CN108133212A - A kind of quota invoice amount identifying system based on deep learning - Google Patents
A kind of quota invoice amount identifying system based on deep learning Download PDFInfo
- Publication number
- CN108133212A CN108133212A CN201810011763.4A CN201810011763A CN108133212A CN 108133212 A CN108133212 A CN 108133212A CN 201810011763 A CN201810011763 A CN 201810011763A CN 108133212 A CN108133212 A CN 108133212A
- Authority
- CN
- China
- Prior art keywords
- module
- deep learning
- picture
- image
- result
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/243—Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
-
- 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
- G06F18/2163—Partitioning the feature space
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Character Input (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of quota invoice amount identifying systems based on deep learning, and including image collection module, image rotation module, picture recognition module and result memory module, described image acquisition module is used to obtain picture file；Described image rotary module is used to be corrected the picture file；Described image identification module obtains the specific location for the picture file of being identified using deep learning model, and carries out image identification；The result memory module is used to store final recognition result.The present invention can improve OCR discriminations when image is contaminated.
Description
Technical field
The present invention relates to image identification technical fields, know more particularly to a kind of quota invoice amount based on deep learning
Other system.
Background technology
The concept of OCR (Optical Character Recognition, optical character identification) earlier than nineteen twenty generation just by
It proposes, is always research direction important in area of pattern recognition.
In recent years, with the quick update iteration of mobile equipment and the fast development of mobile Internet so that OCR has
Scene is more widely applied, from the character recognition of previous scanning file, is applied to picture character in natural scene till now
Identification, as identified the word in identity card, bank card, doorplate, bill and disparate networks picture.
Traditional OCR technique is as follows：
String localization first is then corrected into line tilt text, after being partitioned into individual character later, and individual character is identified, finally
Semantic error correction is carried out based on statistical model (such as hidden Markov chain, HMM).Three phases can be divided by processing mode：Pre- place
Reason stage, cognitive phase and post-processing stages.Wherein key is pretreatment stage, and the quality of pretreatment stage directly determines
Final recognition effect, therefore lower pretreatment stage detailed herein.
Three steps are contained in pretreatment stage：
(1) character area in picture is positioned, and the method that text detection is based primarily upon connected domain analysis, main thought are
The mode clustered using text color, brightness, marginal information is more flowed come quick separating character area and non-legible region
Two capable algorithms are respectively：Maximum extreme value stability region (MSER) algorithm and stroke width transformation (SWT) algorithm, and in nature
Because being interfered by intensity of illumination, picture shooting quality and class character background in scene so that comprising very more in testing result
Non-legible region, and distinguish the main two methods of real character area from candidate region at present, judged with rule or light weight
The neural network model of grade distinguishes；
(2) text filed image flame detection, is based primarily upon rotation transformation and affine transformation；
(3) ranks segmentation extracts individual character, this step using word between ranks there are the feature in gap, pass through binaryzation
And ranks cut-point is found out in the projected, when the discrimination in word and background is preferable, effect is fine, and in the picture shot
Illumination, the influence of pickup quality, and when character background is difficult to differentiate between, often result in the situation of erroneous segmentation.
It can be seen that the step of tradition OCR identification frameworks is more, therefore it is easy to cause the final identification of error accumulation influence
As a result.
Invention content
The technical problems to be solved by the invention are to provide a kind of quota invoice amount identifying system based on deep learning,
OCR discriminations when image is contaminated can be improved.
The technical solution adopted by the present invention to solve the technical problems is：A kind of quota invoice based on deep learning is provided
Amount of money identifying system, including image collection module, image rotation module, picture recognition module and result memory module, the figure
As acquisition module is used to obtain picture file；Described image rotary module is used to be corrected the picture file；The figure
As identification module using deep learning model obtains the specific location for the picture file of being identified, and carry out image identification；It is described
As a result memory module is used to store final recognition result.
Described image rotary module adjusts direction in a manner that opencv rotation adjusting angles are combined using tesseract
The picture file is corrected.
Described image rotary module extracts straight line by Hough transform, is calculated respectively since straight line apical pixel point more
The corresponding origin of a angle is to the distance of straight line；The pixel above-mentioned steps of whole image are traversed, find out the most distance of repetition,
The linear equation of the line correspondences is obtained, finally obtains rotation angle.
Described image rotary module obtains the rotation angle of pictograph using tesseract.
Described image identification module includes sample process unit, image training unit and test cell；The sample process
Unit is used for the samples pictures that finishing collecting arrives, and picture classification is marked, and obtains the corresponding xml document of picture, described
Xml document includes the classification information and location information of picture；Described image training unit is using 24 convolutional layers and 2 full links
Layer, wherein, convolutional layer is for extracting feature, and full linking layer is used for prediction result, and the output of last layer is k dimension, wherein, k
=S*S (B*5+C), k include class prediction and the prediction of bbox coordinates, and S is the grid number divided, and B is responsible for mesh for each grid
Number is marked, C is classification number；The test cell utilizes classification information and bounding the box prediction of each grid forecasting
Authentication information is multiplied, and just obtains the best score of each boundingbox, it is low as a result, to retaining that setting threshold value filters score
Result carry out NMS processing obtain final testing result.
Advantageous effect
As a result of above-mentioned technical solution, compared with prior art, the present invention having the following advantages that and actively imitating
Fruit：The present invention reduces step compared to traditional OCR identification frameworks, reduces the shadow to final recognition result because of error accumulation
It rings.The present invention realizes the combination of deep learning and the identification of OCR images, can greatly improve OCR discriminations when image is contaminated,
It is easy to operate；This system can be applied to accounting field, can improve the working efficiency of accountant, can be them from cumbersome
It is freed in work.
Description of the drawings
Fig. 1 is the system block diagram of the present invention；
Fig. 2 is the internal structure chart of the present invention；
Fig. 3 A-3B are using the recognition result figure after embodiment of the present invention.
Specific embodiment
With reference to specific embodiment, the present invention is further explained.It should be understood that these embodiments are merely to illustrate the present invention
Rather than it limits the scope of the invention.In addition, it should also be understood that, after reading the content taught by the present invention, people in the art
Member can make various changes or modifications the present invention, and such equivalent forms equally fall within the application the appended claims and limited
Range.
Embodiments of the present invention are related to a kind of quota invoice amount identifying system based on deep learning, as shown in Figure 1,
Including image collection module, image rotation module, picture recognition module and result memory module, described image acquisition module is used for
Obtain picture file；Described image rotary module is used to be corrected the picture file；Described image identification module utilizes
Deep learning model obtains the specific location for the picture file of being identified, and carries out image identification；The result memory module is used
In the recognition result that storage is final.
As shown in Fig. 2, present embodiment on the basis of the quota invoice that customer scans are entered, identifies the amount of money and hair above
Ticket code, number.Since the picture that client uploads is likely to be inclination or reversing, for the ease of identifying below, increase
Spin step, the picture after rotating through obtain field above by ocr.
Image rotation is directed to have some pictures to be inclination or upside down in the picture of scanning input upload.This
Image rotation module in embodiment adjusts direction using tesseract and rotates the side for adjusting low-angle and being combined with opencv
Formula is corrected picture.
1.opencv rotations are adjusted
What present embodiment was mainly used is the method for opencv lines detections, then the angle of inclination of straight line is obtained, at this
In the method for lines detection be Hough transform.
Any point O (x, y) in rectangular coordinate system, any one straight line by O can all meet Y=kX+b (except vertical
The straight line of straight X-axis).Due to this special circumstances, so coordinate system is converted to polar coordinate system to meet this case.
In polar coordinate system, arbitrary straight line can be represented with ρ=xCos θ+ySin θ.
Assuming that having straight line in the image of a width 10*10, angle is calculated respectively since graph line apical pixel point
Corresponding origin is spent when being 180 °, 135 °, 90 °, 45 °, 0 ° to the distance of straight line.It is repeated in the pixel of traversal whole image firm
The step of, finds out the most distance of repetition, has just obtained corresponding linear equation, and obtain angle.
When a width figure finds out a plurality of straight line, rotation angle of the highest angle of angular frequency as image is taken.
2.Tesseract rotates
Tesseract be Ray Smithf in 1985 between nineteen ninety-five at one of Hewlett-Packard's Bristol development in laboratory
OCR engine once came out at the top in the test of 1995UNLV accuracy.But exploitation is stopped after 1996 substantially.2006,
Google invites Smith to join, and restarts the project.The licensing of project is Apache 2.0 at present.The project is supported at present
The Mainstream Platforms such as Windows, Linux and Mac OS.But as an engine, it only provides command-line tool.
Tessersact can identify most text languages (including Chinese), it can be obtained in word on picture
Hold the rotation angle (270 °, 180 °, 90 °, 0 °) with picture character, since its accuracy of identification is not high, present embodiment only uses
Tesseract obtains the rotation angle of pictograph.Tesseract only receives gray-scale map, so the cromogram of input needs to convert
For gray-scale map.
Picture recognition module goes to identify using the method for deep learning in present embodiment, here using deep learning target
The method yolo (Youonly lookonce) of detection.
The thought of YOLO：Directly the classification belonging to the position of bounding box and bounding box is returned in output layer
(input of the whole figure as network, the problem of Object Detection is converted to a Regression problem).
1. sample process：
The samples pictures being collected into are put in order, using this software tags picture classification of labelme, are obtained corresponding
Xml document, file contain the information of classification and position in picture.
2. image is trained：
Picture is normalized into 448*448 first, the center that picture segmentation obtains 7*7 grids (cell) a object is fallen at this
This grid is just responsible for predicting this object in a grid.
CNN extracts feature and prediction：Convolution is responsible for putting forward feature；It is responsible for prediction in full link part.The output of last layer is k
Dimension.Wherein
K=S*S (B*5+C) (1)
K includes class prediction and the prediction of bbox coordinates.S is the grid number divided, and B is responsible for target for each grid
Number, C are classification number.Wherein 5 center point coordinates comprising prediction, wide high and class prediction.The prediction of bbox coordinates is expressed as：
Wherein if there is ground true box (object of handmarking) are fallen in a grid cell, first item takes
1, otherwise take 0.Section 2 is the IOU values between the bounding box of prediction and the groundtruthbox of reality.
Network structure has used for reference GoogLeNet.24 convolutional layers, 2 full linking layers.(with 1 × 1reduction
Layers is immediately following the inceptionmodules of 3 × 3convolutional layers substitutions Goolenet)
The design object of loss function is exactly to allow coordinate (x, y, w, h), confidence, classification this three
A aspect reaches good balance.
In predicting different size of bbox, compared to big bbox predictions partially a bit, small box predictions can not a little be born partially
By.And it is the same to similarly deviating loss in total weighting loss.In order to mitigate this problem, present embodiment is by box
Width and height make even root replace script width and height.
The multiple bounding box of one grid forecasting, in training it is desirable that each classification (groundtruebox)
Only it is responsible for specially (mono- bbox of an object) there are one bounding box.Specific practice is and ground truebox
(object) the bounding box of IOU maximums are responsible for the prediction of the groundtruebox (object).This way is referred to as
The specialization (serve full time) of bounding boxpredictor.Each fallout predictor can be to specific (sizes, aspect
Ratio or classed of object) groundtruebox prediction become better and better.
3. test module：
When test, the classification information Pr (Class of each grid forecastingi| Object) and bounding box predictions
Confidence informationIt is multiplied, just obtains the best score of each bounding box.It obtains every
After the best score of a bbox, threshold value is set, filters the low boxes of score, NMS processing is carried out to the boxes of reservation, it must
To final testing result.Fig. 3 A-3B are recognition result figures after applying the present invention.
It is not difficult to find that the present invention reduces step compared to traditional OCR identification frameworks, reduce because error accumulation is to final
The influence of recognition result.The present invention realizes the combination of deep learning and the identification of OCR images, can greatly improve image and be contaminated
When OCR discriminations, it is easy to operate；This system can be applied to accounting field, can improve the working efficiency of accountant, can be
They free from cumbersome work.
Claims (5)
1. a kind of quota invoice amount identifying system based on deep learning, including image collection module, image rotation module, figure
As identification module and result memory module, which is characterized in that described image acquisition module is used to obtain picture file；Described image
Rotary module is used to be corrected the picture file；Described image identification module obtains being known using deep learning model
The specific location of other picture file, and carry out image identification；The result memory module is used to store final recognition result.
2. the quota invoice amount identifying system according to claim 1 based on deep learning, which is characterized in that the figure
It is rotated by the way of adjusting angle is combined to the picture file with opencv as rotary module adjusts direction using tesseract
It is corrected.
3. the quota invoice amount identifying system according to claim 2 based on deep learning, which is characterized in that the figure
Picture rotary module extracts straight line by Hough transform, calculates the corresponding original of multiple angles respectively since straight line apical pixel point
Point arrives the distance of straight line；The pixel above-mentioned steps of whole image are traversed, the most distance of repetition is found out, obtains the line correspondences
Linear equation, finally obtain rotation angle.
4. the quota invoice amount identifying system according to claim 2 based on deep learning, which is characterized in that the figure
As rotary module obtains using tesseract the rotation angle of pictograph.
5. the quota invoice amount identifying system according to claim 1 based on deep learning, which is characterized in that the figure
As identification module includes sample process unit, image training unit and test cell；The sample process unit is received for arranging
The samples pictures collected, and picture classification is marked, the corresponding xml document of picture is obtained, the xml document includes picture
Classification information and location information；Described image training unit uses 24 convolutional layers and 2 full linking layers, wherein, convolutional layer
For extracting feature, full linking layer is used for prediction result, and the output of last layer is k dimension, wherein, k=S*S (B*5+C), k
Include class prediction and the prediction of bbox coordinates, S is the grid number divided, and B is responsible for target number for each grid, and C is classification
Number；The test cell is multiplied using the classification information of each grid forecasting with the bounding box authentication informations predicted,
The best score of each bounding box is just obtained, threshold value is set to filter, and score is low as a result, the result to reservation carries out NMS
Processing obtains final testing result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810011763.4A CN108133212B (en) | 2018-01-05 | 2018-01-05 | Quota invoice amount recognition system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810011763.4A CN108133212B (en) | 2018-01-05 | 2018-01-05 | Quota invoice amount recognition system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108133212A true CN108133212A (en) | 2018-06-08 |
| CN108133212B CN108133212B (en) | 2021-06-29 |
Family
ID=62399437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810011763.4A Active CN108133212B (en) | 2018-01-05 | 2018-01-05 | Quota invoice amount recognition system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108133212B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109002768A (en) * | 2018-06-22 | 2018-12-14 | 深源恒际科技有限公司 | Medical bill class text extraction method based on the identification of neural network text detection |
| CN109816118A (en) * | 2019-01-25 | 2019-05-28 | 上海深杳智能科技有限公司 | A kind of method and terminal of the creation structured document based on deep learning model |
| CN109886257A (en) * | 2019-01-30 | 2019-06-14 | 四川长虹电器股份有限公司 | Using the method for deep learning correction invoice picture segmentation result in a kind of OCR system |
| CN109948617A (en) * | 2019-03-29 | 2019-06-28 | 南京邮电大学 | A kind of invoice image position method |
| CN109993160A (en) * | 2019-02-18 | 2019-07-09 | 北京联合大学 | A kind of image flame detection and text and location recognition method and system |
| CN110348346A (en) * | 2019-06-28 | 2019-10-18 | 苏宁云计算有限公司 | A kind of bill classification recognition methods and system |
| WO2019238063A1 (en) * | 2018-06-15 | 2019-12-19 | 众安信息技术服务有限公司 | Text detection and analysis method and apparatus, and device |
| CN111160395A (en) * | 2019-12-05 | 2020-05-15 | 北京三快在线科技有限公司 | Image recognition method and device, electronic equipment and storage medium |
| CN111401371A (en) * | 2020-06-03 | 2020-07-10 | 中邮消费金融有限公司 | Text detection and identification method and system and computer equipment |
| WO2020223859A1 (en) * | 2019-05-05 | 2020-11-12 | 华为技术有限公司 | Slanted text detection method, apparatus and device |
| CN112464872A (en) * | 2020-12-11 | 2021-03-09 | 广东电网有限责任公司 | Automatic extraction method and device based on NLP (non-line segment) natural language |
| WO2021047182A1 (en) * | 2019-09-11 | 2021-03-18 | 深圳壹账通智能科技有限公司 | Ocr-based picture data recognition method and apparatus, and computer device |
| CN112686319A (en) * | 2020-12-31 | 2021-04-20 | 南京太司德智能电气有限公司 | Merging method of electric power signal model training files |
| CN113159086A (en) * | 2020-12-31 | 2021-07-23 | 南京太司德智能电气有限公司 | Efficient power signal description model training method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103617415A (en) * | 2013-11-19 | 2014-03-05 | 北京京东尚科信息技术有限公司 | Device and method for automatically identifying invoice |
| CN104573688A (en) * | 2015-01-19 | 2015-04-29 | 电子科技大学 | Mobile platform tobacco laser code intelligent identification method and device based on deep learning |
| CN106096607A (en) * | 2016-06-12 | 2016-11-09 | 湘潭大学 | A kind of licence plate recognition method |
| CN107341523A (en) * | 2017-07-13 | 2017-11-10 | 浙江捷尚视觉科技股份有限公司 | Express delivery list information identifying method and system based on deep learning |
| CN107358232A (en) * | 2017-06-28 | 2017-11-17 | 中山大学新华学院 | Invoice recognition methods and identification and management system based on plug-in unit |
-
2018
- 2018-01-05 CN CN201810011763.4A patent/CN108133212B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103617415A (en) * | 2013-11-19 | 2014-03-05 | 北京京东尚科信息技术有限公司 | Device and method for automatically identifying invoice |
| CN104573688A (en) * | 2015-01-19 | 2015-04-29 | 电子科技大学 | Mobile platform tobacco laser code intelligent identification method and device based on deep learning |
| CN106096607A (en) * | 2016-06-12 | 2016-11-09 | 湘潭大学 | A kind of licence plate recognition method |
| CN107358232A (en) * | 2017-06-28 | 2017-11-17 | 中山大学新华学院 | Invoice recognition methods and identification and management system based on plug-in unit |
| CN107341523A (en) * | 2017-07-13 | 2017-11-10 | 浙江捷尚视觉科技股份有限公司 | Express delivery list information identifying method and system based on deep learning |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019238063A1 (en) * | 2018-06-15 | 2019-12-19 | 众安信息技术服务有限公司 | Text detection and analysis method and apparatus, and device |
| CN109002768A (en) * | 2018-06-22 | 2018-12-14 | 深源恒际科技有限公司 | Medical bill class text extraction method based on the identification of neural network text detection |
| CN109816118A (en) * | 2019-01-25 | 2019-05-28 | 上海深杳智能科技有限公司 | A kind of method and terminal of the creation structured document based on deep learning model |
| CN109816118B (en) * | 2019-01-25 | 2022-12-06 | 上海深杳智能科技有限公司 | Method and terminal for creating structured document based on deep learning model |
| CN109886257A (en) * | 2019-01-30 | 2019-06-14 | 四川长虹电器股份有限公司 | Using the method for deep learning correction invoice picture segmentation result in a kind of OCR system |
| CN109886257B (en) * | 2019-01-30 | 2022-10-18 | 四川长虹电器股份有限公司 | Method for correcting invoice image segmentation result by adopting deep learning in OCR system |
| CN109993160B (en) * | 2019-02-18 | 2022-02-25 | 北京联合大学 | Image correction and text and position identification method and system |
| CN109993160A (en) * | 2019-02-18 | 2019-07-09 | 北京联合大学 | A kind of image flame detection and text and location recognition method and system |
| CN109948617A (en) * | 2019-03-29 | 2019-06-28 | 南京邮电大学 | A kind of invoice image position method |
| WO2020223859A1 (en) * | 2019-05-05 | 2020-11-12 | 华为技术有限公司 | Slanted text detection method, apparatus and device |
| CN110348346A (en) * | 2019-06-28 | 2019-10-18 | 苏宁云计算有限公司 | A kind of bill classification recognition methods and system |
| WO2021047182A1 (en) * | 2019-09-11 | 2021-03-18 | 深圳壹账通智能科技有限公司 | Ocr-based picture data recognition method and apparatus, and computer device |
| CN111160395A (en) * | 2019-12-05 | 2020-05-15 | 北京三快在线科技有限公司 | Image recognition method and device, electronic equipment and storage medium |
| CN111401371A (en) * | 2020-06-03 | 2020-07-10 | 中邮消费金融有限公司 | Text detection and identification method and system and computer equipment |
| CN112464872A (en) * | 2020-12-11 | 2021-03-09 | 广东电网有限责任公司 | Automatic extraction method and device based on NLP (non-line segment) natural language |
| CN113159086A (en) * | 2020-12-31 | 2021-07-23 | 南京太司德智能电气有限公司 | Efficient power signal description model training method |
| CN112686319A (en) * | 2020-12-31 | 2021-04-20 | 南京太司德智能电气有限公司 | Merging method of electric power signal model training files |
| CN113159086B (en) * | 2020-12-31 | 2024-04-30 | 南京太司德智能电气有限公司 | Efficient electric power signal description model training method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108133212B (en) | 2021-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108133212A (en) | A kind of quota invoice amount identifying system based on deep learning | |
| Nguyen et al. | Robust minutiae extractor: Integrating deep networks and fingerprint domain knowledge | |
| Li et al. | Robust rooftop extraction from visible band images using higher order CRF | |
| Sirmacek et al. | Urban-area and building detection using SIFT keypoints and graph theory | |
| Costea et al. | Creating roadmaps in aerial images with generative adversarial networks and smoothing-based optimization | |
| Zhang et al. | CNN based suburban building detection using monocular high resolution Google Earth images | |
| CN111611643B (en) | Household vectorization data acquisition method and device, electronic equipment and storage medium | |
| CN107316036B (en) | Insect pest identification method based on cascade classifier | |
| TWI434220B (en) | A method for recognizing the identity of user by palm vein biometric | |
| US20100189316A1 (en) | Systems and methods for graph-based pattern recognition technology applied to the automated identification of fingerprints | |
| CN111738055B (en) | Multi-category text detection system and bill form detection method based on same | |
| CN108446699A (en) | Identity card pictorial information identifying system under a kind of complex scene | |
| CN105335760A (en) | Image number character recognition method | |
| CN110348289A (en) | A kind of finger vein identification method based on binary map | |
| JP6871658B2 (en) | Water area identification methods and equipment based on iterative classification | |
| Tokarczyk et al. | Beyond hand-crafted features in remote sensing | |
| Diaz-Escobar et al. | Natural Scene Text Detection and Segmentation Using Phase‐Based Regions and Character Retrieval | |
| Wang et al. | Big Map R-CNN for object detection in large-scale remote sensing images. | |
| CN116543325A (en) | Unmanned aerial vehicle image-based crop artificial intelligent automatic identification method and system | |
| Yin et al. | A self-supervised learning method for shadow detection in remote sensing imagery | |
| Hong et al. | A real-time critical part detection for the blurred image of infrared reconnaissance balloon with boundary curvature feature analysis | |
| Jiao et al. | Individual Building Rooftop and Tree Crown Segmentation from High‐Resolution Urban Aerial Optical Images | |
| Lukashevich et al. | Building detection on aerial and space images | |
| Du et al. | Shadow detection in high-resolution remote sensing image based on improved K-means | |
| Fan et al. | Vegetation coverage detection from very high resolution satellite imagery |
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 |