:
Training speed of the present invention is fast, and precision is high, can effectively promote the efficiency of manual sort's bill.
Detailed description of the invention
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
In required practical attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description is only the one of the present embodiment
A little embodiments for those of ordinary skill in the art without creative efforts, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1: whole implementation flow chart of the invention；
Fig. 2: STNEfficientNet of the invention constructs process.
Fig. 3: business datum of the invention collects process.
Fig. 4: data set of the invention divides logic.
Fig. 5: STNEfficientNet training process of the invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, technical solution of the present invention will be carried out below
Detailed description.Obviously, the described embodiment is only a part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art without making creative work it is obtained it is all its
Its embodiment belongs to the range that the present invention is protected.
In any embodiment, as shown in Figure 1, a kind of bill method for sorting based on deep learning of the invention, including
Following steps:
S1. business datum is collected, filing mark is carried out: using the VGG16 network of ImageNet pre-training plus unsupervised
Clustering algorithm K-Means carries out rough sort to the business picture of collection, then by manually carrying out accurately sorting out to obtain marking
Bill data；
S2. it divides data set: TotalDataset being divided into training set TrainDataset and verifying collects
ValidationDataset, based on each specific bill type number of samples number selectively divide trained and verify data
Collection, the case where training data and verify data division level consider imbalanced training sets；
S3. it trains STNEfficientNet: being tested using TrainDataset training and using ValidationDataset
Demonstrate,prove network structure STNEfficientNet；
S4. the model that deployment training is completed uploads client new using trained network STNEfficientNet
Bill carry out automatic sorting.
In a specific embodiment, the process according to Figure of description 3 collects business datum, specifically includes following several
A step:
Step1-1: the n bills that client uploads are collected and are achieved, image_000001.png, image_ are labeled as
0000002.png ..., image_n.png；
Step1-2: using the VGG16 network of ImageNet pre-training, 4096 dimensional vectors of its 2nd layer of output reciprocal are obtained
(image_i.png, vec_i_4096), wherein i indicates i-th invoice；
Step1-3: the 4096 dimensional vector Invoice_Vec=(image_ of every bill are obtained using step Step1-2
I.png, vec_i_4096), i ∈ [1, n] is clustered Invoice_Vec for N=200 class using K-Means algorithm；
Step1-4: this n picture is clustered as the picture after N=200 class, artificial nucleus to and filed corresponding
Bill type, concrete operations process are as follows:
Step1-4-1: taking any type N_j in N class, therefrom take 20 pictures at random, if picture in the cluster classification
Sum is less than 20, then takes whole；
Step1-4-2: then actual invoice type belonging to each picture in 20 picture of artificial judgment takes invoice class
The most type of number of pictures is the preset kind of N_j in type；
Step1-4-3: judging whether the preset kind of every one kind N_j in N class cluster result has obtained, if then turning to walk
Rapid Step1-4-4, otherwise turns Step1-4-1；
Step1-4-4: merging the identical N_j1 and N_j2 of preset kind is same type M_j1_j2, for example, N_1 type and
The preset kind of N_20 type is identical, then is merged into same preset kind, as M_1_20, if without type and N_j phase
Together, then its amalgamation result is separately denoted as M_j；
Step1-4-5: all data completed that merge of note are M=[M_1, M_2 ..., M_j ..., M_x]；
Step1-4-6: taking each data subset M_j in M, successively carries out bill classification completion to wherein every invoice and returns
Shelves；
Step1-5: the picture after all filings is denoted as TotalDataset, including two parts: part 1 is institute
There are n bill pictures of collection, part 2 is the corresponding bill classification of every bill picture.
It in a specific embodiment, is training set by logical partitioning described in TotalDataset by specification attached drawing 4
TrainDataset and verifying collection ValidationDataset, specifically includes the following steps:
Step2-1: counting the sample number of all kinds of bills, and training set TrainDataset is arranged as sky, and setting verifying collects
ValidationDataset is sky；
Does Step2-2: the one kind in bill type for taking N category to be poured in judge that its sample number is greater than 500 if then
Step2-3 is gone to step, if it is not, then going to step Step2-4；
Step2-3: randomly choosing 100 from the bill type, is directly placed into verifying collection ValidationDataset,
The bill type residue sample is put into training set TrainDataset；
Step2-4: randomly choosing 100 from the bill type, replicates this 100 and is put into verifying collection
ValidationDataset, the bill type whole sample are put into training set TrainDataset；
Step2-5: whether the sample standard deviation of all bill types has been divided to training set TrainDataset and verifying collection
ValidationDataset, if then turning Step3, if otherwise turning Step2-2.
In a specific embodiment, it is said using TrainDataset training and using ValidationDataset verifying
Network structure STNEfficientNet constructed by bright book attached drawing 2, wherein training method is using mode described in Figure of description 5
Training, specific training step are as follows:
Step3-1: set algorithm parameter specifically includes initial learning rate learning rate=1e-4, batch of training
Secondary size batch-size=16, network reference services device optimizer are band momentum momentum=0.9 and weight decays
The stochastic gradient descent method SGD of weight_decay=1e-4, algorithm total cycle of training are EPOACH=50, initial epoach
=0；
Step3-2: establishing network model model=STNEfficientNet (), constructs the specific steps of network model such as
Under:
Step3-1: a spatial alternation network STN is established；
Step3-2: EfficientNet network is established；
Step3-3: by STN cascade EfficientNet network, STNEfficientNet network is formed；
Step3-3: judging whether epoach is less than EPOACH, if then going to step Step3-4, if otherwise turning S4；
Step3-4: taking batch-size invoice from TrainDataset, be trained to model, and utilizes SGD
Carry out model algorithm update；
Step3-5: judge whether that training is completed in all invoices in TrainDataset, if so, going to step
Step3-6, if otherwise going to step Step3-4；
Step3-6: verifying the precision of model model with ValidationDataset, and save trained model, turns
Step3-3。
Frame in figure with vertical line indicates that this step is completed by a submodule.
It was verified that the present invention is able to achieve efficiently, accurately, bill is sorted.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Specific technical features described in the above specific embodiments, in not lance
In the case where shield, can be combined in any appropriate way, in order to avoid unnecessary repetition, the present invention to it is various can
No further explanation will be given for the combination of energy.Various embodiments of the present invention can be combined randomly, only
Want it without prejudice to thought of the invention, it should also be regarded as the disclosure of the present invention.
Claims (4)
1. a kind of bill method for sorting based on deep learning, which comprises the following steps:
S1. business datum is collected, filing mark is carried out: using the VGG16 network of ImageNet pre-training plus Unsupervised clustering
Algorithm K-Means carries out rough sort to the business picture of collection, then obtains the ticket marked by manually carrying out accurate classification
According to data；
S2. it divides data set: TotalDataset being divided into training set TrainDataset and verifying collects
ValidationDataset, based on each specific bill type number of samples number selectively divide trained and verify data
Collection, the case where training data and verify data division level consider imbalanced training sets；
S3. it trains STNEfficientNet: verifying net using TrainDataset training and using ValidationDataset
Network structure STNEfficientNet；
S4. the model that deployment training is completed, using trained network STNEfficientNet, the new ticket that client is uploaded
According to progress automatic sorting.
2. a kind of bill method for sorting based on deep learning as described in claim 1, which is characterized in that S1 specific steps are such as
Under:
Step1-1: the n bills that client uploads are collected and are achieved, image_000001.png, image_ are labeled as
0000002.png ..., image_n.png；
Step1-2: using the VGG16 network of ImageNet pre-training, 4096 dimensional vectors of its 2nd layer of output reciprocal are obtained
(image_i.png, vec_i_4096), wherein i indicates i-th invoice；
Step1-3: using step Step1-2 obtain every bill 4096 dimensional vector Invoice_Vec=(image_i.png,
Vec_i_4096), [1, n] i ∈ is clustered Invoice_Vec for N=200 class using K-Means algorithm；
Step1-4: this n picture is clustered as the picture after N=200 class, artificial nucleus to and filed corresponding bill
Type, concrete operations process are as follows:
Step1-4-1: taking any type N_j in N class, therefrom take 20 pictures at random, if picture sum in the cluster classification
Less than 20, then whole is taken；
Step1-4-2: then actual invoice type belonging to each picture in 20 picture of artificial judgment takes in invoice type
The most type of number of pictures is the preset kind of N_j；
Step1-4-3: judging whether the preset kind of every one kind N_j in N class cluster result has obtained, if then going to step
Otherwise Step1-4-4 turns Step1-4-1；
Step1-4-4: merging the identical N_j1 and N_j2 of preset kind is same type M_j1_j2, such as N_1 type and N_20
The preset kind of type is identical, then is merged into same preset kind, as M_1_20, if identical as N_j without type,
Its amalgamation result is separately denoted as M_j；
Step1-4-5: all data completed that merge of note are M=[M_1, M_2 ..., M_j ..., M_x]；
Step1-4-6: taking each data subset M_j in M, successively carries out bill classification to wherein every invoice and completes filing；
Step1-5: the picture after all filings is denoted as TotalDataset, including two parts: part 1 is all receipts
The n of collection bill pictures, part 2 is the corresponding bill classification of every bill picture.
3. a kind of bill method for sorting based on deep learning as described in claim 1, which is characterized in that S2 specific steps are such as
Under:
Step2-1: counting the sample number of all kinds of bills, and training set TrainDataset is arranged as sky, and setting verifying collects
ValidationDataset is sky；
Does Step2-2: the one kind in bill type for taking N category to be poured in judge that its sample number is greater than 500 if then turning to walk
Rapid Step2-3, if it is not, then going to step Step2-4；
Step2-3: randomly choosing 100 from the bill type, is directly placed into verifying collection ValidationDataset, the ticket
Training set TrainDataset is put into according to type residue sample；
Step2-4: randomly choosing 100 from the bill type, replicates this 100 and is put into verifying collection
ValidationDataset, the bill type whole sample are put into training set TrainDataset；
Step2-5: whether the sample standard deviation of all bill types has been divided to training set TrainDataset and verifying collection
ValidationDataset, if then turning Step3, if otherwise turning Step2-2.
4. a kind of bill method for sorting based on deep learning as described in claim 1, which is characterized in that S3 specific steps are such as
Under:
Step3-1: set algorithm parameter specifically includes initial learning rate learningrate=1e-4, and trained batch is big
Small batch-size=16, network reference services device optimizer are band momentum momentum=0.9 and weight decays
The stochastic gradient descent method SGD of weight_decay=1e-4, algorithm total cycle of training are EPOACH=50, initial epoach
=0；
Step3-2: establishing network model model=STNEfficientNet (), and constructing network model, specific step is as follows:
Step3-1: a spatial alternation network STN is established；
Step3-2: EfficientNet network is established；
Step3-3: by STN cascade EfficientNet network, STNEfficientNet network is formed；
Step3-3: judging whether epoach is less than EPOACH, if then going to step Step3-4, if otherwise turning S4；
Step3-4: taking batch-size invoice from TrainDataset, be trained to model, and is carried out using SGD
Model algorithm updates；
Step3-5: judging whether that training is completed in all invoices in TrainDataset, if so, Step3-6 is gone to step,
If otherwise going to step Step3-4；
Step3-6: verifying the precision of model model with ValidationDataset, and save trained model, turns
Step3-3。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910787567.0A CN110490267B (en) | 2019-08-23 | 2019-08-23 | Bill sorting method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910787567.0A CN110490267B (en) | 2019-08-23 | 2019-08-23 | Bill sorting method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110490267A true CN110490267A (en) | 2019-11-22 |
| CN110490267B CN110490267B (en) | 2022-03-01 |
Family
ID=68553919
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910787567.0A Active CN110490267B (en) | 2019-08-23 | 2019-08-23 | Bill sorting method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110490267B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111523579A (en) * | 2020-04-14 | 2020-08-11 | 燕山大学 | Vehicle type recognition method and system based on improved deep learning |
| CN111666872A (en) * | 2020-06-04 | 2020-09-15 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN112115943A (en) * | 2020-09-16 | 2020-12-22 | 四川长虹电器股份有限公司 | Bill rotation angle detection method based on deep learning |
| CN116778513A (en) * | 2023-08-24 | 2023-09-19 | 国网山西省电力公司太原供电公司 | Intelligent archiving control method for bills in power industry |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7165041B1 (en) * | 1999-05-27 | 2007-01-16 | Accenture, Llp | Web-based architecture sales tool |
| US20120002008A1 (en) * | 2010-07-04 | 2012-01-05 | David Valin | Apparatus for secure recording and transformation of images to light for identification, and audio visual projection to spatial point targeted area |
| CN105045780A (en) * | 2015-07-15 | 2015-11-11 | 广州敦和信息技术有限公司 | Method and device for identifying semantic information of invoice brief notes |
| CN106096667A (en) * | 2016-07-26 | 2016-11-09 | 江苏鸿信系统集成有限公司 | Bill images sorting technique based on SVM |
| US20170124497A1 (en) * | 2015-10-28 | 2017-05-04 | Fractal Industries, Inc. | System for automated capture and analysis of business information for reliable business venture outcome prediction |
| CN107292823A (en) * | 2017-08-20 | 2017-10-24 | 平安科技（深圳）有限公司 | Electronic installation, the method for invoice classification and computer-readable recording medium |
| US20180039914A1 (en) * | 2016-08-04 | 2018-02-08 | Loom Systems LTD. | Machine learning techniques for providing enriched root causes based on machine-generated data |
| CN108446621A (en) * | 2018-03-14 | 2018-08-24 | 平安科技（深圳）有限公司 | Bank slip recognition method, server and computer readable storage medium |
| CN110070665A (en) * | 2019-04-24 | 2019-07-30 | 武汉华创欣网科技有限公司 | A kind of insurance bill classification method based on deep learning |
-
2019
- 2019-08-23 CN CN201910787567.0A patent/CN110490267B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7165041B1 (en) * | 1999-05-27 | 2007-01-16 | Accenture, Llp | Web-based architecture sales tool |
| US20120002008A1 (en) * | 2010-07-04 | 2012-01-05 | David Valin | Apparatus for secure recording and transformation of images to light for identification, and audio visual projection to spatial point targeted area |
| CN105045780A (en) * | 2015-07-15 | 2015-11-11 | 广州敦和信息技术有限公司 | Method and device for identifying semantic information of invoice brief notes |
| US20170124497A1 (en) * | 2015-10-28 | 2017-05-04 | Fractal Industries, Inc. | System for automated capture and analysis of business information for reliable business venture outcome prediction |
| CN106096667A (en) * | 2016-07-26 | 2016-11-09 | 江苏鸿信系统集成有限公司 | Bill images sorting technique based on SVM |
| US20180039914A1 (en) * | 2016-08-04 | 2018-02-08 | Loom Systems LTD. | Machine learning techniques for providing enriched root causes based on machine-generated data |
| CN107292823A (en) * | 2017-08-20 | 2017-10-24 | 平安科技（深圳）有限公司 | Electronic installation, the method for invoice classification and computer-readable recording medium |
| CN108446621A (en) * | 2018-03-14 | 2018-08-24 | 平安科技（深圳）有限公司 | Bank slip recognition method, server and computer readable storage medium |
| CN110070665A (en) * | 2019-04-24 | 2019-07-30 | 武汉华创欣网科技有限公司 | A kind of insurance bill classification method based on deep learning |
Non-Patent Citations (5)
| Title |
|---|
| KAREN SIMONYAN 等: "VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION", 《ICLR0 2015》 * |
| MAX JADERBERG 等: "Spatial Transformer Networks", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
| MINGXING TAN 等: "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks", 《 PROCEEDINGS OF THE 36TH INTERNATIONAL CONFERENCE ON MACHINE LEARNING》 * |
| 陈增照 等: "基于多特征融合的票据分类技术及应用", 《计算机工程与应用》 * |
| 高欣 等: "基于模型自适应选择融合的智能电表故障多分类方法", 《电网技术》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111523579A (en) * | 2020-04-14 | 2020-08-11 | 燕山大学 | Vehicle type recognition method and system based on improved deep learning |
| CN111523579B (en) * | 2020-04-14 | 2022-05-03 | 燕山大学 | Vehicle type recognition method and system based on improved deep learning |
| CN111666872A (en) * | 2020-06-04 | 2020-09-15 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN111666872B (en) * | 2020-06-04 | 2022-08-05 | 电子科技大学 | Efficient behavior identification method under data imbalance |
| CN112115943A (en) * | 2020-09-16 | 2020-12-22 | 四川长虹电器股份有限公司 | Bill rotation angle detection method based on deep learning |
| CN116778513A (en) * | 2023-08-24 | 2023-09-19 | 国网山西省电力公司太原供电公司 | Intelligent archiving control method for bills in power industry |
| CN116778513B (en) * | 2023-08-24 | 2023-10-27 | 国网山西省电力公司太原供电公司 | Intelligent archiving control method for bills in power industry |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110490267B (en) | 2022-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110490267A (en) | A kind of bill method for sorting based on deep learning | |
| CN102521656B (en) | Integrated transfer learning method for classification of unbalance samples | |
| CN104331498B (en) | A kind of method that web page contents to internet user access are classified automatically | |
| CN104239858B (en) | A kind of method and apparatus of face characteristic checking | |
| CN108921166A (en) | Medical bill class text detection recognition method and system based on deep neural network | |
| CN105844287B (en) | A kind of the domain adaptive approach and system of classification of remote-sensing images | |
| CN107067044A (en) | A kind of finance reimbursement unanimous vote is according to intelligent checks system | |
| CN109299741B (en) | Network attack type identification method based on multi-layer detection | |
| CN102156885B (en) | Image classification method based on cascaded codebook generation | |
| CN108345904A (en) | A kind of Ensemble Learning Algorithms of the unbalanced data based on the sampling of random susceptibility | |
| CN103679160B (en) | Human-face identifying method and device | |
| CN109190594A (en) | Optical Character Recognition system and information extracting method | |
| CN109491914A (en) | Defect report prediction technique is influenced based on uneven learning strategy height | |
| CN107644057A (en) | A kind of absolute uneven file classification method based on transfer learning | |
| CN106446602A (en) | Prediction method and system for RNA binding sites in protein molecules | |
| CN206975668U (en) | A kind of finance reimbursement unanimous vote is according to intelligent checks system | |
| CN109784368A (en) | A kind of determination method and apparatus of application program classification | |
| CN114663002A (en) | Method and equipment for automatically matching performance assessment indexes | |
| CN109614866A (en) | Method for detecting human face based on cascade deep convolutional neural networks | |
| CN109598307A (en) | Data screening method, apparatus, server and storage medium | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN110110845A (en) | A kind of learning method based on parallel multi-level width neural network | |
| CN103971136A (en) | Large-scale data-oriented parallel structured support vector machine classification method | |
| CN107392230A (en) | A kind of semi-supervision image classification method for possessing maximization knowledge utilization ability | |
| CN108416369A (en) | Based on Stacking and the random down-sampled sorting technique of overturning, system, medium and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |