CN116844181A - Intelligent recognition system and method for metering equipment to bill - Google Patents

Intelligent recognition system and method for metering equipment to bill Download PDF

Info

Publication number
CN116844181A
CN116844181A CN202310609685.9A CN202310609685A CN116844181A CN 116844181 A CN116844181 A CN 116844181A CN 202310609685 A CN202310609685 A CN 202310609685A CN 116844181 A CN116844181 A CN 116844181A
Authority
CN
China
Prior art keywords
module
recognition
identification
information
arrival
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
Application number
CN202310609685.9A
Other languages
Chinese (zh)
Inventor
胡厚鹏
欧家祥
王吉
肖艳红
周密
何沛林
唐建林
李航峰
罗奕
陈泽瑞
李富盛
邓钥丹
高正浩
吴欣
李鹏程
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CSG Electric Power Research Institute
Guizhou Power Grid Co Ltd
Original Assignee
CSG Electric Power Research Institute
Guizhou Power Grid Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by CSG Electric Power Research Institute, Guizhou Power Grid Co Ltd filed Critical CSG Electric Power Research Institute
Priority to CN202310609685.9A priority Critical patent/CN116844181A/en
Publication of CN116844181A publication Critical patent/CN116844181A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06V30/40—Document-oriented image-based pattern recognition
    • G06V30/42—Document-oriented image-based pattern recognition based on the type of document
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • G06V30/18—Extraction of features or characteristics of the image

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Character Discrimination (AREA)

Abstract

The application discloses an intelligent recognition system and method for metering equipment to a manifest. The intelligent recognition system comprises a arrival bill acquisition module, an image preprocessing module, an OCR recognition module, a recognition judgment module, a template recognition module, an intelligent recognition module, a similarity check module, a manual processing module and a database. The application solves the problems of low recognition efficiency, high rejection rate and high false recognition rate of the existing uniform recognition methods for different bill templates, and solves the problems of high calculation complexity, long recognition waiting time, incapability of releasing manual labor force to the maximum extent, unreliable and inadequately intelligent recognition results and the like.

Description

Intelligent recognition system and method for metering equipment to bill
Technical Field
The application relates to the field of image recognition, in particular to an intelligent recognition system for metering equipment to a manifest, and also relates to an intelligent recognition and management method for metering equipment to the manifest.
Background
Check-in includes check-in, unpacking, full performance testing, sampling detection, etc. performed after the metering device arrives. The metering equipment comprises a single three-phase electric energy meter, a special transformer terminal, a mutual inductor, a three-phase digital standard electric energy meter, a high-precision time base source and a three-phase electric energy meter calibrating device. At present, the information about the check and acceptance of the metering equipment is managed and maintained, and the information needs to be added, modified, deleted, inquired and exported. The metering equipment check-in information mainly comprises: equipment name, technical model number, factory number, shipment information (contract number, contract name, supplier, factory number, factory date, etc.), shipment details (asset number, factory number, shipment date, etc.), acceptance details (asset number, acceptance type, acceptance date, acceptance conclusion, acceptance person, acceptance attachment, etc.). The collection work of the metering equipment to the manifest data is mainly completed manually, the arrival manifest of the original metering equipment is firstly tidied, then scanned, and finally the manifest acceptance information of the metering equipment is manually input and checked. Even if the intelligent recognition method is adopted, the situation that the recognition error word is even unrecognizable often occurs, so that the work efficiency of acquiring the acquired manifest data is very low, and much time and labor are consumed.
On the other hand, because the names of the business fields in the bill are greatly different due to different issuers from the metering equipment to the bill, the electronic equipment cannot intelligently identify the business fields with different names but the same meaning, and the business efficiency of the arrival and the filing is required to be improved.
The application number is CN202210322687.5, and the Chinese application patent application with publication date being 2022.06.17 discloses a bill identification method, a system and a computer readable storage medium, wherein the method comprises the following steps: constructing a journey information model and a bill major model; receiving a form filling request, and identifying the general types of the receipts in the form filling request; if the bill major class is identified as the travel class, identifying the bill on the basis of the trained travel information model and the bill major class model in sequence, and outputting respective identification results to the corresponding bill; if the bill major class is not the travel class, the bill is identified based on the bill major class model, and the identification result is output to the corresponding bill. By the scheme, the form filling accuracy and the form filling efficiency can be improved, and the filling error rate and the initial review check-out rate of the traditional manual account reporting are reduced.
However, the above-described technique has at least the following technical problems: most of the existing bill identification systems and methods do not have a division identification method, and a unified identification method is adopted for different bill templates, so that the problems of low identification efficiency and high rejection rate and false identification rate exist; moreover, the calculation complexity is high, the recognition waiting time is long, the artificial labor force cannot be liberated to the maximum extent, and the recognition result is not reliable enough and intelligent enough.
Disclosure of Invention
The application provides an intelligent recognition system and method for metering equipment to a manifest, which aims at: (1) The method solves the problems that the existing different bill templates adopt a uniform identification method, and the identification efficiency is low, and the rejection rate and the false recognition rate are high; (2) The method solves the problems that the calculation complexity is high, the recognition waiting time is long, the artificial labor force cannot be liberated to the maximum extent, and the recognition result is not reliable enough and intelligent enough.
The technical scheme of the application is as follows:
an intelligent recognition system for metering equipment to a manifest comprises a manifest acquisition module, an image preprocessing module, an OCR recognition module, a recognition judging module, a template recognition module, an intelligent recognition module, a similarity checking module, a manual processing module and a database;
the manifest acquisition module is connected with the image preprocessing module, the image preprocessing module is connected with the database and is also connected with the OCR recognition module, the OCR recognition module is connected with the recognition judgment module, the recognition judgment module is respectively connected with the template recognition module and the intelligent recognition module, the template recognition module is also connected with the intelligent recognition module, the template recognition module and the intelligent recognition module are connected with the similarity checking module, and the similarity checking module is connected with the manual processing module.
Further improvements to the intelligent identification system of the metering device to the manifest:
the arrival bill acquisition module is used for acquiring an arrival bill image and sending the arrival bill image to the image preprocessing module in a data transmission mode;
the image preprocessing module is used for carrying out preliminary processing on the arrival bill image and respectively transmitting the preprocessed arrival bill image to the OCR module and the database in a data transmission mode;
the OCR module is used for identifying texts in the arrival bill image so as to obtain a primary identification result of the target text, and is also used for sending the primary identification result to the identification judgment module in a data transmission mode;
the identification judging module is used for judging the type of the arrival bill identification of the metering equipment according to the primary identification result, wherein the arrival bill identification type comprises template identification and complete identification, and is also used for sending the identified primary identification result to the template identification module or the intelligent identification module in a data transmission mode according to the type of the identification result.
Further improvements to the intelligent identification system of the metering device to the manifest:
the template identification module comprises a content matching unit 501 and a filling unit 502; the content matching unit 501 is configured to match the specific identified content that follows the attribute information in the initial identification result of the current manifest with the text information that is already in the information base of the database;
the filling unit 502 is configured to fill specific content successfully matched with text information in the information base to a corresponding position;
the template identification module is also used for transmitting the arrival order image of the unrecognized area to the intelligent identification module in a data transmission mode.
Further improvements to the intelligent identification system of the metering device to the manifest:
the intelligent recognition module is used for constructing an intelligent recognition neural network model, and efficiently recognizing the arrival order image of the metering equipment to obtain the arrival acceptance information; the intelligent recognition module is also used for sending the check-in information of the arrival goods to the similarity checking module in a data transmission mode.
The application also discloses an intelligent recognition method of the metering equipment to the manifest, which comprises the following steps:
s1, acquiring an arrival bill image of metering equipment, acquiring a primary identification result of the arrival bill image by adopting an OCR technology, and judging the type of arrival bill identification according to attribute information and position information in the primary identification result;
s2, selecting a template identification method or a complete identification method to identify according to the type of the arrival bill identification;
the template identification method is used for matching specific contents of attribute information in the initial identification result of the manifest with text information existing in an information base, filling the successfully matched specific contents to corresponding positions, and further identifying the unfilled specific contents through an intelligent identification neural network model;
the complete recognition method is used for carrying out complete recognition on all information on the arrival bill image through the intelligent recognition neural network model, matching the recognized arrival bill acceptance information from the information base through a text similarity algorithm and a semantic model based on a depth network, and updating the information base to the bill image base and the information base.
As a further improvement of the intelligent recognition method from the metering device to the manifest: in step S1, the specific judgment method of the arrival bill identification type is as follows: determining attribute information and corresponding position information in each template, searching the attribute information from the primary identification result and obtaining the position information of the attribute information, so as to calculate the matching degree of the current primary identification result and the template; if the number of the attribute information with the matching degree meeting the preset threshold range reaches a preset percentage, the current primary identification result is indicated to be matched with the current template; otherwise, matching the current primary identification result with the next template until the corresponding template is matched or the matching calculation with all templates is finished;
the method for judging the type of the manifest identification according to the initial identification result comprises the following steps: if the typesetting of the goods receipt information in the initial recognition result is matched with the existing goods receipt template, template recognition is adopted, and if the typesetting of the goods receipt information in the initial recognition result is not matched with the existing goods receipt template, complete recognition is used.
As a further improvement of the intelligent recognition method from the metering device to the manifest: using a template recognition module to perform template recognition; the template recognition module comprises a content matching unit 501 and a filling unit;
the content matching unit matches the specific content which is identified and follows the attribute information in the initial identification result of the current manifest with the text information in the information base, and the filling unit fills the specific content which is successfully matched with the text information in the information base to the corresponding position; and carrying out high-efficiency identification on the unfilled specific content through an intelligent identification neural network model, so as to perfect the goods acceptance information.
As a further improvement of the intelligent recognition method from the metering device to the manifest:
the intelligent recognition neural network model is used for complete recognition, and the realization process of the intelligent recognition neural network model is as follows: acquiring historical arrival bill images and arrival acceptance information in a corresponding information base thereof from an arrival bill image base as training samples of the intelligent recognition neural network, inputting the arrival bill images into the intelligent recognition neural network, outputting text information corresponding to the arrival bill images through deep learning of the neural network, performing error calculation with actual arrival bill acceptance information, correcting parameters in the intelligent recognition neural network according to the errors until the output precision of the intelligent recognition neural network reaches an expected effect, and thus finishing training of the intelligent recognition neural network;
the intelligent recognition neural network comprises an input layer, a multi-size convolution layer, a fusion layer, a clustering layer, a mapping layer and an output layer.
As a further improvement of the intelligent recognition method from the metering device to the manifest: the process of using the intelligent recognition neural network model for complete recognition is as follows: leading the metering equipment to the bill image sample information into an input layer, and transmitting the bill image sample information to a multi-size convolution layer by the input layer, wherein the multi-size convolution layer comprises N convolution layers with different sizes of convolution kernels, and extracting N corresponding image characteristic information under the different sizes of convolution kernels of the bill image sample information by the multi-size convolution layer; the multi-size convolution layers transmit the image features extracted by each convolution layer to the fusion layer;
the fusion layer fuses the multi-size image features, and the fusion layer transmits the fused features to the clustering layer;
the clustering layer clusters according to the distribution condition of the image features so as to divide different characters in the image, the clustering layer transmits the image features of the divided different characters to the mapping layer, the mapping layer maps the image features of the characters from the image space to the text space by using a mapping function, and the mapping layer transmits the obtained text information of the characters to the output layer;
the output layer outputs the text information recognized in the bill image, namely the check-in information.
As a further improvement of the intelligent recognition method from the metering device to the manifest: after the complete arrival acceptance information is obtained through complete identification, a new template is constructed according to the identification result, and a template library is updated.
Compared with the prior art, the application has the following beneficial effects:
(1) According to the application, the initial recognition result obtained by adopting the OCR technology on the metering equipment arrival bill image is further recognized, the matching degree of the initial recognition result and the template is calculated, the gradient of difference calculation is smaller, so that training is more stable, the difference calculation convergence is higher, and the matching reaches higher precision, so that a more proper recognition method is selected for recognizing the metering equipment arrival bill image, the recognition efficiency is improved, the recognition rejection rate and the false recognition rate are reduced, the reliability of the recognition result is improved, the calculation complexity is reduced, and the recognition waiting time is shortened.
(2) The method comprises the steps of respectively selecting a template recognition method and a complete recognition method for intelligent recognition of the arrival image of the metering equipment, constructing an intelligent recognition neural network model, extracting image feature information of different sizes of arrival image sample information of the arrival image by a multi-size convolution layer, carrying out feature fusion, mapping the image features of characters from an image space to a text space by a mapping layer by using a mapping function, obtaining arrival acceptance information of the metering equipment with higher recognition accuracy, and improving accuracy of the arrival image recognition of the arrival image of the metering equipment.
(3) The one-key uploading of the arrival information of the metering equipment can be realized, the business efficiency of the arrival filing of the equipment is greatly improved, and the standardization, convenience and reliability of arrival bill identification are improved.
Drawings
FIG. 1 is a schematic diagram of an identification system according to the present application;
FIG. 2 is a flow chart of the identification and management method of the present application.
Detailed Description
The technical scheme of the application is described in detail below with reference to the accompanying drawings:
the embodiment provides an intelligent recognition system and method for metering equipment to a manifest, wherein the overall thought is as follows:
further recognizing the initial recognition result obtained by OCR technology on the arrival bill image of the metering equipment, calculating the matching degree of the initial recognition result and the template, and calculating the gradient of the difference to be smaller, so that the training is more stable, the difference calculation convergence is higher, and the matching is higher in precision, thereby selecting a proper recognition method for recognizing the arrival bill image, and reducing the calculation complexity; the method comprises the steps of respectively selecting a template identification method and a complete identification method for intelligent identification of a metering equipment arrival order image, constructing an intelligent identification neural network model, extracting image characteristic information of different sizes of arrival order image sample information by a multi-size convolution layer, carrying out characteristic fusion, mapping the image characteristic of characters from an image space to a text space by a mapping layer by using a mapping function, obtaining metering equipment arrival order acceptance information with higher identification accuracy, improving the accuracy of metering equipment arrival order image identification, realizing one-key uploading of metering equipment arrival information, greatly improving the business efficiency of arrival and filing of the metering equipment, and improving the standard, convenience and reliability of arrival order identification.
In order to better understand the above technical solutions, the following description will explain the above technical solutions in detail with reference to the accompanying drawings.
Referring to fig. 1, an intelligent identification system for metering equipment to a manifest includes the following: the system comprises a receipt acquisition module 10, an image preprocessing module 20, an OCR (optical character recognition) module 30, a recognition judging module 40, a template recognition module 50, an intelligent recognition module 60, a similarity check module 70, a manual processing module 80 and a database 90.
The order acquisition module 10 is configured to acquire an order image, and send the order image to the image preprocessing module 20 by means of data transmission.
The image preprocessing module 20 is configured to perform preliminary processing on the arrival order image, including operations such as image graying, binarization, noise reduction, smoothing, etc., and the image preprocessing module 20 sends the preprocessed arrival order image to the OCR recognition module 30 and the database 90 in a data transmission manner.
The OCR (optical character recognition) module 30 is configured to recognize the text in the arrival order image by using an OCR (optical character recognition) technology, so as to obtain a primary recognition result of the target text, and the OCR module 30 sends the primary recognition result to the recognition judging module 40 through data transmission.
The identification judging module 40 is configured to judge the type of the bill identification according to the initial identification result, where the type of the bill identification includes template identification and complete identification, and the identification judging module 40 sends the identified initial identification result to the template identifying module 50 or the intelligent identifying module 60 in a data transmission manner according to the type of the bill identification.
The template recognition module 50 includes a content matching unit 501 and a filling unit 502. The content matching unit 501 is configured to match the specific identified content that follows the attribute information in the initial identification result of the current manifest with the text information that is already in the information base of the database 90; the filling unit 502 is configured to fill the specific content successfully matched with the text information in the information base to a corresponding position, and the template recognition module 50 further sends the arrival order image of the unrecognized area to the intelligent recognition module 60 through data transmission.
The intelligent recognition module 60 is configured to construct an intelligent recognition neural network model, and perform efficient recognition on the arrival order image to obtain the arrival acceptance information. The intelligent recognition module 60 transmits the check-in information to the similarity checking module 70 through data transmission.
The similarity checking module 70 is configured to match the check-in information identified by the intelligent identification neural network model with the information base of the database 90 through a text similarity algorithm. If there is an unmatched text, matching the unmatched text with the text in the information base by using a semantic model based on the depth network, updating the successfully matched text information into the information base, and transmitting the successfully unmatched text information to the manual processing module 80 in a data transmission mode.
The manual processing module 80 is configured to display text information that cannot be matched to a user, and perform confirmation by a person, and the manual processing module 80 updates the confirmation result to the database 90.
The database 90 comprises a receipt image library and an information library, wherein the receipt image library stores the pretreated receipt image into the database 90; the information base stores all text information in the history to manifest.
Referring to fig. 2, the intelligent recognition method for metering equipment to a manifest according to the application comprises the following steps:
s1, acquiring a manifest image, acquiring a primary recognition result of the manifest image by adopting an OCR technology, and judging the type of manifest recognition according to attribute information and position information in the primary recognition result.
After receiving the arrival bill of the metering equipment, the arrival bill intelligent recognition system needs to intelligently recognize the arrival bill of the metering equipment, recognize the arrival acceptance information recorded on the arrival bill, manage and maintain the recognized arrival acceptance information, support the new addition, modification, deletion, inquiry and export of the information, and support the rapid import in a photographing or picture mode through a mobile phone browser.
And the metering equipment arrival bill intelligent recognition system acquires an arrival bill image.
In some embodiments of the present application, the order image may be obtained by photographing or importing through a mobile phone browser by a user, and the order image is automatically transferred to the order acquisition module 10.
In an alternative embodiment, the metering device arrival image may be an arrival image obtained by the user scanning the arrival by the scanner, and the arrival image is automatically transmitted to the arrival acquisition module 10.
It will be appreciated that the metering device may acquire the bill image in a plurality of ways, and the method is not limited to the above-mentioned ways of photographing, importing and scanning, and the application is not limited thereto.
The image preprocessing module 20 performs preliminary processing on the arrival order image, including operations of image graying, binarization, noise reduction, smoothing, etc., and the preprocessing methods are all well known in the art.
After preprocessing the image, the OCR recognition module 30 recognizes the text in the metering device to the bill image using OCR (optical character recognition) technology to obtain the primary recognition result of the target text. However, the existing OCR software has high rejection rate and false recognition rate, so that only a fuzzy result of the goods acceptance information can be obtained.
The identification judgment module 40 judges the type of the arrival bill identification of the metering device according to the initial identification result, wherein the arrival bill identification type comprises template identification and complete identification. If the typesetting of the goods receipt information in the initial recognition result is matched with the existing goods receipt template, template recognition is adopted, and if the typesetting of the goods receipt information in the initial recognition result is not matched with the existing goods receipt template, complete recognition is used.
The specific judging method of the arrival bill identification type is as follows:
attribute information in the check-in information is first determined, the attribute information including:
(1) Arrival information: contract number, contract name, vendor, equipment major class, equipment minor class, equipment name, quantity of arrival, date of arrival, recipient, asset number, equipment code, model specification, factory number, factory date;
(2) To the goods details: asset number, equipment name, technical model number, factory number, date of arrival, whether tracing is needed, supplier, contract number;
(3) Acceptance details: asset number, type of acceptance (send-check, self-check, must-check, other verification), date of acceptance, conclusion of acceptance, person of acceptance, value tracing result, acceptance attachment.
Determining attribute information and corresponding position information in each template, finding out the attribute information from the primary identification result and obtaining the position information of the attribute information, thereby calculating the matching degree of the current primary identification result and the template, wherein the specific calculation method comprises the following steps:
wherein md i Representing the position matching degree of the initial recognition result and the ith attribute information in the template, p dem Representing attribute location information, p, in a template dis And the position information of the corresponding attribute in the initial recognition result is represented. If the number of the attribute information with the matching degree meeting the preset threshold range reaches a preset percentage, the current primary identification result is indicated to be matched with the current template; otherwise, the current initial recognition result is matched with the next template until the corresponding template is matched or the matching calculation with all templates is finished, so that whether the template recognition or the complete recognition is needed for the arrival order of the current metering equipment is determined.
The beneficial effects of the step S1 are as follows: the initial recognition result obtained by the OCR technology is further recognized, the matching degree of the initial recognition result and the template is calculated, the gradient of difference calculation is small, training is stable, the matching precision is high due to high convergence of the difference calculation, and therefore a proper recognition method is selected for recognition of the metering equipment arrival order image, and the calculation complexity is reduced.
S2, the template identification method matches specific contents of attribute information in the initial identification result of the manifest with text information existing in an information base, the successfully matched specific contents are filled in corresponding positions, and unfilled specific contents are efficiently identified through the intelligent identification neural network model. The complete recognition method is used for carrying out complete recognition on all information on the arrival bill image through the intelligent recognition neural network model, matching the recognized arrival bill acceptance information from the information base through a text similarity algorithm and a semantic model based on a depth network, and updating the information base to the bill image base and the information base.
The intelligent recognition system for the arrival bill is provided with a database 90, wherein the database 90 comprises an arrival bill image library and an information library. The order image library stores the pretreated order image into the database 90. The information base stores all text information in the history to manifest.
If the recognition judging module 40 judges that the template recognition method is needed for the current arrival bill, the specific content matching unit 501 in the template recognition module 50 matches the recognized specific content following the attribute information in the initial recognition result of the current arrival bill with the text information existing in the information base, and the filling unit 502 fills the specific content successfully matched with the text information in the information base to the corresponding position.
The unfilled concrete content is efficiently identified by the intelligent recognition neural network model in the intelligent recognition module 60, so that the goods acceptance information is perfected.
The specific implementation process of the intelligent recognition neural network model is as follows:
acquiring historical arrival bill images and arrival acceptance information in a corresponding information base thereof from an arrival bill image base, taking the arrival bill images as training samples of the intelligent recognition neural network, inputting the arrival bill images into the intelligent recognition neural network, outputting text information corresponding to the arrival bill images through deep learning of the neural network, performing error calculation with actual arrival bill acceptance information, correcting parameters in the intelligent recognition neural network according to the errors until the output precision of the intelligent recognition neural network reaches an expected effect, and thus finishing training of the intelligent recognition neural network.
The intelligent recognition neural network comprises an input layer, a multi-size convolution layer, a fusion layer, a clustering layer, a mapping layer and an output layer.
The method comprises the steps that arrival order image sample information is imported into an input layer, the input layer transmits the arrival order image sample information to a multi-size convolution layer, the multi-size convolution layer comprises N convolution layers with different sizes of convolution kernels, the multi-size convolution layer extracts N corresponding image characteristic information under the different sizes of convolution kernels of the arrival order image sample information, and an activation function of the multi-size convolution layer is as follows:
f＝max(Z，m*(1-expβ))*R(σ*Z)+δ；
wherein f is an activation function of a multi-size convolution layer, Z represents the image sample information of the arrival order after convolution operation, m represents the number of image channels, beta represents a convolution kernel parameter, R (·) is an image feature extractor, sigma represents the convolution kernel size, and delta represents an adjustment factor. The multi-size convolution layer transmits the image features extracted by each convolution layer to the fusion layer, and the output of the multi-size convolution layer is as follows:
wherein OP j Represents the output of the j-th convolution layer, j E [1, N]ω represents the weight of the manifest image sample information,the convolution operation is represented by z, the manifest image sample information, and b, the deviation amount.
The fusion layer fuses the multi-size image features, and the fusion calculation formula is as follows:
wherein F represents fusion characteristics, MLP represents multi-layer perceptron operation,representing a join operation. The fusion layer transmits the fusion features to the clustering layer.
The clustering layer clusters according to the distribution condition of the image features so as to divide different characters in the image, and the clustering method adopts the prior art, such as k-clustering and the like. The clustering layer transmits the image features of the separated different words to the mapping layer.
The mapping layer maps the image features of the characters from an image space to a text space by using a mapping function, and the mapping method comprises the following steps:
wherein MF represents a mapping function of image features of any one character, Δd x Representing the coordinate difference value delta d of the current character image characteristic in the horizontal direction y Representing the coordinate difference value of the current character image feature in the vertical direction, G (x, y) represents the pixel point coordinate function of the current character image, x represents the coordinate in the horizontal direction, y represents the coordinate in the vertical direction, x max And x min Respectively representing the maximum value and the minimum value of the horizontal coordinate of the current character image characteristic, y max And y min Respectively representing the maximum value and the minimum value of the current character image feature in the vertical direction coordinate. The mapping layer transmits the obtained literal text information to the output layer. The output layer outputs the text information recognized in the bill image, namely the check-in information.
The similarity checking module 70 matches the check-in information identified by the intelligent identification neural network model with the information base through a text similarity algorithm, wherein the text similarity algorithm can select a cosine similarity algorithm and the like. If the text information which cannot be matched exists, matching the unmatched text information with the text information in the information base through a semantic model based on the depth network, and updating the successfully matched text information into the information base. The semantic model based on the depth network is the prior art. Text information which cannot be matched through the semantic model based on the depth network is sent to the manual processing module 80 and is confirmed by the manual.
If the identification judgment module 40 judges that the complete identification method is needed for the current arrival bill, all information on the arrival bill is required to be re-identified through the intelligent identification neural network model, and complete arrival acceptance information is obtained according to the method. Meanwhile, a new template is constructed according to the identification result, and a template library is updated. And finally, one-key uploading of the metering equipment arrival information is realized, and the business efficiency of the equipment arrival filing is greatly improved.
The beneficial effects of the step S2 are as follows: the method comprises the steps of respectively selecting a template identification method and a complete identification method for intelligent identification of the arrival bill image, constructing an intelligent identification neural network model, extracting image characteristic information of different sizes of arrival bill image sample information by a multi-size convolution layer, carrying out characteristic fusion, mapping the image characteristic of characters from an image space to a text space by a mapping layer by using a mapping function, obtaining arrival bill acceptance information with higher identification accuracy, improving the accuracy of arrival bill image identification, realizing one-key uploading of arrival bill information of metering equipment, greatly improving the service efficiency of arrival bill filing of equipment, and improving the standard, convenience and reliability of arrival bill identification.
In summary, the intelligent recognition system and the management method for the arrival order of the metering equipment are completed.
The technical scheme provided by the embodiment of the application at least has the following technical effects or advantages:
1. the initial recognition result obtained by OCR technology is further recognized, the matching degree of the initial recognition result and the template is calculated, the gradient of difference calculation is small, training is stable, the convergence of the difference calculation is high, matching achieves higher precision, and therefore a proper recognition method is selected for recognition of the arrival bill image, and the calculation complexity is reduced.
2. The method comprises the steps of respectively selecting a template identification method and a complete identification method for intelligent identification of a metering device arrival order image, constructing an intelligent identification neural network model, extracting image characteristic information of different sizes of arrival order image sample information by a multi-size convolution layer, carrying out characteristic fusion, mapping the image characteristic of characters from an image space to a text space by a mapping layer by using a mapping function, obtaining arrival order acceptance information with higher identification accuracy, improving accuracy of arrival order image identification, realizing one-key uploading of metering device arrival order information, greatly improving service efficiency of arrival order construction of the metering device, and improving standard, convenience and reliability of arrival order identification.
The method and the system can effectively solve the problems that the existing metering equipment bill recognition system and method are not divided and the uniform recognition method is adopted for different metering equipment bill templates, the recognition efficiency is low, the rejection rate and the false recognition rate are high, the calculation complexity is high, the recognition waiting time is long, the manual labor force cannot be liberated to the maximum extent, the recognition result is not reliable enough and intelligent enough, and through a series of effect investigation, the system or the method can finally realize one-key uploading of metering equipment arrival information through verification, the service efficiency of equipment arrival filing is greatly improved, and the standard, convenience and reliability of arrival bill recognition are improved.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Claims (10)

1. An intelligent recognition system for metering equipment to a manifest, which is characterized in that: the system comprises a manifest acquisition module (10), an image preprocessing module (20), an OCR (optical character recognition) module (30), a recognition judging module (40), a template recognition module (50), an intelligent recognition module (60), a similarity checking module (70), a manual processing module (80) and a database (90);
the automatic bill checking system comprises a manifest acquisition module (10), an image preprocessing module (20), a database (90), an OCR (optical character recognition) module (30), a recognition judging module (40), a template recognition module (50) and an intelligent recognition module (60), wherein the manifest acquisition module (10) is connected with the image preprocessing module (20), the image preprocessing module (20) is connected with the database (90), the OCR judging module (30) is connected with the recognition judging module (40), the recognition judging module (40) is respectively connected with the template recognition module (50) and the intelligent recognition module (60), the template recognition module (50) and the intelligent recognition module (60) are further connected with the intelligent recognition module (60), and the similarity checking module (70) is connected with the artificial processing module (80).
2. The intelligent metering device-to-manifest identification system of claim 1, wherein:
the arrival bill acquisition module (10) is used for acquiring an arrival bill image and sending the arrival bill image to the image preprocessing module (20) in a data transmission mode;
the image preprocessing module (20) is used for carrying out preliminary processing on the arrival bill image and respectively transmitting the preprocessed arrival bill image to the OCR module (30) and the database (90) in a data transmission mode;
the OCR module (30) is used for identifying texts in the arrival bill image so as to obtain a primary identification result of the target texts, and is also used for sending the primary identification result to the identification judging module (40) in a data transmission mode;
the identification judging module (40) is used for judging the type of the arrival bill identification of the metering equipment according to the primary identification result, wherein the arrival bill identification type comprises template identification and complete identification, and is also used for sending the identified primary identification result to the template identification module (50) or the intelligent identification module (60) in a data transmission mode according to the type of the identification result.
3. The intelligent metering device-to-manifest identification system of claim 2, wherein:
the template identification module (50) comprises a content matching unit 501 and a filling unit 502; the content matching unit 501 is configured to match the specific identified content that follows the attribute information in the initial identification result of the current arrival manifest with text information that is already in the information base of the database (90);
the filling unit 502 is configured to fill specific content successfully matched with text information in the information base to a corresponding position;
the template identification module (50) is also used for transmitting the arrival order image of the unrecognized area to the intelligent identification module (60) in a data transmission mode.
4. A metering device-to-manifest intelligent identification system as claimed in claim 3, wherein:
the intelligent recognition module (60) is used for constructing an intelligent recognition neural network model, and efficiently recognizing the arrival order image of the metering equipment to obtain the arrival acceptance information; the intelligent identification module (60) is also used for sending the check-in information to the similarity checking module (70) in a data transmission mode.
5. An intelligent recognition method for metering equipment to a manifest is characterized by comprising the following steps:
s1, acquiring an arrival bill image of metering equipment, acquiring a primary identification result of the arrival bill image by adopting an OCR technology, and judging the type of arrival bill identification according to attribute information and position information in the primary identification result;
s2, selecting a template identification method or a complete identification method to identify according to the type of the arrival bill identification;
the template identification method is used for matching specific contents of attribute information in the initial identification result of the manifest with text information existing in an information base, filling the successfully matched specific contents to corresponding positions, and further identifying the unfilled specific contents through an intelligent identification neural network model;
the complete recognition method is used for carrying out complete recognition on all information on the arrival bill image through the intelligent recognition neural network model, matching the recognized arrival bill acceptance information from the information base through a text similarity algorithm and a semantic model based on a depth network, and updating the information base to the bill image base and the information base.
6. The intelligent metering device-to-manifest identification method of claim 5, wherein: in step S1, the specific judgment method of the arrival bill identification type is as follows: determining attribute information and corresponding position information in each template, searching the attribute information from the primary identification result and obtaining the position information of the attribute information, so as to calculate the matching degree of the current primary identification result and the template; if the number of the attribute information with the matching degree meeting the preset threshold range reaches a preset percentage, the current primary identification result is indicated to be matched with the current template; otherwise, matching the current primary identification result with the next template until the corresponding template is matched or the matching calculation with all templates is finished;
the method for judging the type of the manifest identification according to the initial identification result comprises the following steps: if the typesetting of the goods receipt information in the initial recognition result is matched with the existing goods receipt template, template recognition is adopted, and if the typesetting of the goods receipt information in the initial recognition result is not matched with the existing goods receipt template, complete recognition is used.
7. The intelligent metering device-to-manifest identification method of claim 5, wherein: performing template recognition using a template recognition module (50); the template identification module (50) comprises a content matching unit (501) and a filling unit (502);
the content matching unit (501) matches the specific content which is identified and follows the attribute information in the initial identification result of the current manifest with the text information in the information base, and the filling unit (502) fills the specific content which is successfully matched with the text information in the information base to the corresponding position; and carrying out high-efficiency identification on the unfilled specific content through an intelligent identification neural network model, so as to perfect the goods acceptance information.
8. The intelligent metering device-to-manifest identification method of claim 5, wherein:
the intelligent recognition neural network model is used for complete recognition, and the realization process of the intelligent recognition neural network model is as follows: acquiring historical arrival bill images and arrival acceptance information in a corresponding information base thereof from an arrival bill image base as training samples of the intelligent recognition neural network, inputting the arrival bill images into the intelligent recognition neural network, outputting text information corresponding to the arrival bill images through deep learning of the neural network, performing error calculation with actual arrival bill acceptance information, correcting parameters in the intelligent recognition neural network according to the errors until the output precision of the intelligent recognition neural network reaches an expected effect, and thus finishing training of the intelligent recognition neural network;
the intelligent recognition neural network comprises an input layer, a multi-size convolution layer, a fusion layer, a clustering layer, a mapping layer and an output layer.
9. The intelligent metering device-to-manifest identification method of claim 8, wherein: the process of using the intelligent recognition neural network model for complete recognition is as follows: leading the metering equipment to the bill image sample information into an input layer, and transmitting the bill image sample information to a multi-size convolution layer by the input layer, wherein the multi-size convolution layer comprises N convolution layers with different sizes of convolution kernels, and extracting N corresponding image characteristic information under the different sizes of convolution kernels of the bill image sample information by the multi-size convolution layer; the multi-size convolution layers transmit the image features extracted by each convolution layer to the fusion layer;
the fusion layer fuses the multi-size image features, and the fusion layer transmits the fused features to the clustering layer;
the clustering layer clusters according to the distribution condition of the image features so as to divide different characters in the image, the clustering layer transmits the image features of the divided different characters to the mapping layer, the mapping layer maps the image features of the characters from the image space to the text space by using a mapping function, and the mapping layer transmits the obtained text information of the characters to the output layer;
the output layer outputs the text information recognized in the bill image, namely the check-in information.
10. The intelligent metering device-to-manifest identification method of claim 9, wherein: after the complete arrival acceptance information is obtained through complete identification, a new template is constructed according to the identification result, and a template library is updated.
CN202310609685.9A 2023-05-26 2023-05-26 Intelligent recognition system and method for metering equipment to bill Pending CN116844181A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310609685.9A CN116844181A (en) 2023-05-26 2023-05-26 Intelligent recognition system and method for metering equipment to bill

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310609685.9A CN116844181A (en) 2023-05-26 2023-05-26 Intelligent recognition system and method for metering equipment to bill

Publications (1)

Publication Number Publication Date
CN116844181A true CN116844181A (en) 2023-10-03

Family

ID=88166061

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310609685.9A Pending CN116844181A (en) 2023-05-26 2023-05-26 Intelligent recognition system and method for metering equipment to bill

Country Status (1)

Country Link
CN (1) CN116844181A (en)

Similar Documents

Publication Publication Date Title
CN109887153B (en) Finance and tax processing method and system
CN109325116B (en) Urban event automatic classification dispatching method and device based on deep learning
CN114862540B (en) Bill auditing system and method thereof
CN107918916A (en) Self-service Claims Resolution application processing method, device, computer equipment and storage medium
CN109377240B (en) Commercial tenant management method and device based on neural network, computer equipment and storage medium
US8676731B1 (en) Data extraction confidence attribute with transformations
CN110188336B (en) Method and device for generating reimbursement bill based on OA application form
CN107729576A (en) The management method and system of archives
CN107295039A (en) Data access treating method and apparatus
CN105787418A (en) Method and apparatus for intelligently identifying original certificates and automatically generating document of identification information
CN110569856A (en) sample labeling method and device, and damage category identification method and device
CN111444793A (en) Bill recognition method, equipment, storage medium and device based on OCR
CN112418812A (en) Distributed full-link automatic intelligent clearance system, method and storage medium
CN110647784A (en) Equipment asset management method and device based on deep learning
CN115017272A (en) Intelligent verification method and device based on registration data
CN114998920B (en) Supply chain financial file management method and system based on NLP semantic recognition
CN116409575A (en) Warehouse-in and warehouse-out management and control system of intelligent storage shelf
CN113066223A (en) Automatic invoice verification method and device
CN116863122A (en) Ammeter meter reading processing method, device, cloud end, system and medium
US11227174B1 (en) License plate recognition
CN116844181A (en) Intelligent recognition system and method for metering equipment to bill
CN116798061A (en) Bill auditing and identifying method, device, terminal and storage medium
CN115907673A (en) Supply chain system
CN116911852B (en) RPA user dynamic information monitoring method and system
CN111223230A (en) Invoice file authenticity identification method based on CRNN algorithm

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