CN108875739A - A kind of accurate detecting method of digital displaying meter reading - Google Patents
A kind of accurate detecting method of digital displaying meter reading Download PDFInfo
- Publication number
- CN108875739A CN108875739A CN201810608757.7A CN201810608757A CN108875739A CN 108875739 A CN108875739 A CN 108875739A CN 201810608757 A CN201810608757 A CN 201810608757A CN 108875739 A CN108875739 A CN 108875739A
- Authority
- CN
- China
- Prior art keywords
- image
- meter reading
- network model
- training
- instrument image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/242—Aligning, centring, orientation detection or correction of the image by image rotation, e.g. by 90 degrees
-
- 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
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/02—Recognising information on displays, dials, clocks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses the accurate detecting methods of one of pattern-recognition and field of artificial intelligence digital displaying meter reading, including(1)Data acquisition:It include the Instrument image of reading area using picture pick-up device shooting;(2)Data processing:Meter reading value is manually marked, and carries out Random-Rotation, stretching and the translation transformation of Instrument image;(3)The building of depth network model and training:Instrument image and corresponding labeled data are input to depth network model to be trained;(4)Meter reading detection:Instrument image is inputted, system returns to the meter reading image of specification.The present invention overcomes the shortcomings of existing meter reading extracting method, make full use of the parameter learning ability of depth network model, the physical significance of confrontation type learning ability and anti-pass residual error based on depth network model, pass through the distribution of learning data sample, digital display meter reading is accurately detected, have the characteristics that strong real-time, accuracy rate are high, there is preferable practical value.
Description
Technical field
The present invention relates to pattern-recognitions and field of artificial intelligence, in particular to a kind of digital displaying meter is read
Accurate detecting method.
Background technique
The extraction of meter reading has a wide range of applications in various measurements and monitoring system, such as the meter reading of water, electricity and gas heat
In charging application, need periodically to read meter reading;In monitoring system, it is also desirable to meter reading is read periodically or in real time,
To realize monitoring and control to system.
There are mainly two types of modes at present for the extraction of meter reading:First is that by the way of manually reading, such as manual metering.
This mode is time-consuming and laborious, is also unfavorable for the automation of system.Second is that carrying out digital improvement, Ke Yizhi using to measuring instrumentss
Connect output digitized indications.This mode generally requires biggish cost input.Such as digital improvement is carried out to existing water meter,
It needs replacing digital water gauge, puts into larger, also make troubles to using.
Meter reading based on computer vision is extracted, and is shot Instrument image by camera, is utilized computer vision skill
Art automatically extracts meter reading.This method has plug and play, without being transformed original measuring instrumentss, it is low in cost the features such as.But
It is the meter reading extracting method currently based on computer vision, the main method using traditional images processing passes through image two
The separation modules such as value, edge detection, image segmentation realize the extraction of instrument reading area.This method is easy to be made an uproar by image
The interference of sound, accuracy rate be not high;And detect that speed is slow, practicability is low.
Drawbacks described above is worth solving.
Summary of the invention
In order to overcome the shortcomings of existing meter reading extracting method, the parameter learning energy of depth network model is made full use of
Power, the physical significance of confrontation type learning ability and anti-pass residual error based on depth network model pass through point of learning data sample
Cloth, provides a kind of accurate detecting method of digital displaying meter reading, and the present invention has the characteristics that strong real-time, accuracy rate are high, has
Preferable practical value.
Technical solution of the present invention is as described below:
A kind of accurate detecting method of digital displaying meter reading, which is characterized in that include the following steps:
S1:Data acquisition:It include the Instrument image of reading area using picture pick-up device shooting;
S2:Data processing:Meter reading value is manually marked, and carry out Instrument image low-angle Random-Rotation,
And stretching and the translation transformation of small scale;
S3:The building of depth network model and training:Instrument image and corresponding labeled data are input to depth network model
It is trained, Weakly supervised parameter learning is done using identification model in training process;
S4:Meter reading detection:Instrument image is inputted, system returns to the meter reading image of a specification.
According to the present invention of above scheme, which is characterized in that in the step S1, in the Instrument image of shooting, instrument
Table dial reading part occupies 2/3rds of image area or more.
According to the present invention of above scheme, which is characterized in that in the step S2, the image of shooting is labeled,
Marked content is meter reading, uses ", " to separate between each of reading number.
According to the present invention of above scheme, which is characterized in that the step S3 is specifically included:
S31:Construct deep neural network model;
S32:The setting of network model training parameter;
S33:The training of deep neural network is carried out under random initializtion parameter.
Further, the deep neural network model in the step S32 includes parameter prediction module and resampling module:
The parameter prediction module extracts the position feature information of reading from Instrument image, obtains advanced spy by convolutional neural networks
Sign figure, and predicted to obtain the parameter of reflection image rotation, zooming and panning characteristic by advanced features figure;The resampling module root
It is predicted that parameter image is rotated, zooming and panning, the Instrument image of a specification is obtained after resampling.
Further, the network model training parameter in the step S32 include the number of iterations, optimizer, learning rate with
And weight attenuation coefficient.
Further, the number of iterations is 1000000, and the optimizer uses ADADELTA method, the study
Rate is 1.0, and the weight attenuation coefficient is 0.0005.
Further, in the step S33, network model training in using residual error return algorithm, by from last
Layer calculates transmitting residual error, successively transmits, updates all parameters of network model.
Further, in the step S33, training process uses Weakly supervised Training strategy, and steps are as follows:
(1) using the Instrument image and markup information of acquisition, one general depth Network Recognition model of training works as input
It when one image, exports as the character string in image, trained loss function is CTC loss function;
(2) accurate detection network model is docking together with general depth Network Recognition model, it is accurate to detect network mould
The output of type is the input of general depth Network Recognition model;
(3) finally, continuing to use above-mentioned Instrument image and markup information is trained, i.e., Instrument image is input to accurately
Network model is detected, CTC loss is calculated according to the output of general depth Network Recognition model, and carry out gradient passback, carries out essence
The really training of detection network model.
According to the present invention of above scheme, which is characterized in that the step S4 concrete operation step is:Input an instrument
Image, depth network model accurately detect image, and return to the dial plate image of a specification.
According to the present invention of above scheme, the beneficial effect is that:
(1) present invention learning objective position distribution automatic from image data, the parameter learnt are suitable for really
The feature of image of scene, the prediction probability of success are higher.
(2) stringent Labeling Coordinate is not needed in network training, saves a large amount of manpower and material resources;Pass through Weakly supervised side
Formula, the gradient information with physical significance returned using identification model, effectively instructs parameter prediction module, thus logical
The adaptive ability for crossing network finds the image transformation that can most reduce identification difficulty.It can effectively improve in practical application
Detection accuracy is also beneficial to accurately identifying for subsequent detection result.
(3) residual error back propagation algorithm, adjust automatically convolution nuclear parameter, to obtain more robust filter, Neng Goushi are used
Answer the application scenarios such as fuzzy image, perspective transform, light variation.
(4) opposite manual type, the present invention can be automatically performed the detection of reading area, can use manpower and material resources sparingly.
(5) relatively traditional detection method based on computer vision, the present invention be not necessarily to carry out explicit image binaryzation and
Cutting procedure etc. has detection accuracy height, detection speed fast.After meter reading image normalization, it is suitable for image rotation
Equal application scenarios.
Detailed description of the invention
Fig. 1 is overall flow figure of the invention.
Fig. 2 is the structural block diagram of depth network model of the present invention.
Fig. 3 is the schematic diagram of the accurate testing result example of instrument of the invention.
Specific embodiment
With reference to the accompanying drawing and the present invention is further described in embodiment:
As shown in Figure 1, a kind of accurate detecting method of digital displaying meter reading, including data acquisition, data processing, depth net
Network model construction and training and meter reading detection and etc..
One, data acquisition:Include the Instrument image of reading area using picture pick-up device shooting, covers the to be checked of different size
Survey instrument.
Instrument dial plate image is shot using picture pick-up devices such as mobile phone, specialized hardwares.In shooting process, instrument dial plate reading portion
Point face camera lens, it is placed in the middle and occupy image area 2/3rds or more, instrument dial plate allows to tilt to a certain degree, but reads area
Domain needs to shoot complete.
Captured Instrument image should cover the instrument to be detected of different size, and Instrument image quantity is no less than 100000
?.
Two, data processing:The image of shooting is labeled, marked content is meter reading, each of reading number
Between use ", " separate;Then, the stretching and translation of low-angle Random-Rotation and small scale are carried out to the Instrument image of shooting
Transformation.
For just in the reading position of carry, annotation formatting be X.5, such as certain reading position value between 7 and 8, then need
It is labeled as " 7.5 ".Every shared 0-9 i.e. 10 digital states, in addition carry status, then every co-exists in 20 classes mark shape
State.Marked content is without including any coordinate information.
Three, the building of depth network model and training:Instrument image and corresponding labeled data are input to depth network model
It is trained, Weakly supervised parameter learning is done using identification model in training process.Specifically comprise the steps of:
1, deep neural network model is constructed
As shown in Fig. 2, constructed deep neural network model includes parameter prediction module and resampling module.Parameter is pre-
It surveys module and extracts the position feature information of reading from Instrument image, obtain advanced features figure by convolutional neural networks, and by
Advanced features figure is predicted to obtain 6 parameters of reflection image rotation, zooming and panning characteristic;Resampling module is according to the 6 of prediction
A parameter rotates image, zooming and panning, and the Instrument image of a specification is obtained after resampling.
In this norm image, meter reading region is located at image center, and direction is horizontal and reading area occupies image and surpasses
The area for crossing 90% can be used for search and storage of data image of the automatic identification, Instrument image data of meter reading etc..
The network structure of parameter prediction module is as shown in the table:
In parameter prediction module:The welt operation of convolutional layer is that a row/column is respectively sticked on four sides up and down in former characteristic pattern
Pixel, pixel value 0;Non-linear layer uses ReLU activation primitive;Pond layer is using maximum pond mode.
Parameter prediction module is in structure using the strategy that pond layer is preferential.As shown above, it is after input layer
Pond layer, pond layer are located at before convolutional layer.Calculation amount can be effectively reduced in this way, avoid the input of a large amount of noises, improve
Module robustness.
The input of parameter prediction module is instrument dial plate image, is exported as 6 parameters for resampling module.Parameter is pre-
It surveys module and has used depth network model, the neural net layer in upper table is the form being linked in sequence, and returns algorithm using residual error
Update the parameter in neural network.The module is obtained by Weakly supervised training, study close to accurate prediction result.
Resampling module uses STN (Spatial Transformer Networks) network structure, and treatment process is:It is first
Resampling, the meter reading image standardized are first carried out by STN network;Then the meter reading image of specification is sent into identification
Model carries out the Weakly supervised parameter training of whole detection network.
Wherein, the sampling process of STN network is shown below:
Wherein:θij, (i=1,2;J=1,2,3) totally 6 parameters, are the output of parameter prediction module.θ11, θ12, θ21, θ22
Input coordinate point is subjected to Two Dimensional Rotating and scaling, θ18, θ28Input coordinate point is translated.In detection network model training
Incipient stage, the biasing of the full articulamentum of network is initialized as " 1,0,0,0,1,0 ", bring into formula (1), then have STN net
The coordinate of network output is as before, is identical transformation.(xin,yin) it is pixel coordinate in input picture, (xout,yout) be
Coordinate after the resampling of the pixel.
2, the setting of network model training parameter:
The number of iterations:1000000;
Optimizer:Using ADADELTA method;
Learning rate:1.0 (learning rate more new strategies:It immobilizes);
Weight decay (weight attenuation coefficient):0.0005.
3, the training of deep neural network is carried out under random initializtion parameter.It is returned in network model training using residual error
Algorithm successively transmits by calculating transmitting residual error from the last layer, updates all parameters of network model.Training process uses
Weakly supervised Training strategy, steps are as follows:
(1) Instrument image and markup information of acquisition, one general depth Network Recognition model of training, the identification are utilized
Generally by convolutional layer-length, memory unit forms model in short-term.When image is opened in input one, export as the character string in image,
Trained loss function is CTC loss function.
(2) accurate detection network model is docking together with the general depth Network Recognition model, the former output is
The input of the latter.
(3) it continues to use above-mentioned Instrument image and markup information is trained, i.e., Instrument image is input to accurate detection
Network model calculates CTC loss according to the output of general depth Network Recognition model, and carries out gradient passback, is accurately examined
Survey the training of network model.
Due to not using any location coordinate information in the training process, but there is position by what identification model returned
The gradient of information physical meaning is a Weakly supervised parameter learning process to supervise the training process of accurate detection network model.
When accurately detection network model prediction close to it is correct when, identify network model recognition result can be easier correctly, confidence level
It is higher, so that positive feedback improves detection accuracy in accurately detection network model in gradient passback.
Four, meter reading detects:An Instrument image is inputted, depth network model accurately detects image, and returns
Return the meter reading image of a specification.
The meter reading image of specification can be used for Recognition of Reading, the search of Instrument image data and the storage of data image
Deng.The present invention can be automatically performed the detection of reading area, without carrying out explicit image binaryzation and cutting procedure etc.
As shown in figure 3, which show the accurate testing results of meter reading image.The model that the present invention provides is in practical survey
In examination, meter reading region can be made to occupy detection and obtain the area that image is more than 90%, detection accuracy is high.It can be used for
Search and the storage of data image of the automatic identification, Instrument image data of meter reading etc..In addition, the present invention also has detection
Fireballing feature, every Instrument image detection time are no more than 10 milliseconds, can satisfy the needs of real-time application.
It should be understood that for those of ordinary skills, it can be modified or changed according to the above description,
And all these modifications and variations should all belong to the protection domain of appended claims of the present invention.
Illustrative description has been carried out to the invention patent above in conjunction with attached drawing, it is clear that the realization of the invention patent not by
The limitation of aforesaid way, if the method concept of the invention patent and the various improvement of technical solution progress are used, or without
It improves and the conception and technical scheme of the invention patent is directly applied into other occasions, be within the scope of the invention.
Claims (10)
1. a kind of accurate detecting method of digital displaying meter reading, which is characterized in that include the following steps:
S1:Data acquisition:It include the Instrument image of reading area using picture pick-up device shooting;
S2:Data processing:Meter reading value is manually marked, and carries out the Random-Rotation of Instrument image, stretch and put down
Move transformation;
S3:The building of depth network model and training:Instrument image and corresponding labeled data are input to depth network model to carry out
It trains, Weakly supervised parameter learning is done using identification model in training process;
S4:Meter reading detection:Instrument image is inputted, system returns to the meter reading image of a specification.
2. the accurate detecting method of digital displaying meter reading according to claim 1, which is characterized in that in the step S1
In, in the Instrument image of shooting, instrument dial plate reading portion point occupies 2/3rds of image area or more.
3. the accurate detecting method of digital displaying meter reading according to claim 1, which is characterized in that in the step S2
In, the image of shooting is labeled, marked content is meter reading, uses ", " to separate between each of reading number.
4. the accurate detecting method of digital displaying meter reading according to claim 1, which is characterized in that the step S3 is specific
Including:
S31:Construct deep neural network model;
S32:The setting of network model training parameter;
S33:The training of deep neural network is carried out under random initializtion parameter.
5. the accurate detecting method of digital displaying meter reading according to claim 4, which is characterized in that in the step S32
Deep neural network model include parameter prediction module and resampling module:The parameter prediction module is mentioned from Instrument image
Several position feature information is read, obtains advanced features figure by convolutional neural networks, and predict to obtain instead by advanced features figure
Reflect the parameter of image rotation, zooming and panning characteristic;The resampling module rotates image according to the parameter of prediction, is contracted
It puts and translates, the Instrument image of a specification is obtained after resampling.
6. the accurate detecting method of digital displaying meter reading according to claim 4, which is characterized in that in the step S32
Network model training parameter include the number of iterations, optimizer, learning rate and weight attenuation coefficient.
7. the accurate detecting method of digital displaying meter according to claim 6 reading, which is characterized in that the number of iterations is
1000000, the optimizer uses ADADELTA method, and the learning rate is 1.0, and the weight attenuation coefficient is 0.0005.
8. the accurate detecting method of digital displaying meter reading according to claim 4, which is characterized in that in the step S33
In, algorithm is returned using residual error in network model training, by calculating transmitting residual error from the last layer, successively transmits, updates net
All parameters of network model.
9. the accurate detecting method of digital displaying meter reading according to claim 4, which is characterized in that in the step S33
In, training process uses Weakly supervised Training strategy, and steps are as follows:
(1)Utilize the Instrument image and markup information of acquisition, one general depth Network Recognition model of training, when input one is opened
It when image, exports as the character string in image, trained loss function is CTC loss function;
(2)Accurate detection network model is docking together with general depth Network Recognition model, accurate detection network model
Output is the input of general depth Network Recognition model;
(3)Finally, continuing to use above-mentioned Instrument image and markup information is trained, i.e., Instrument image is input to accurate detection
Network model calculates CTC loss according to the output of general depth Network Recognition model, and carries out gradient passback, is accurately examined
Survey the training of network model.
10. the accurate detecting method of digital displaying meter reading according to claim 1, which is characterized in that the step S4 tool
Body operating procedure is:An Instrument image is inputted, depth network model accurately detects image, and returns to a specification
Dial plate image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810608757.7A CN108875739A (en) | 2018-06-13 | 2018-06-13 | A kind of accurate detecting method of digital displaying meter reading |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810608757.7A CN108875739A (en) | 2018-06-13 | 2018-06-13 | A kind of accurate detecting method of digital displaying meter reading |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108875739A true CN108875739A (en) | 2018-11-23 |
Family
ID=64338278
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810608757.7A Pending CN108875739A (en) | 2018-06-13 | 2018-06-13 | A kind of accurate detecting method of digital displaying meter reading |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108875739A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109840497A (en) * | 2019-01-30 | 2019-06-04 | 华南理工大学 | A kind of pointer-type water meter reading detection method based on deep learning |
CN109871754A (en) * | 2019-01-08 | 2019-06-11 | 深圳禾思众成科技有限公司 | A kind of instrument read method, equipment and computer readable storage medium |
CN110097041A (en) * | 2019-03-08 | 2019-08-06 | 贵州电网有限责任公司 | A kind of standard character method for quickly identifying for electric instrument inspection |
CN110717462A (en) * | 2019-10-12 | 2020-01-21 | 上海市建筑科学研究院 | Digital instrument reading identification method, device, equipment and medium |
CN111199194A (en) * | 2019-12-25 | 2020-05-26 | 吉林大学 | Automobile intelligent cabin instrument testing method based on machine vision and deep learning |
CN112001383A (en) * | 2020-08-10 | 2020-11-27 | 长沙奇巧匠人软件有限公司 | Water meter code intelligent identification method based on convolutional neural network technology |
CN113688817A (en) * | 2021-08-05 | 2021-11-23 | 同济人工智能研究院(苏州)有限公司 | Instrument identification method and system for automatic inspection |
CN113743405A (en) * | 2021-09-07 | 2021-12-03 | 南方电网数字电网研究院有限公司 | Intelligent meter reading method and device for electric energy meter |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103034838A (en) * | 2012-12-03 | 2013-04-10 | 中国人民解放军63963部队 | Special vehicle instrument type identification and calibration method based on image characteristics |
CN106529537A (en) * | 2016-11-22 | 2017-03-22 | 亿嘉和科技股份有限公司 | Digital meter reading image recognition method |
CN106909941A (en) * | 2017-02-27 | 2017-06-30 | 广东工业大学 | Multilist character recognition system and method based on machine vision |
-
2018
- 2018-06-13 CN CN201810608757.7A patent/CN108875739A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103034838A (en) * | 2012-12-03 | 2013-04-10 | 中国人民解放军63963部队 | Special vehicle instrument type identification and calibration method based on image characteristics |
CN106529537A (en) * | 2016-11-22 | 2017-03-22 | 亿嘉和科技股份有限公司 | Digital meter reading image recognition method |
CN106909941A (en) * | 2017-02-27 | 2017-06-30 | 广东工业大学 | Multilist character recognition system and method based on machine vision |
Non-Patent Citations (5)
Title |
---|
SHUTAO ZHAO ET AL: "Research on Remote Meter Automatic Reading Based on Computer Vision", 《 2005 IEEE/PES TRANSMISSION & DISTRIBUTION CONFERENCE & EXPOSITION: ASIA AND PACIFIC》 * |
YUEZHONG LI ET AL: "Research on water meter reading system based on LoRa communication", 《2017 IEEE INTERNATIONAL CONFERENCE ON SMART GRID AND SMART CITIES (ICSGSC)》 * |
唐梦婷: "基于机器视觉的仪表示值识别系统设计", 《中国优秀硕士学位论文全文数据库(电子期刊)信息科技辑》 * |
肖佳: "基于机器视觉的数字仪表自动读数方法研究", 《万方学位论文库》 * |
莫宇琨等: "深度卷积神经网络在水表字符图像识别上的应用", 《供水技术》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109871754A (en) * | 2019-01-08 | 2019-06-11 | 深圳禾思众成科技有限公司 | A kind of instrument read method, equipment and computer readable storage medium |
CN109840497A (en) * | 2019-01-30 | 2019-06-04 | 华南理工大学 | A kind of pointer-type water meter reading detection method based on deep learning |
CN110097041A (en) * | 2019-03-08 | 2019-08-06 | 贵州电网有限责任公司 | A kind of standard character method for quickly identifying for electric instrument inspection |
CN110717462A (en) * | 2019-10-12 | 2020-01-21 | 上海市建筑科学研究院 | Digital instrument reading identification method, device, equipment and medium |
CN110717462B (en) * | 2019-10-12 | 2023-02-28 | 上海市建筑科学研究院 | Digital instrument reading identification method, device, equipment and medium |
CN111199194A (en) * | 2019-12-25 | 2020-05-26 | 吉林大学 | Automobile intelligent cabin instrument testing method based on machine vision and deep learning |
CN112001383A (en) * | 2020-08-10 | 2020-11-27 | 长沙奇巧匠人软件有限公司 | Water meter code intelligent identification method based on convolutional neural network technology |
CN112001383B (en) * | 2020-08-10 | 2024-09-10 | 长沙奇巧匠人软件有限公司 | Water meter code intelligent identification method based on convolutional neural network technology |
CN113688817A (en) * | 2021-08-05 | 2021-11-23 | 同济人工智能研究院(苏州)有限公司 | Instrument identification method and system for automatic inspection |
CN113743405A (en) * | 2021-09-07 | 2021-12-03 | 南方电网数字电网研究院有限公司 | Intelligent meter reading method and device for electric energy meter |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108875739A (en) | A kind of accurate detecting method of digital displaying meter reading | |
CN106156734B (en) | Water flow velocity measurement method based on convolutional neural network image recognition | |
CN108830271A (en) | A kind of digital displaying meter Recognition of Reading method based on convolutional neural networks | |
CN109583324A (en) | A kind of pointer meters reading automatic identifying method based on the more box detectors of single-point | |
CN109800697B (en) | Transformer target detection and appearance defect identification method based on VGG-net style migration | |
CN109409365A (en) | It is a kind of that method is identified and positioned to fruit-picking based on depth targets detection | |
CN111141653B (en) | Tunnel leakage rate prediction method based on neural network | |
CN105243386A (en) | Face living judgment method and system | |
CN110263790A (en) | A kind of power plant's ammeter character locating and recognition methods based on convolutional neural networks | |
CN108921177A (en) | The instrument localization method of Intelligent Mobile Robot | |
CN110533069A (en) | A kind of two-dimentional chaff distribution character recognition methods based on algorithm of support vector machine | |
CN116229052B (en) | Method for detecting state change of substation equipment based on twin network | |
CN109840497A (en) | A kind of pointer-type water meter reading detection method based on deep learning | |
CN105488541A (en) | Natural feature point identification method based on machine learning in augmented reality system | |
CN110427943A (en) | A kind of intelligent electric meter technique for partitioning based on R-CNN | |
CN112349057A (en) | Deep learning-based indoor smoke and fire detection method | |
CN114861761B (en) | Loop detection method based on twin network characteristics and geometric verification | |
Peng et al. | Research on oil leakage detection in power plant oil depot pipeline based on improved YOLO v5 | |
CN112001964A (en) | Flood evolution process inundation range measuring method based on deep learning | |
CN117372854A (en) | Real-time detection method for hidden danger diseases of deep water structure of dam | |
CN115690614A (en) | Test method for improving embankment piping data acquisition precision | |
CN117541887A (en) | Water deficit detection model training and water deficit detection method, device and equipment | |
CN113034359A (en) | Intelligent diagnosis method and device for generator set stator and rotor defects, computer equipment and storage medium | |
CN116883893A (en) | Tunnel face underground water intelligent identification method and system based on infrared thermal imaging | |
CN109753896A (en) | A kind of unsupervised heterologous method for detecting change of remote sensing image based on general character autocoder |
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: 20181123 |
|
RJ01 | Rejection of invention patent application after publication |