Intelligent identification system and method for metering equipment to cargo manifest

Abstract

The application discloses an intelligent identification system and method for metering equipment to cargo list. The intelligent identification system comprises a cargo list acquisition module, an image preprocessing module, an OCR identification module, an identification judgment module, a template identification module, an intelligent identification module, a similarity checking module, a manual processing module and a database. The application solves the problems of low recognition efficiency, high rejection rate and high misrecognition rate caused by the unified recognition method for different document templates, and solves the problems of high calculation complexity, long recognition waiting time, inability to maximize the liberation of manual labor, unreliable and unintelligent recognition results and the like.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and more particularly to an intelligent recognition system for metering equipment delivery slips, as well as an intelligent recognition and management method for metering equipment delivery slips. Background Technology

[0002] Delivery acceptance includes registration, unpacking inspection, full-performance testing, and sampling inspection of metering equipment upon arrival. Metering equipment includes single-phase and three-phase energy meters, dedicated transformer terminals, instrument transformers, three-phase digital standard energy meters, high-precision time base sources, and three-phase energy meter calibration devices. Currently, managing and maintaining delivery acceptance information for metering equipment requires the ability to add, modify, delete, query, and export information. Delivery acceptance information mainly includes: equipment name, technical model, serial number, delivery information (contract number, contract name, supplier, serial number, date of manufacture, etc.), delivery details (asset number, serial number, date of delivery, etc.), and acceptance details (asset number, acceptance type, acceptance date, acceptance conclusion, inspector, acceptance attachments, etc.). The collection of delivery order data is mainly done manually. First, the original delivery orders for metering equipment are compiled, then scanned, and finally, the delivery acceptance information is manually entered and reviewed. Even with intelligent recognition methods, errors in character recognition or even failure to recognize characters often occur, resulting in very low efficiency in the data collection of delivery notes and requiring a lot of time and manpower.

[0003] On the other hand, due to the different issuers of the delivery notes for metering equipment, the names of the business fields in the documents vary greatly. Electronic equipment cannot intelligently identify business fields with different names but the same meaning, and the efficiency of the delivery filing business needs to be improved.

[0004] Chinese invention patent application CN202210322687.5, published on June 17, 2022, discloses a document recognition method, system, and computer-readable storage medium. The method includes: constructing a travel information model and a document category model; receiving a form-filling request and identifying the document category of the document in the request; if the document category is identified as travel, then the document is sequentially identified based on the trained travel information model and document category model, and the identification results are output to the corresponding document; if the document category is not travel, then the document is identified based on the document category model, and the identification results are output to the corresponding document. This solution can improve the accuracy and efficiency of form filling, and reduce the error rate and initial rejection rate of traditional manual reimbursement.

[0005] However, the above-mentioned technologies have at least the following technical problems: most existing document recognition systems and methods do not differentiate between recognition methods, and use a uniform recognition method for different document templates, resulting in low recognition efficiency, high rejection rate and false recognition rate; moreover, the computational complexity is high, the recognition waiting time is long, and it cannot maximize the liberation of manual labor, and the recognition results are not reliable or intelligent enough. Summary of the Invention

[0006] This invention proposes an intelligent identification system and method for metering equipment delivery notes. Its purpose is to: (1) solve the problem that the current different document templates all use the same identification method, which results in low identification efficiency, high rejection rate and high false recognition rate; (2) solve the problems of high computational complexity, long identification waiting time, inability to maximize the liberation of manual labor, and unreliable and unintelligent identification results.

[0007] The technical solution of this invention is as follows:

[0008] An intelligent recognition system for metering equipment delivery notes includes a delivery note acquisition module, an image preprocessing module, an OCR recognition module, a recognition judgment module, a template recognition module, an intelligent recognition module, a similarity screening module, a manual processing module, and a database.

[0009] The delivery note acquisition module is connected to the image preprocessing module, which is connected to the database and the OCR recognition module. The OCR recognition module is connected to the recognition judgment module, which is connected to the template recognition module and the intelligent recognition module respectively. The template recognition module is also connected to the intelligent recognition module. The template recognition module and the intelligent recognition module are connected to the similarity screening module, which is connected to the manual processing module.

[0010] As a further improvement to the intelligent recognition system for the delivery note of the aforementioned metering equipment:

[0011] The delivery note acquisition module is used to acquire the delivery note image and send the delivery note image to the image preprocessing module via data transmission.

[0012] The image preprocessing module is used to perform preliminary processing on the delivery note image, and then sends the preprocessed delivery note image to the OCR recognition module and the database respectively via data transmission.

[0013] The OCR recognition module is used to recognize the text in the delivery note image to obtain the initial recognition result of the target text, and is also used to send the initial recognition result to the recognition judgment module through data transmission.

[0014] The identification and judgment module is used to determine the type of metering equipment delivery note identification based on the initial identification result. The delivery note identification type includes template identification and complete identification. It is also used to send the initial identification result after identification to the template identification module or the intelligent identification module through data transmission according to its type.

[0015] As a further improvement to the intelligent recognition system for the delivery note of the aforementioned metering equipment:

[0016] The template recognition module includes a content matching unit 501 and a filling unit 502; the content matching unit 501 is used to match the specific content identified after the attribute information in the initial recognition result of the current delivery order with the existing text information in the database.

[0017] The filling unit 502 is used to fill the specific content that successfully matches the text information in the information database into the corresponding position;

[0018] The template recognition module is also used to send the arrival order image of the unrecognized area to the intelligent recognition module via data transmission.

[0019] As a further improvement to the intelligent recognition system for the delivery note of the aforementioned metering equipment:

[0020] The intelligent recognition module is used to construct an intelligent recognition neural network model to efficiently recognize the arrival order image of the metering equipment and obtain the arrival and acceptance information; the intelligent recognition module is also used to send the arrival and acceptance information to the similarity screening module through data transmission.

[0021] This invention also discloses an intelligent identification method for metering equipment delivery notes, comprising the following steps:

[0022] S1. Obtain the image of the metering equipment delivery note, use OCR technology to obtain the initial recognition result of the delivery note image, and determine the type of delivery note recognition based on the attribute information and location information in the initial recognition result;

[0023] S2. Select either template recognition method or complete recognition method for recognition based on the type of delivery note identification;

[0024] The template recognition method matches the specific content of the attribute information in the initial recognition result of the delivery note with the existing text information in the information database, and fills the successfully matched specific content into the corresponding position. The unfilled specific content is then further recognized by the intelligent recognition neural network model.

[0025] The complete recognition method uses an intelligent recognition neural network model to fully recognize all information on the delivery note image. The recognized delivery and acceptance information is then matched with the information database using a text similarity algorithm and a semantic model based on a deep network, and the delivery note image database and information database are updated accordingly.

[0026] As a further improvement to the intelligent identification method for the metering equipment delivery note: In step S1, the specific method for determining the delivery note identification type is as follows: Determine the attribute information and its corresponding location information in each template, find the attribute information and obtain its location information from the initial identification result, and then calculate the matching degree between the current initial identification result and the template; if the number of attribute information whose matching degree meets the preset threshold range reaches the preset percentage, it indicates that the current initial identification result matches the current template; otherwise, match the current initial identification result with the next template until the corresponding template is matched or the matching calculation with all templates is completed;

[0027] The method for determining the type of delivery note recognition based on the initial recognition results is as follows: if the layout of the delivery and acceptance information in the initial recognition results matches an existing delivery note template, then template recognition is used; if the layout of the delivery and acceptance information in the initial recognition results cannot match an existing delivery note template, then complete recognition is used.

[0028] As a further improvement to the intelligent identification method for the metering equipment delivery note: a template identification module is used for template identification; the template identification module includes a content matching unit 501 and a filling unit;

[0029] The content matching unit matches the attribute information following the initial identification result of the current delivery note with the existing text information in the information database. The filling unit fills the specific content that successfully matches the text information in the information database into the corresponding position. The unfilled specific content is efficiently identified by the intelligent recognition neural network model, thereby improving the delivery and acceptance information.

[0030] As a further improvement to the intelligent identification method for the aforementioned metering equipment delivery note:

[0031] Complete recognition is performed using an intelligent recognition neural network model. The implementation process of the intelligent recognition neural network model is as follows: Historical delivery note images and corresponding delivery acceptance information from the delivery note image library are obtained as training samples for the intelligent recognition neural network. The delivery note images are input into the intelligent recognition neural network. After deep learning, the neural network outputs the text information corresponding to the delivery note images. Then, the error is calculated with the actual delivery acceptance information, and the parameters in the intelligent recognition neural network are corrected according to the error until the output accuracy of the intelligent recognition neural network reaches the expected effect, thereby completing the training of the intelligent recognition neural network.

[0032] The intelligent recognition neural network includes an input layer, a multi-size convolutional layer, a fusion layer, a clustering layer, a mapping layer, and an output layer.

[0033] As a further improvement to the intelligent recognition method for the metering equipment delivery note, the process of complete recognition using an intelligent recognition neural network model is as follows: Image sample information of the metering equipment delivery note is imported into the input layer; the input layer transmits the delivery note image sample information to a multi-size convolutional layer, which contains N convolutional layers with convolutional kernels of different sizes; the multi-size convolutional layer extracts N types of image feature information corresponding to different convolutional kernel sizes from the delivery note image sample information; the multi-size convolutional layer transmits the image features extracted by each convolutional layer to the fusion layer.

[0034] The fusion layer fuses features from images of multiple sizes, and then passes the fused features to the clustering layer.

[0035] The clustering layer clusters images based on the distribution of image features, thereby separating different texts in the image. The clustering layer then transmits the image features of the separated texts to the mapping layer. The mapping layer uses a mapping function to map the image features of the texts from the image space to the text space. The mapping layer then transmits the obtained text information to the output layer.

[0036] The output layer outputs the text information recognized from the delivery note image, i.e., the delivery and acceptance information.

[0037] As a further improvement to the intelligent identification method for the metering equipment delivery note: after obtaining complete delivery and acceptance information through complete identification, a new template is constructed based on the identification results, and the template library is updated.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) This invention further recognizes the initial recognition results obtained by OCR technology for the metering equipment delivery note image, calculates the matching degree between the initial recognition result and the template, and the gradient of the difference calculation is small, which makes the training more stable. Moreover, the convergence of the difference calculation is high, which makes the matching more accurate. Thus, a more suitable recognition method is selected for the recognition of the metering equipment delivery note image, which improves the recognition efficiency, reduces the rejection rate and false recognition rate, improves the reliability of the recognition results, reduces the computational complexity, and shortens the recognition waiting time.

[0040] (2) The arrival order images of metering equipment are intelligently identified by selecting template recognition method and complete recognition method respectively. An intelligent recognition neural network model is constructed. The image feature information of different sizes of the arrival order image sample information is extracted by multi-size convolutional layer and feature fusion is performed. The mapping layer uses the mapping function to map the image features of the text from the image space to the text space, so as to obtain the arrival acceptance information of metering equipment with higher recognition accuracy and improve the accuracy of metering equipment arrival order image recognition.

[0041] (3) It can realize one-click uploading of metering equipment arrival information, greatly improve the business efficiency of equipment arrival filing, and improve the standardization, convenience and reliability of arrival order recognition. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the identification system of the present invention;

[0043] Figure 2 This is a flowchart illustrating the identification and management method of the present invention. Detailed Implementation

[0044] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings:

[0045] This embodiment provides an intelligent identification system and method for metering equipment delivery notes. The overall concept is as follows:

[0046] The initial recognition results obtained by OCR technology on the metering equipment arrival note images are further recognized. The matching degree between the initial recognition results and the template is calculated. The gradient of the difference calculation is small, which makes the training more stable. The high convergence of the difference calculation results in higher matching accuracy, thus selecting a suitable recognition method for the arrival note images and reducing the computational complexity. The metering equipment arrival note images are intelligently recognized by selecting template recognition method and complete recognition method respectively. An intelligent recognition neural network model is constructed. Multi-size convolutional layers extract image feature information of different sizes from the arrival note image sample information and perform feature fusion. The mapping layer uses a mapping function to map the image features of the text from the image space to the text space, resulting in metering equipment arrival and acceptance information with higher recognition accuracy. This improves the accuracy of metering equipment arrival note image recognition, realizes one-click uploading of metering equipment arrival information, greatly improves the business efficiency of equipment arrival filing, and improves the standardization, convenience and reliability of arrival note recognition.

[0047] To better understand the above technical solution, the following will provide a detailed explanation of the technical solution in conjunction with the accompanying drawings.

[0048] See attached document Figure 1An intelligent recognition system for metering equipment delivery notes includes the following components: a delivery note acquisition module 10, an image preprocessing module 20, an OCR recognition module 30, a recognition judgment module 40, a template recognition module 50, an intelligent recognition module 60, a similarity screening module 70, a manual processing module 80, and a database 90.

[0049] The delivery note acquisition module 10 is used to acquire the delivery note image and send the delivery note image to the image preprocessing module 20 via data transmission.

[0050] The image preprocessing module 20 is used to perform preliminary processing on the delivery note image, including operations such as image grayscale conversion, binarization, noise reduction, and smoothing. The image preprocessing module 20 sends the preprocessed delivery note image to the OCR recognition module 30 and the database 90 via data transmission.

[0051] The OCR recognition module 30 is used to recognize the text in the delivery note image using OCR (Optical Character Recognition) technology to obtain the initial recognition result of the target text. The OCR recognition module 30 sends the initial recognition result to the recognition judgment module 40 through data transmission.

[0052] The identification and judgment module 40 is used to determine the type of delivery note identification based on the initial identification result. The delivery note identification type includes template identification and complete identification. The identification and judgment module 40 sends the initial identification result after identification to the template identification module 50 or the intelligent identification module 60 through data transmission according to its type.

[0053] The template recognition module 50 includes a content matching unit 501 and a filling unit 502. The content matching unit 501 is used to match the specific content identified after the attribute information in the initial recognition result of the current delivery note with the existing text information in the information database 90. The filling unit 502 is used to fill the specific content that successfully matches the text information in the information database into the corresponding position. The template recognition module 50 also sends the delivery note image of the unrecognized area to the intelligent recognition module 60 via data transmission.

[0054] The intelligent recognition module 60 is used to construct an intelligent recognition neural network model to efficiently recognize the delivery note image and obtain delivery acceptance information. The intelligent recognition module 60 sends the delivery acceptance information to the similarity screening module 70 via data transmission.

[0055] The similarity screening module 70 is used to match the arrival and acceptance information identified by the intelligent recognition neural network model with the information database 90 using a text similarity algorithm. If there are unmatchable characters, the unmatched characters are matched with the text in the information database using a semantic model based on a deep network. The successfully matched text information is updated in the information database, and the unmatched text information is sent to the manual processing module 80 via data transmission.

[0056] The manual processing module 80 is used to display unmatched text information to the user for manual confirmation, and the manual processing module 80 updates the confirmation result to the database 90.

[0057] The database 90 includes a delivery note image library and an information library. The delivery note image library stores pre-processed delivery note images in the database 90. The information library stores all text information from historical delivery notes.

[0058] See attached document Figure 2 The intelligent identification method for metering equipment delivery notes described in this application includes the following steps:

[0059] S1. Obtain the delivery note image, use OCR technology to obtain the initial recognition result of the delivery note image, and determine the type of delivery note recognition based on the attribute information and location information in the initial recognition result.

[0060] Upon receiving the delivery note for the metering equipment, the delivery note intelligent recognition system needs to intelligently recognize the delivery note, identify the delivery and acceptance information recorded on the delivery note, and manage and maintain the identified delivery and acceptance information. It supports adding, modifying, deleting, querying and exporting information, and supports quick import via mobile browser by taking photos or images.

[0061] The intelligent recognition system for metering equipment delivery notes acquires images of the delivery notes.

[0062] In some embodiments of this application, the delivery note image can be obtained by the user through taking a photo via a mobile browser or by importing an image, and the delivery note image is automatically transmitted to the delivery note acquisition module 10.

[0063] In an optional implementation, the delivery note image for the metering equipment can be obtained by the user scanning the delivery note with a scanner, and the delivery note image is automatically transmitted to the delivery note acquisition module 10.

[0064] It is understood that there are multiple ways to obtain images of the metering equipment delivery note, not limited to the above-mentioned methods of taking pictures, importing, and scanning. This invention does not limit these methods.

[0065] The image preprocessing module 20 performs preliminary processing on the delivery note image, including operations such as image grayscale conversion, binarization, noise reduction, and smoothing. The preprocessing methods all adopt existing technologies, which are relatively mature.

[0066] After image preprocessing, the OCR recognition module 30 uses OCR (Optical Character Recognition) technology to recognize the text in the metering equipment delivery note image to obtain the initial recognition result of the target text. However, due to the high rejection rate and false recognition rate of existing OCR recognition software, only fuzzy results of the delivery and acceptance information can be obtained.

[0067] The identification and judgment module 40 determines the type of metering equipment delivery note identification based on the initial identification result. The delivery note identification type includes template identification and complete identification. If an existing delivery note template is matched with the layout of the delivery and acceptance information in the initial identification result, template identification is used. If an existing delivery note template cannot be matched with the layout of the delivery and acceptance information in the initial identification result, complete identification is used.

[0068] The specific method for determining the type of delivery note is as follows:

[0069] First, determine the attribute information in the goods receipt and acceptance information, including:

[0070] (1) Arrival information: Contract number, contract name, supplier, equipment category, equipment subcategory, equipment name, quantity received, arrival date, recipient, asset number, equipment code, model and specifications, serial number, and date of manufacture;

[0071] (2) Arrival details: asset number, equipment name, technical model, serial number, arrival date, whether traceability is required, supplier, contract number;

[0072] (3) Acceptance details: asset number, acceptance type (sent to calibration, self-calibration, sent for inspection, self-inspection, mandatory verification, other verification), acceptance date, acceptance conclusion, inspector, traceability results of measurement values, and acceptance attachments.

[0073] The attribute information and its corresponding location information in each template are determined. The attribute information and its location information are retrieved from the initial identification results, and the matching degree between the current initial identification result and the template is calculated. The specific calculation method is as follows:

[0074]

[0075] Among them, md i p represents the degree of matching between the initial recognition result and the position of the i-th attribute information in the template. dem p represents the attribute location information in the template. disThis indicates the location information of the corresponding attribute in the initial identification result. If the number of attribute information with a matching degree within the preset threshold range reaches the preset percentage, it indicates that the current initial identification result matches the current template; otherwise, the current initial identification result is matched with the next template until the corresponding template is matched or the matching calculation with all templates is completed, thereby determining whether the current metering equipment delivery note needs to undergo template identification or complete identification.

[0076] The beneficial effects of step S1 are as follows: the initial recognition results obtained by using OCR technology on the metering equipment delivery note image are further recognized, the matching degree between the initial recognition results and the template is calculated, the gradient of the difference calculation is small, which makes the training more stable, and the convergence of the difference calculation is high, which makes the matching more accurate. Thus, a suitable recognition method is selected for the recognition of the metering equipment delivery note image, and the computational complexity is reduced.

[0077] S2. The template recognition method matches the specific content of the attribute information in the initial recognition result of the delivery note with the existing text information in the information database. Successfully matched content is filled into the corresponding positions, while unfilled content is efficiently recognized using an intelligent recognition neural network model. The complete recognition method uses an intelligent recognition neural network model to fully recognize all information on the delivery note image. The recognized delivery and acceptance information is then matched against the information database using a text similarity algorithm and a deep network-based semantic model, updating both the delivery note image database and the information database.

[0078] The intelligent delivery note recognition system includes a database 90, which comprises a delivery note image library and an information library. The delivery note image library stores pre-processed delivery note images. The information library stores all text information from historical delivery notes.

[0079] If the recognition and judgment module 40 determines that the current delivery order needs to use the template recognition method, the specific content matching unit 501 in the template recognition module 50 will match the specific content identified after the attribute information in the initial recognition result of the current delivery order with the existing text information in the information database, and the filling unit 502 will fill the specific content that successfully matches the text information in the information database into the corresponding position.

[0080] The missing information is efficiently identified by the intelligent recognition neural network model in the intelligent recognition module 60, thereby improving the arrival and acceptance information.

[0081] The specific implementation process of the intelligent recognition neural network model is as follows:

[0082] Historical delivery note images and corresponding delivery acceptance information from the information database are obtained from the delivery note image library and used as training samples for the intelligent recognition neural network. The delivery note images are input into the intelligent recognition neural network, and after deep learning, the neural network outputs the text information corresponding to the delivery note images. The error is then calculated by comparing the text information with the actual delivery acceptance information, and the parameters in the intelligent recognition neural network are corrected according to the error until the output accuracy of the intelligent recognition neural network reaches the expected effect, thus completing the training of the intelligent recognition neural network.

[0083] The intelligent recognition neural network includes an input layer, a multi-size convolutional layer, a fusion layer, a clustering layer, a mapping layer, and an output layer.

[0084] The arrival order image sample information is imported into the input layer. The input layer then transmits the arrival order image sample information to a multi-size convolutional layer. This multi-size convolutional layer contains N convolutional layers with convolutional kernels of different sizes. The multi-size convolutional layer extracts N types of image feature information corresponding to the different convolutional kernel sizes of the arrival order image sample information. The activation function of the multi-size convolutional layer is:

[0085] f=max(Z,m*(1-expβ))*R(σ*Z)+δ;

[0086] Where f is the activation function of the multi-size convolutional layer, Z represents the image sample information of the delivery note after convolution, m represents the number of image channels, β represents the convolution kernel parameter, R(·) is the image feature extractor, σ represents the convolution kernel size, and δ represents the adjustment factor. The multi-size convolutional layer passes the image features extracted by each convolutional layer to the fusion layer. The output of the multi-size convolutional layer is:

[0087]

[0088] Among them, OP j This represents the output of the j-th convolutional layer, j∈[1,N], and ω represents the weights of the bill of lading image sample information. denoted by , z represents the image sample information of the waybill, and b represents the deviation.

[0089] The fusion layer fuses features from multiple image sizes, and the fusion calculation formula is as follows:

[0090]

[0091] Where F represents the fused feature, and MLP represents the multilayer perceptron operation. This indicates a join operation. The fusion layer passes the fused features to the clustering layer.

[0092] The clustering layer performs clustering based on the distribution of image features, thereby separating different characters in the image. The clustering method employs existing techniques, such as k-clustering. The clustering layer then transmits the image features of the separated characters to the mapping layer.

[0093] The mapping layer uses a mapping function to map the image features of text from the image space to the text space. The mapping method is as follows:

[0094]

[0095] Where MF represents the mapping function of the image features of any character, Δd x Δd represents the horizontal coordinate difference of the current text image features. y G(x, y) represents the vertical coordinate difference of the current text image features, where x represents the horizontal coordinate and y represents the vertical coordinate. max and x min These represent the maximum and minimum values ​​of the current text image features in the horizontal direction, y and y', respectively. max and y min These represent the maximum and minimum values ​​of the current text image features in the vertical direction, respectively. The mapping layer transmits the obtained text information to the output layer. The output layer outputs the text information recognized from the delivery note image, i.e., the delivery and acceptance information.

[0096] The similarity screening module 70 uses a text similarity algorithm to match the arrival and acceptance information identified by the intelligent recognition neural network model with the information database. The text similarity algorithm can be cosine similarity, etc. If there are unmatched text information, a deep network-based semantic model is used to match the unmatched text information with the text information in the information database, and the successfully matched text information is updated in the information database. The deep network-based semantic model is existing technology. Text information that still cannot be matched by the deep network-based semantic model needs to be sent to the manual processing module 80 for manual confirmation.

[0097] If the identification and judgment module 40 determines that the current delivery note requires a complete identification method, then all information on the delivery note needs to be re-identified using the intelligent identification neural network model, and the complete delivery and acceptance information is obtained according to the above method. Simultaneously, a new template is constructed based on the identification results, and the template library is updated. Ultimately, this enables one-click uploading of metering equipment delivery information, significantly improving the efficiency of equipment delivery documentation.

[0098] The beneficial effects of step S2 are as follows: the delivery note images are intelligently recognized by selecting template recognition method and complete recognition method respectively, and an intelligent recognition neural network model is constructed. The image feature information of different sizes of the delivery note image sample information is extracted by multi-size convolutional layer and feature fusion is performed. The mapping layer uses the mapping function to map the image features of the text from the image space to the text space, thereby obtaining delivery acceptance information with higher recognition accuracy, improving the accuracy of delivery note image recognition, realizing one-click uploading of metering equipment delivery information, greatly improving the business efficiency of equipment delivery filing, and improving the standardization, convenience and reliability of delivery note recognition.

[0099] In summary, the intelligent identification system and management method for metering equipment delivery notes described in this application have been completed.

[0100] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0101] 1. Further recognition is performed on the initial recognition results obtained by OCR technology on the arrival note image of the metering equipment. The matching degree between the initial recognition result and the template is calculated. The gradient of the difference calculation is small, which makes the training more stable. The high convergence of the difference calculation makes the matching more accurate. Thus, a suitable recognition method is selected for the recognition of the arrival note image, and the computational complexity is reduced.

[0102] 2. The metering equipment arrival note images are intelligently recognized using both template recognition and complete recognition methods. An intelligent recognition neural network model is constructed, which extracts image feature information of different sizes from the arrival note image samples through multi-size convolutional layers and performs feature fusion. The mapping layer uses a mapping function to map the image features of the text from the image space to the text space, resulting in more accurate arrival and acceptance information. This improves the accuracy of arrival note image recognition, enables one-click uploading of metering equipment arrival information, significantly improves the business efficiency of equipment arrival filing, and enhances the standardization, convenience, and reliability of arrival note recognition.

[0103] This embodiment effectively addresses the problems of existing metering equipment document recognition systems and methods, which mostly lack differentiation in recognition methods and use a uniform recognition method for different metering equipment document templates. This results in low recognition efficiency, high rejection and false recognition rates, high computational complexity, long recognition waiting time, failure to maximize the reduction of manual labor, and unreliable and unintelligent recognition results. Furthermore, the aforementioned system or method has undergone a series of effectiveness surveys and verifications, ultimately enabling one-click uploading of metering equipment arrival information, significantly improving the business efficiency of equipment arrival filing, and enhancing the standardization, convenience, and reliability of arrival document recognition.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

Claims

1. An intelligent identification system for metering equipment delivery notes, characterized in that: It includes a delivery note acquisition module (10), an image preprocessing module (20), an OCR recognition module (30), a recognition judgment module (40), a template recognition module (50), an intelligent recognition module (60), a similarity screening module (70), a manual processing module (80), and a database (90); The delivery note acquisition module (10) is connected to the image preprocessing module (20), the image preprocessing module (20) is connected to the database (90) and also to the OCR recognition module (30), the OCR recognition module (30) is connected to the recognition judgment module (40), the recognition judgment module (40) is connected to the template recognition module (50) and the intelligent recognition module (60) respectively, the template recognition module (50) is also connected to the intelligent recognition module (60), the template recognition module (50) and the intelligent recognition module (60), the intelligent recognition module (60) is connected to the similarity screening module (70), and the similarity screening module (70) is connected to the manual processing module (80); The OCR recognition module (30) is used to recognize the text in the delivery note image to obtain the initial recognition result of the target text, and is also used to send the initial recognition result to the recognition judgment module (40) through data transmission. The identification and judgment module (40) is used to determine the type of metering equipment delivery note identification based on the initial identification result. The delivery note identification type includes template identification and complete identification. It is also used to send the initial identification result after identification to the template identification module (50) or the intelligent identification module (60) through data transmission according to its type.

2. The intelligent identification system for metering equipment delivery notes as described in claim 1, characterized in that: The delivery note acquisition module (10) is used to acquire the delivery note image and send the delivery note image to the image preprocessing module (20) via data transmission. The image preprocessing module (20) is used to perform preliminary processing on the delivery note image and send the preprocessed delivery note image to the OCR recognition module (30) and the database (90) respectively via data transmission.

3. The intelligent identification system for metering equipment delivery notes as described in claim 2, characterized in that: The template recognition module (50) includes a content matching unit (501) and a filling unit (502); the content matching unit (501) is used to match the specific content identified after the attribute information in the initial recognition result of the current delivery order with the existing text information in the information base of the database (90); The filling unit (502) is used to fill the corresponding position with the specific content that successfully matches the text information in the information database; The template recognition module (50) is also used to send the delivery note image of the unrecognized area to the intelligent recognition module (60) via data transmission.

4. The intelligent identification system for metering equipment delivery notes as described in claim 3, characterized in that: The intelligent recognition module (60) is used to construct an intelligent recognition neural network model to efficiently recognize the image of the metering equipment delivery note and obtain the delivery acceptance information; the intelligent recognition module (60) is also used to send the delivery acceptance information to the similarity screening module (70) through data transmission. The intelligent recognition module (60) uses an intelligent recognition neural network model for complete recognition. The implementation process of the intelligent recognition neural network model is as follows: historical delivery order images and corresponding delivery acceptance information in the information database are obtained from the delivery order image library as training samples for the intelligent recognition neural network. The delivery order images are input into the intelligent recognition neural network. After deep learning of the neural network, the text information corresponding to the delivery order images is output. Then, the error is calculated with the actual delivery acceptance information. The parameters in the intelligent recognition neural network are corrected according to the error until the output accuracy of the intelligent recognition neural network reaches the expected effect, thereby completing the training of the intelligent recognition neural network. The intelligent recognition neural network includes an input layer, a multi-size convolutional layer, a fusion layer, a clustering layer, a mapping layer, and an output layer; The process of complete recognition using an intelligent recognition neural network model is as follows: Image sample information of the metering equipment arrival note is imported into the input layer. The input layer then transmits the arrival note image sample information to a multi-size convolutional layer. This multi-size convolutional layer contains N convolutional layers with convolutional kernels of different sizes. The multi-size convolutional layer extracts N types of image feature information corresponding to the convolutional kernels of different sizes from the arrival note image sample information. The multi-size convolutional layer then transmits the image features extracted by each convolutional layer to the fusion layer. The fusion layer fuses features from images of multiple sizes, and then passes the fused features to the clustering layer. The clustering layer clusters images based on the distribution of image features, thereby separating different texts in the image. The clustering layer then transmits the image features of the separated texts to the mapping layer. The mapping layer uses a mapping function to map the image features of the texts from the image space to the text space. The mapping layer then transmits the obtained text information to the output layer. The output layer outputs the text information recognized from the delivery note image, i.e., the delivery and acceptance information.

5. A method for intelligent identification of delivery notes for metering equipment, characterized in that... Includes the following steps: S1. Obtain the image of the metering equipment delivery note, use OCR technology to obtain the initial recognition result of the delivery note image, and determine the type of delivery note recognition based on the attribute information and location information in the initial recognition result; S2. Select either template recognition method or complete recognition method for recognition based on the type of delivery note identification; The template recognition method matches the specific content of the attribute information in the initial recognition result of the delivery note with the existing text information in the information database, and fills the successfully matched specific content into the corresponding position. The unfilled specific content is then further recognized by the intelligent recognition neural network model. The complete recognition method uses an intelligent recognition neural network model to fully recognize all information on the delivery note image. The recognized delivery and acceptance information is then matched with the information database using a text similarity algorithm and a semantic model based on a deep network, and the delivery note image database and information database are updated accordingly. Complete recognition is performed using an intelligent recognition neural network model. The implementation process of the intelligent recognition neural network model is as follows: Historical delivery note images and corresponding delivery acceptance information from the delivery note image library are obtained as training samples for the intelligent recognition neural network. The delivery note images are input into the intelligent recognition neural network. After deep learning, the neural network outputs the text information corresponding to the delivery note images. Then, the error is calculated with the actual delivery acceptance information, and the parameters in the intelligent recognition neural network are corrected according to the error until the output accuracy of the intelligent recognition neural network reaches the expected effect, thereby completing the training of the intelligent recognition neural network. The intelligent recognition neural network includes an input layer, a multi-size convolutional layer, a fusion layer, a clustering layer, a mapping layer, and an output layer; The process of complete recognition using an intelligent recognition neural network model is as follows: Image sample information of the metering equipment arrival note is imported into the input layer. The input layer then transmits the arrival note image sample information to a multi-size convolutional layer. This multi-size convolutional layer contains N convolutional layers with convolutional kernels of different sizes. The multi-size convolutional layer extracts N types of image feature information corresponding to the convolutional kernels of different sizes from the arrival note image sample information. The multi-size convolutional layer then transmits the image features extracted by each convolutional layer to the fusion layer. The fusion layer fuses features from images of multiple sizes, and then passes the fused features to the clustering layer. The clustering layer clusters images based on the distribution of image features, thereby separating different texts in the image. The clustering layer then transmits the image features of the separated texts to the mapping layer. The mapping layer uses a mapping function to map the image features of the texts from the image space to the text space. The mapping layer then transmits the obtained text information to the output layer. The output layer outputs the text information recognized from the delivery note image, i.e., the delivery and acceptance information.

6. The intelligent identification method for metering equipment delivery notes as described in claim 5, characterized in that: In step S1, the specific method for determining the type of delivery note identification is as follows: determine the attribute information and its corresponding location information in each template, find the attribute information and obtain its location information from the initial identification result, and then calculate the matching degree between the current initial identification result and the template. If the number of attribute information that matches the preset threshold range reaches the preset percentage, it indicates that the current initial identification result matches the current template. Otherwise, the current initial identification result is matched with the next template until the corresponding template is matched or the matching calculation with all templates is completed; The method for determining the type of delivery note recognition based on the initial recognition results is as follows: if the layout of the delivery and acceptance information in the initial recognition results matches an existing delivery note template, then template recognition is used; if the layout of the delivery and acceptance information in the initial recognition results cannot match an existing delivery note template, then complete recognition is used.

7. The intelligent identification method for metering equipment delivery notes as described in claim 5, characterized in that: Template recognition is performed using a template recognition module (50); the template recognition module (50) includes a content matching unit (501) and a filling unit (502); The content matching unit (501) matches the attribute information following the initial identification result of the current delivery order with the existing text information in the information database. The filling unit (502) fills the specific content that is successfully matched with the text information in the information database into the corresponding position. The unfilled specific content is efficiently identified by the intelligent recognition neural network model, thereby improving the delivery acceptance information.

8. The intelligent identification method for metering equipment delivery notes as described in claim 5, characterized in that: After obtaining complete arrival and acceptance information through complete identification, a new template is built based on the identification results, and the template library is updated.

Images