CN113743336B

CN113743336B - Invoice information identification method and device based on deep learning and computer equipment

Info

Publication number: CN113743336B
Application number: CN202111052157.5A
Authority: CN
Inventors: 余宪
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2021-09-08
Filing date: 2021-09-08
Publication date: 2023-06-20
Anticipated expiration: 2041-09-08
Also published as: CN113743336A

Abstract

The application relates to the field of artificial intelligence and is simultaneously applicable to the field of digital medical treatment, and provides an invoice information identification method, device, computer equipment and storage medium based on deep learning, wherein the method comprises the following steps: detecting the medical invoice image to obtain a first detection frame corresponding to the target field; obtaining a second detection frame based on the first detection frame; performing matching processing on the second detection frames in each block to obtain a first matching frame; filtering the first matching frame based on the general recognition result of the item frame to obtain a second matching frame; filtering and matching all detection frames contained in the second matching frame based on the text filling rule to obtain a third matching frame; and calculating the second text information in the third matching box to generate a first value of all the self-charge amount fields and a second value of part of the self-charge amount fields. The intelligent medical invoice information extraction method and device can improve the intelligence of medical invoice information extraction. The present application may also be applied to the blockchain domain, where the first value may be stored on the blockchain.

Description

Invoice information identification method and device based on deep learning and computer equipment

Technical Field

The application relates to the technical field of artificial intelligence, in particular to an invoice information identification method, device and computer equipment based on deep learning.

Background

With the development of computer vision technology, OCR (Optical Character Recognition ) technology has been widely applied to scenes such as medical insurance. The medical invoice recognition system usually adopts a mode of mainly machine input and manually input as auxiliary. When the key information of the medical invoice is extracted, the existing medical invoice identification scheme follows what you see is what you get, namely key fields such as name, number, date, amount and the like on the medical invoice are automatically identified and structured and output through a deep learning algorithm, so that the quick completion of claim settlement operation is supported. However, in general, when the claim settlement operation of the medical invoice is performed, some other important key information needs to be obtained from the medical invoice, such as the self-payment amount information and part of the self-payment amount information of the charge invoice of the Tianjin medical clinic. However, the existing medical invoice identification scheme cannot accurately extract the self-fee amount information and part of the self-fee amount information from the medical invoice, and the content identification of the medical invoice is not intelligent. And by manually carrying out information accounting according to various related content data in the item/medicine details of the medical invoice, a large amount of manpower and material resources are required to be consumed, so that the processing efficiency of generating the self-charge amount information and part of the self-charge amount information is low.

Disclosure of Invention

The main purpose of the application is to provide an invoice information identification method, device, computer equipment and storage medium based on deep learning, which aims to solve the technical problems that the existing medical invoice identification scheme can not accurately extract self-charge amount information and partial self-charge amount information from medical invoices, the content identification of the medical invoice is lack of intelligence, and a large amount of manpower and material resources are required to be consumed by carrying out information accounting according to various related content data in the item/medicine details of the medical invoice by means of manpower, so that the processing efficiency of generating the self-charge amount information and partial self-charge amount information is low.

The application provides an invoice information identification method based on deep learning, which comprises the following steps:

detecting the medical invoice image based on a preset detection model to obtain a first detection frame corresponding to the target field; the target field comprises a negative increasing field and an amount field, and the first detection frame comprises a first negative increasing detection frame corresponding to the negative increasing field and a first amount detection frame corresponding to the amount field;

processing the detection frames according to a first preset rule based on the spatial position information of the first detection frames to obtain second detection frames after block division;

Matching the second increasing and decreasing detection frames contained in the second detection frames in each block with the second amount detection frames according to a second preset rule to obtain corresponding first matching frames; each first matching frame comprises a pair of second increasing negative detection frames and second amount detection frames which have a matching relationship;

obtaining a general recognition result of a preset project frame in the medical invoice image, filtering the first matching frame based on the general recognition result to obtain a corresponding second matching frame, and writing corresponding first text information in the second matching frame based on the general recognition result;

filtering all detection frames contained in the second matching frame based on a preset text filling rule to obtain a corresponding third detection frame, and matching the third detection frame according to the second preset rule to obtain a corresponding third matching frame;

calculating the second text information in the third matching frame according to a preset numerical calculation rule corresponding to all the self-charge amount fields and part of the self-charge amount fields, and generating a first numerical value of all the self-charge amount fields and a second numerical value of part of the self-charge amount fields; the total self-fee amount field is a field corresponding to the sum of amounts spent by all the self-fee items contained in the medical invoice image, and the partial self-fee amount field is a field corresponding to the sum of self-fee amounts in the amounts spent by all the partial self-fee items contained in the medical invoice image.

Optionally, the step of processing the detection frame according to a first preset rule based on the spatial position information of the first detection frame to obtain a second detection frame after block division includes:

based on the coordinate information of the first detection frame, filtering the first detection frame through a non-maximum suppression algorithm to obtain a corresponding fourth detection frame;

performing block division processing on the fourth detection frame based on the abscissa of the central point of the fourth detection frame to obtain a fifth detection frame after block division;

performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the fifth detection frame to obtain a corresponding sixth detection frame;

performing preset row correction processing on the sixth detection frame based on second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame, and performing filtering processing on the seventh detection frame through the non-maximum suppression algorithm to obtain a corresponding eighth detection frame;

and taking the eighth detection frame as the second detection frame.

Optionally, the step of filtering the first detection frame by a non-maximum suppression algorithm based on the coordinate information of the first detection frame to obtain a corresponding fourth detection frame includes:

Acquiring the confidence coefficient of each first detection frame and the coordinate value of each first detection frame, and acquiring the coordinate position information of the item frame;

filtering detection frames with confidence coefficient of 0 from all the first detection frames and detection frames with filtering coordinate values not in the range of the coordinate position information to obtain corresponding ninth detection frames;

the confidence degrees of all the ninth detection frames are ordered in a descending order, and corresponding first ordering results are obtained;

screening a tenth detection frame with highest confidence from the first sequencing result;

calculating the overlapping area between the tenth detection frame and other detection frames in the ninth detection frame;

and acquiring a preset area threshold, deleting the detection frames with the overlapping area larger than the area threshold in the ninth detection frames, sequencing the rest detection frames in descending order of confidence, deleting the detection frames with the overlapping area larger than the area threshold again until the number of the detection frames with the overlapping area larger than the area threshold is 0, and taking all the rest detection frames as the fourth detection frames.

Optionally, the step of performing block division processing on the fourth detection frame based on the abscissa of the center point of the fourth detection frame to obtain a fifth detection frame after block division includes:

Calculating the average value of the abscissa of the central points of all the first specified detection frames according to a first preset formula; the first appointed detection frame is a third negative increasing detection frame corresponding to the negative increasing field or a third amount detection frame corresponding to the amount field in the fourth detection frame;

calculating standard deviations corresponding to the abscissa of the central points of all the first specified detection frames according to a second preset formula;

acquiring a first preset threshold value, a second preset threshold value and a width value of the item frame;

determining a block division result of each first specified detection frame based on the mean value, the standard deviation, the first preset threshold value, the second preset threshold value and the width value;

and dividing each first appointed detection frame based on the block division result to obtain a fifth detection frame after block division.

Optionally, the step of performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the third detection frame to obtain a corresponding sixth detection frame includes:

acquiring all second specified detection frames in the first block; the first block is any block after block division, and the second specified detection frame is a fourth negative-increasing detection frame corresponding to the negative-increasing field or a fourth amount detection frame corresponding to the amount field, which is included in a fifth detection frame in the first block;

Sequencing all the second specified detection frames from large to small according to the numerical value of the ordinate of the central point to obtain a corresponding second sequencing result;

acquiring the abscissa of the central point, the average width value of the detection frames and the average height value of the first detection frames corresponding to all the second specified detection frames;

acquiring the number of all the second specified detection frames;

according to the ordering of the second ordering result, calculating the interval of the central abscissa between two adjacent second specified detection frames;

if the number is equal to 2, screening all the second specified detection frames according to a first screening rule corresponding to the interval of the central abscissa between the second specified detection frames to obtain a sixth detection frame;

and if the number is greater than 2, screening all the second specified detection frames according to a second screening rule corresponding to the distance between the central abscissa of the two adjacent second specified detection frames, the average value of the width of the detection frames and the average value of the height of the first detection frames, so as to obtain the sixth detection frame.

Optionally, the step of performing preset row correction processing on the sixth detection frame based on the second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame includes:

Judging whether the number of fifth negative increasing detection frames corresponding to the negative increasing field contained in a sixth detection frame in a second block is smaller than the number of fifth metal amount detection frames corresponding to the amount field contained in the second block; wherein the second block is any block after block division;

if not, traversing all the fifth increasing negative detection frames according to a first preset sequence to obtain the longitudinal distance between two adjacent fifth increasing negative detection frames;

obtaining average longitudinal distances of all the fifth negative increasing detection frames;

calculating a first difference absolute value of the longitudinal spacing and the average longitudinal spacing, and judging whether the first difference absolute value is larger than a preset numerical threshold;

if yes, acquiring the abscissa of the two adjacent fifth increasing and decreasing detection frames, and acquiring the average height values of the second detection frames of all the fifth increasing and decreasing detection frames;

and carrying out frame adding processing in the two adjacent fifth negative-increasing detection frames based on the abscissa and the average height value of the second detection frame to obtain an added fifth negative-increasing detection frame, carrying out matching processing between the longitudinal spacing and the average longitudinal spacing of the two adjacent fifth negative-increasing detection frames again on the obtained added fifth negative-increasing detection frame, and carrying out frame adding processing in the two fifth negative-increasing detection frames with the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing being greater than the preset numerical threshold until the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing of any two fifth negative-increasing detection frames is not greater than the preset numerical threshold, thereby obtaining the seventh detection frame.

Optionally, the step of performing matching processing on a second increasing negative detection frame and a second amount detection frame included in the second detection frame in each block according to a second preset rule to obtain a corresponding first matching frame includes:

acquiring a preset average slope;

traversing a sixth increasing and decreasing detection frame corresponding to the increasing and decreasing field, which is included in a second detection frame in a third block, according to a second preset sequence, and respectively calculating the slope between the sixth increasing and decreasing detection frame traversed currently and the sixth amount detection frame corresponding to each amount field for the sixth increasing and decreasing detection frame traversed currently; the sixth negative increasing detection frame traversed currently is marked as a designated negative increasing detection frame;

acquiring the slope with the smallest value from all the slopes, and judging whether the absolute value of the second difference value between the smallest slope and the average slope is in a preset value range;

if so, acquiring a specified amount detection frame corresponding to the slope with the smallest value, judging that the specified increase and decrease detection frame has a matching relationship with the specified amount detection frame, and establishing a matching relationship for the specified increase and decrease detection frame and the specified detection frame until the matching processing of all the sixth increase and decrease detection frames is completed to obtain the corresponding first matching frame.

The application also provides an invoice information recognition device based on deep learning, which comprises:

the detection module is used for detecting the medical invoice image based on a preset detection model to obtain a first detection frame corresponding to the target field; the target field comprises a negative increasing field and an amount field, and the first detection frame comprises a first negative increasing detection frame corresponding to the negative increasing field and a first amount detection frame corresponding to the amount field;

the first processing module is used for processing the detection frames according to a first preset rule based on the spatial position information of the first detection frames to obtain second detection frames after block division;

the matching module is used for carrying out matching processing on a second increasing negative detection frame and a second amount detection frame which are contained in a second detection frame in each block according to a second preset rule to obtain a corresponding first matching frame; each first matching frame comprises a pair of second increasing negative detection frames and second amount detection frames which have a matching relationship;

the second processing module is used for acquiring a general recognition result of a preset project frame in the medical invoice image, filtering the first matching frame based on the general recognition result to obtain a corresponding second matching frame, and writing corresponding first text information in the second matching frame based on the general recognition result;

The third processing module is used for filtering all detection frames contained in the second matching frame based on a preset text filling rule to obtain a corresponding third detection frame, and matching the third detection frame according to the second preset rule to obtain a corresponding third matching frame;

the calculation module is used for calculating the second text information in the third matching frame according to a preset numerical calculation rule corresponding to all the self-payment amount fields and part of the self-payment amount fields, and generating a first numerical value of all the self-payment amount fields and a second numerical value of part of the self-payment amount fields; the total self-fee amount field is a field corresponding to the sum of amounts spent by all the self-fee items contained in the medical invoice image, and the partial self-fee amount field is a field corresponding to the sum of self-fee amounts in the amounts spent by all the partial self-fee items contained in the medical invoice image.

The application also provides a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the steps of the above method when executing the computer program.

The present application also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the above method.

The invoice information identification method, device, computer equipment and storage medium based on deep learning provided by the application have the following beneficial effects:

according to the deep learning-based invoice information identification method, device, computer equipment and storage medium, unlike the existing mode of manually carrying out information accounting according to various relevant content data in the item/medicine details of the medical invoice, after the medical invoice image to be processed is acquired, all self-charge amount information and part of self-charge amount information can be accurately extracted from the medical invoice in an automatic mode based on the use of a preset detection model, a text filling rule, a numerical calculation rule and spatial position information and content information in the medical invoice image, and the accuracy and the intelligence of the content identification of the medical invoice are effectively improved. In addition, because the information accounting is not needed by relying on manpower according to the relevant content data of each item in the item/medicine detail of the medical invoice, the manual workload can be greatly reduced, and the processing efficiency of generating all self-charge amount information and part of self-charge amount information is effectively improved. And this scheme can be applied to in the wisdom medical field to promote the construction of wisdom city.

Drawings

FIG. 1 is a flow diagram of a deep learning based invoice information recognition method according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a deep learning based invoice information recognition device according to an embodiment of the present application;

fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present application.

The realization, functional characteristics and advantages of the present application will be further described with reference to the embodiments, referring to the attached drawings.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

The embodiment of the application can acquire and process the related data based on the artificial intelligence technology. Among these, artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use knowledge to obtain optimal results.

Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

Referring to fig. 1, a deep learning-based invoice information recognition method according to an embodiment of the present application includes:

s1: detecting the medical invoice image based on a preset detection model to obtain a first detection frame corresponding to the target field; the target field comprises a negative increasing field and an amount field, and the first detection frame comprises a first negative increasing detection frame corresponding to the negative increasing field and a first amount detection frame corresponding to the amount field;

s2: processing the detection frames according to a first preset rule based on the spatial position information of the first detection frames to obtain second detection frames after block division;

s3: matching the second increasing and decreasing detection frames contained in the second detection frames in each block with the second amount detection frames according to a second preset rule to obtain corresponding first matching frames; each first matching frame comprises a pair of second increasing negative detection frames and second amount detection frames which have a matching relationship;

S4: obtaining a general recognition result of a preset project frame in the medical invoice image, filtering the first matching frame based on the general recognition result to obtain a corresponding second matching frame, and writing corresponding first text information in the second matching frame based on the general recognition result;

s5: filtering all detection frames contained in the second matching frame based on a preset text filling rule to obtain a corresponding third detection frame, and matching the third detection frame according to the second preset rule to obtain a corresponding third matching frame;

s6: calculating the second text information in the third matching frame according to a preset numerical calculation rule corresponding to all the self-charge amount fields and part of the self-charge amount fields, and generating a first numerical value of all the self-charge amount fields and a second numerical value of part of the self-charge amount fields; the total self-fee amount field is a field corresponding to the sum of amounts spent by all the self-fee items contained in the medical invoice image, and the partial self-fee amount field is a field corresponding to the sum of self-fee amounts in the amounts spent by all the partial self-fee items contained in the medical invoice image.

As described in steps S1 to S6, the execution subject of the embodiment of the method is an invoice information recognition device based on deep learning. In practical application, the invoice information recognition device based on deep learning may be implemented by a virtual device, for example, a software code, or may be implemented by an entity device in which related execution codes are written or integrated, and may perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device. According to the deep learning-based invoice information identification device, text information corresponding to the charge increasing field and the amount field can be accurately acquired from the medical invoice image, and further, the first value of the self-charge amount field and the second value of the partial self-charge amount field are rapidly and accurately generated based on the acquired text information, so that the intelligence and the accuracy of medical invoice information extraction are improved. Specifically, before detecting a medical invoice image based on a preset detection model, a medical invoice image to be processed is first acquired. The medical invoice image is an electronic image of a medical invoice. The medical invoice at least comprises a charge-up field and an amount field, and can also comprise fields such as a name, a number, a date and the like, and the medical invoice can be a charging invoice of Tianjin medical clinic.

Then detecting the medical invoice image based on a preset detection model to obtain a first detection frame corresponding to the target field; the target field comprises a plus-minus field and an amount field, and the first detection frame comprises a first plus-minus detection frame corresponding to the plus-minus field and a first amount detection frame corresponding to the amount field. The medical invoice image is detected by using the preset detection model, a target area which corresponds to a target field and is to be subjected to data information extraction of the target field can be detected, the target area is a first detection frame which corresponds to the target field, and coordinate information of the first detection frame can be output, wherein the coordinate information can comprise a center point abscissa and a center point ordinate. The preset detection model can be obtained based on the existing detection model. Specifically, in order to enable the detection model to learn the characteristics of the target area corresponding to the target field, the corresponding detection model needs to be trained in advance according to the detection requirement of the target area. For example, if it is necessary to detect the amount information on the medical invoice, it is necessary to train the detection model in advance using a large number of medical invoice images as training samples. And during training, the medical invoice image is taken as input, the medical invoice image of the area marked with the monetary information is taken as output, and finally the detection model is obtained through training. When the trained detection model is used for inputting the medical invoice image, the detection model only identifies the area corresponding to the amount information, and other areas except the area cannot be detected.

And then processing the detection frames according to a first preset rule based on the spatial position information of the first detection frames to obtain second detection frames after block division. The processing the detection frame according to a first preset rule based on the spatial position information of the first detection frame to obtain a second detection frame after block division may include: based on the coordinate information of the first detection frame, filtering the first detection frame through a non-maximum suppression algorithm to obtain a corresponding fourth detection frame; performing block division processing on the fourth detection frame based on the abscissa of the central point of the fourth detection frame to obtain a fifth detection frame after block division; performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the fifth detection frame to obtain a corresponding sixth detection frame; performing preset row correction processing on the sixth detection frame based on second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame, and performing filtering processing on the seventh detection frame through the non-maximum suppression algorithm to obtain a corresponding eighth detection frame; and taking the eighth detection frame as the second detection frame.

After the second detection frame is obtained, carrying out matching processing on a second increasing negative detection frame and a second amount detection frame contained in the second detection frame in each block according to a second preset rule to obtain a corresponding first matching frame; each first matching frame comprises a pair of second increasing negative detection frames and second amount detection frames which have matching relations. The first matching frame is obtained by comparing the average slope with all the obtained slopes. After a first matching frame is obtained, a general identification result of a project frame preset in the medical invoice image is obtained, filtering processing is carried out on the first matching frame based on the general identification result, a corresponding second matching frame is obtained, and corresponding first text information is written in the second matching frame based on the general identification result. The item box is a region for filling data of each item in the medical invoice, for example, the item box is a rectangle formed by a lower region of an item/specification field and an upper region of an amount (upper case) field. In addition, the text recognition can be performed on the medical invoice image by using a preset text detection recognition model, so that a general recognition result is output according to the text condition in the item boxes in the medical invoice, and the general recognition result comprises a plurality of text detection boxes and corresponding text contents. The text detection recognition model may be an existing text detection recognition model, for example, a text detection recognition model obtained by training an MSER algorithm (Maximally Stable Extremal Regions, based on a maximum stable extremum region). Specifically, the text detection recognition model can be called to perform text recognition processing on the item boxes to obtain a plurality of text detection boxes and text information corresponding to the text detection boxes; judging whether a matched appointed text detection box exists in the first matching box or not; if the first matching frame exists, reserving the first matching frame, writing text information in the appointed text detection frame in the first matching frame, recording the text information as first text information, if the first matching frame does not exist, eliminating the first matching frame, and the like, and obtaining the second matching frame after the filtering processing of all the first matching frames is completed.

And filtering all detection frames contained in the second matching frame based on a preset text filling rule to obtain a corresponding third detection frame, and matching the third detection frame according to the second preset rule to obtain a corresponding third matching frame. The text filling rule refers to that filling data corresponding to a target field is required to follow a standard target data format, if text information filled in a detection frame accords with the text filling rule, the text filling rule is reserved, and if the text information filled in the detection frame does not accord with the text filling rule, the text filling rule is deleted. And finally, calculating the second text information in the third matching frame according to a preset numerical calculation rule corresponding to all the self-charge amount fields and part of the self-charge amount fields, and generating the first numerical value of all the self-charge amount fields and the second numerical value of part of the self-charge amount fields. The total self-fee amount field is a field corresponding to the sum of amounts spent by all the self-fee items contained in the medical invoice image, and the partial self-fee amount field is a field corresponding to the sum of self-fee amounts in the amounts spent by all the partial self-fee items contained in the medical invoice image. Additionally, the numerical calculation rule may include: each third matching frame corresponds to one medical item in the medical invoice image, which can be simply called an item, and the information in the third matching frames is the information of the corresponding item. Each third matching frame comprises a pair of specific increasing and decreasing detection frames with matching relation and a specific amount detection frame, text information written in the specific increasing and decreasing detection frames is a numerical value corresponding to an increasing and decreasing field, and text information written in the specific amount detection frames is a numerical value corresponding to an amount field. If the text information corresponding to the specific increasing and decreasing detection frame is 100% or 1, the item corresponding to the specific increasing and decreasing detection frame is all self-fee item, and all self-fee amount value corresponding to all self-fee amount field, namely the first value is the sum of the amount spent for all self-fee items contained in the medical invoice image; similarly, if the text information corresponding to the specific increment detecting frame is smaller than 1 and larger than 0, the item corresponding to the specific increment detecting frame is a partial expense item, and the partial expense amount corresponding to the partial expense amount field is a partial expense amount value, that is, the second value is the sum of the expense amounts of all the partial expense items contained in the medical invoice image, and the expense amount of the partial expense item is the product of the expense amount of the partial expense item and the text information in the increment detecting frame corresponding to the partial expense item.

In this embodiment, unlike the existing method that manually performs information accounting according to various relevant content data in the item/drug details of the medical invoice, the method of the embodiment can accurately extract all the self-payment amount information and part of the self-payment amount information from the medical invoice in an automatic manner based on the use of the preset detection model, the text filling rule, the numerical calculation rule and the spatial position information and the content information in the medical invoice image after the medical invoice image to be processed is acquired, thereby effectively improving the accuracy and the intelligence of the content identification of the medical invoice. In addition, because the information accounting is not needed by relying on manual work according to the relevant content data of each item in the item/medicine detail of the medical invoice, the manual work load can be greatly reduced, and the processing efficiency of generating all self-charge amount information and part of self-charge amount information is effectively improved. And this scheme can be applied to in the wisdom medical field to promote the construction of wisdom city.

Further, in an embodiment of the present application, the step S2 includes:

s20: based on the coordinate information of the first detection frame, filtering the first detection frame through a non-maximum suppression algorithm to obtain a corresponding fourth detection frame;

S21: performing block division processing on the fourth detection frame based on the abscissa of the central point of the fourth detection frame to obtain a fifth detection frame after block division;

s22: performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the fifth detection frame to obtain a corresponding sixth detection frame;

s23: performing preset row correction processing on the sixth detection frame based on second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame, and performing filtering processing on the seventh detection frame through the non-maximum suppression algorithm to obtain a corresponding eighth detection frame;

s24: and taking the eighth detection frame as the second detection frame.

As described in the above steps S20 to S24, the step of processing the detection frame according to a first preset rule based on the spatial position information of the first detection frame to obtain a second detection frame after block division may specifically include: and firstly, filtering the first detection frame through a non-maximum suppression algorithm based on the coordinate information of the first detection frame to obtain a corresponding fourth detection frame. The coordinate information may refer to a center point coordinate of the first detection frame, where the center point coordinate is formed by a center point abscissa and a center point ordinate. In addition, based on the coordinate information of the first detection frame, the first detection frame is filtered by a non-maximum suppression algorithm, and the implementation process of obtaining the corresponding second detection frame will be further described in the following specific embodiments, which are not described herein. And then carrying out block division processing on the fourth detection frame based on the abscissa of the central point of the fourth detection frame to obtain a fifth detection frame after block division. The method comprises the steps of calculating the mean value and standard deviation of the central point abscissas of all fourth detection frames through a preset calculation formula, obtaining the width value of a project frame in a preset medical invoice, comparing the relation between the central point abscissas of the fourth detection frames and the width value, the mean value and the standard deviation in sequence for each fourth detection frame, and carrying out block division processing on the fourth detection frames according to the relation so as to divide the fourth detection frames into left blocks or right blocks. And then, carrying out preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the fifth detection frame to obtain a corresponding sixth detection frame. The first spatial position information may include information such as a center point ordinate, a center point abscissa, an average width value of the detection frame, and an average height value of the detection frame. In addition, the implementation process of performing the preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the fifth detection frame will be further described in the following embodiments, which are not described herein. And carrying out preset row correction processing on the sixth detection frame based on second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame, and carrying out filtering processing on the seventh detection frame through the non-maximum suppression algorithm to obtain a corresponding eighth detection frame. The second spatial position information comprises longitudinal spacing, average longitudinal spacing, detection frame abscissa, detection frame average height value and the like. In addition, the implementation process of performing the preset row correction processing on the sixth detection frame based on the second spatial position information corresponding to the sixth detection frame will be further described in the following specific embodiments, which are not described herein. And finally, taking the eighth detection frame as the second detection frame. According to the embodiment, the detection frames are processed according to the first preset rule based on the spatial position information of the first detection frames, so that the second detection frames after the block division can be obtained quickly, the subsequent matching processing of the second increasing negative detection frames and the second amount detection frames contained in the second detection frames in each block according to the second preset rule is facilitated, and the corresponding first matching frames can be obtained quickly and accurately.

Further, in an embodiment of the present application, the step S20 includes:

s200: acquiring the confidence coefficient of each first detection frame and the coordinate value of each first detection frame, and acquiring the coordinate position information of the item frame;

s201: filtering detection frames with confidence coefficient of 0 from all the first detection frames and detection frames with filtering coordinate values not in the range of the coordinate position information to obtain corresponding ninth detection frames;

s202: the confidence degrees of all the ninth detection frames are ordered in a descending order, and corresponding first ordering results are obtained;

s203: screening a tenth detection frame with highest confidence from the first sequencing result;

s204: calculating the overlapping area between the tenth detection frame and other detection frames in the ninth detection frame;

s205: and acquiring a preset area threshold, deleting the detection frames with the overlapping area larger than the area threshold in the ninth detection frames, sequencing the rest detection frames in descending order of confidence, deleting the detection frames with the overlapping area larger than the area threshold again until the number of the detection frames with the overlapping area larger than the area threshold is 0, and taking all the rest detection frames as the fourth detection frames.

As described in the above steps S200 to S205, the step of filtering the first detection frame by a non-maximum suppression algorithm based on the coordinate information of the first detection frame to obtain a corresponding fourth detection frame may specifically include: firstly, confidence coefficient of each first detection frame and coordinate value of each first detection frame are obtained, and coordinate position information of the item frame is obtained. In addition, the confidence coefficient is obtained by a confidence function, wherein the confidence function is a function preset in advance, and a confidence coefficient calculation rule and a calculation parameter are arranged in the confidence function. And filtering detection frames with confidence coefficient of 0 from all the first detection frames and detection frames with filtering coordinate values not in the range of the coordinate position information to obtain corresponding ninth detection frames. The coordinate position information is an information range consisting of the horizontal and vertical coordinates of the upper left corner, the lower left corner, the upper right corner and the lower right corner of the project frame. And then, carrying out descending order sequencing on the confidence degrees of all the ninth detection frames to obtain a corresponding first sequencing result. And screening a tenth detection frame with highest confidence from the first sequencing result. And after the ninth detection frame is obtained, calculating the overlapping area between the tenth detection frame and other detection frames in the ninth detection frame. Wherein the overlap area may also be referred to as IoU (Intersect ion over Union, the ratio of the intersection and union of the ninth detection frame with the other detection frames),

A is the area of the tenth detection frame, B is the area of other detection frames, U is the intersection mathematical symbol, and U is the union mathematical symbol. And finally, acquiring a preset area threshold, deleting the detection frames with the overlapping area larger than the area threshold in the ninth detection frames, sequencing the rest detection frames in descending order of confidence, deleting the detection frames with the overlapping area larger than the area threshold again, and the like until the number of the detection frames with the overlapping area larger than the area threshold is 0, and taking all the rest detection frames as the fourth detection frames. Wherein the area threshold is a value preset in advance, and is specific to the area thresholdThe numerical value is not particularly limited. According to the embodiment, the non-maximum suppression algorithm is used for filtering the first detection frame, so that redundant detection frames can be quickly and accurately removed from the first detection frame, and then the corresponding fourth detection frame is obtained, the subsequent block division processing of the fourth detection frame can be quickly completed based on the fourth detection frame, and the accuracy of block division is guaranteed.

Further, in an embodiment of the present application, the step S21 includes:

s210: calculating the average value of the abscissa of the central points of all the first specified detection frames according to a first preset formula; the first appointed detection frame is a third negative increasing detection frame corresponding to the negative increasing field or a third amount detection frame corresponding to the amount field in the fourth detection frame;

S211: calculating standard deviations corresponding to the abscissa of the central points of all the first specified detection frames according to a second preset formula;

s212: acquiring a first preset threshold value, a second preset threshold value and a width value of the item frame;

s213: determining a block division result of each first specified detection frame based on the mean value, the standard deviation, the first preset threshold value, the second preset threshold value and the width value;

s214: and dividing each first appointed detection frame based on the block division result to obtain a fifth detection frame after block division.

As described in the above steps S210 to S214, the step of performing the block division processing on the fourth detection frame based on the abscissa of the center point of the fourth detection frame to obtain the fifth detection frame after the block division may specifically include: firstly, calculating the average value of the abscissa of the central point of all the first specified detection frames according to a first preset formula. The first appointed detection frame is a third negative increasing detection frame corresponding to the negative increasing field or a third amount detection frame corresponding to the amount field in the fourth detection frame. Specifically, the first preset formula is:

For the mean value of the abscissa of the central points of all the first designated detection frames, X _i And (3) the center point abscissa of the ith first specified detection frame, and n is the number of the first specified detection frames. And then calculating standard deviations corresponding to the abscissa of the central points of all the first specified detection frames according to a second preset formula. Specifically, the second preset formula is: />

S is the standard deviation corresponding to the abscissa of the central point of all the first specified detection frames,/->

For the mean value of the abscissa of the central points of all the first designated detection frames, X _i And (3) the center point abscissa of the ith first specified detection frame, and n is the number of the first specified detection frames. And then acquiring a first preset threshold value, a second preset threshold value and a width value of the item frame. The item frame is a region for filling data of each item in the medical invoice, for example, the item frame is a rectangle formed by a lower region of an item/specification field and an upper region of an amount (upper case) field, and the detection frame is located in the region of the item frame. And determining a block division result of each first specified detection frame based on the mean value, the standard deviation, the first preset threshold value, the second preset threshold value and the width value. And for each first specified detection frame, comparing the relation between the abscissa of the central point of the current first specified detection frame and the width value, the mean value and the standard deviation in sequence, and dividing the current first specified detection frame into a left block or a right block according to the relation. Specifically, judging whether the abscissa of the central point of the first specified detection frame is smaller than 1/2 times of the sum value of the width value and the first preset threshold value, and the abscissa of the central point of the first specified detection frame is smaller than or equal to the average value; if all of them meet, make the above-mentioned The first appointed detection frame is divided into left blocks; if not, judging whether the abscissa of the central point of the first specified detection frame is smaller than the sum of the width value which is 1/2 times and the first preset threshold value; if yes, dividing the first appointed detection frame into a left block; if not, judging whether the standard deviation is smaller than or equal to the second preset threshold value, and whether the abscissa of the central point of the first specified detection frame is smaller than or equal to 2/3 times the width value; if both the first specified detection frames are satisfied, dividing the first specified detection frames into left blocks; if not, dividing the first appointed detection frame into right blocks. In addition, the specific values of the first preset threshold and the second preset threshold are not particularly limited, and may be generated by performing statistical analysis according to detection frames marked in the training set of the text detection model in advance. And finally, dividing each first appointed detection frame based on the block division result to obtain a fifth detection frame after block division. According to the embodiment, the fourth detection frame is subjected to block division processing based on the abscissa of the central point of the fourth detection frame, so that a fifth detection frame after block division can be obtained quickly, the subsequent preset column correction processing can be performed on the obtained fifth detection frame, and the corresponding sixth detection frame can be obtained quickly and accurately.

Further, in an embodiment of the present application, the step S22 includes:

s220: acquiring all second specified detection frames in the first block; the first block is any block after block division, and the second specified detection frame is a fourth negative-increasing detection frame corresponding to the negative-increasing field or a fourth amount detection frame corresponding to the amount field, which is included in a fifth detection frame in the first block;

s221: sequencing all the second specified detection frames from large to small according to the numerical value of the ordinate of the central point to obtain a corresponding second sequencing result;

s222: acquiring the abscissa of the central point, the average width value of the detection frames and the average height value of the first detection frames corresponding to all the second specified detection frames;

s223: acquiring the number of all the second specified detection frames;

s224: according to the ordering of the second ordering result, calculating the interval of the central abscissa between two adjacent second specified detection frames;

s225: if the number is equal to 2, screening all the second specified detection frames according to a first screening rule corresponding to the interval of the central abscissa between the second specified detection frames to obtain a sixth detection frame;

S226: and if the number is greater than 2, screening all the second specified detection frames according to a second screening rule corresponding to the distance between the central abscissa of the two adjacent second specified detection frames, the average value of the width of the detection frames and the average value of the height of the first detection frames, so as to obtain the sixth detection frame.

As described in the above steps S220 to S226, the step of performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the third detection frame to obtain a corresponding sixth detection frame may specifically include: all second specified detection frames in the first block are first acquired. The first block is any block after the block division, and the second specified detection frame is a fourth negative-increasing detection frame corresponding to the negative-increasing field or a fourth amount detection frame corresponding to the amount field, which is included in the fifth detection frame in the first block. And then sequencing all the second specified detection frames according to the numerical value of the ordinate of the central point from large to small to obtain a corresponding second sequencing result. And then acquiring the abscissa of the central point, the average width value of the detection frames and the average height value of the first detection frames corresponding to all the second specified detection frames, and acquiring the number of all the second specified detection frames. Wherein the average width value of the detection frames refers to a quotient between a sum of width values of all the second specified detection frames and the number of all the second specified detection frames, and the average height value of the first detection frames refers to a quotient between a sum of height values of all the second specified detection frames and the number of all the second specified detection frames. And calculating the distance between the central abscissa coordinates of two adjacent second designated detection frames according to the sequence of the second sequence results. The distance is a data value obtained by subtracting the central abscissa with a smaller value from the central abscissa with a larger index value. And if the number is equal to 2, screening all the second specified detection frames according to a first screening rule corresponding to the interval of the central abscissa between the second specified detection frames to obtain the sixth detection frame. Specifically, the screening all the second specified detection frames according to the first screening rule corresponding to the interval of the central abscissa between the second specified detection frames, and obtaining the sixth detection frame may include: acquiring a preset interval threshold value, and judging whether the interval of the central abscissa between the second detection frames is larger than the interval threshold value or not; if yes, screening out second specified detection frames with low confidence from all the second specified detection frames, and deleting the second specified detection frames with low confidence to finish screening processing of all the second specified detection frames. And if the number is greater than 2, screening all the second specified detection frames according to a second screening rule corresponding to the distance between the central abscissa of the two adjacent second specified detection frames, the average value of the width of the detection frames and the average value of the height of the first detection frames, so as to obtain the sixth detection frame. Specifically, according to the second screening rule corresponding to the distance between the central abscissa of the two adjacent second designated detection frames, the average value of the width of the detection frames, and the average value of the height of the first detection frames, the screening process is performed on all the second designated detection frames, so as to obtain the sixth detection frame, where the obtaining may include: judging whether the distance between the central abscissa of the two adjacent second designated detection frames is larger than the average value of the width of the detection frames or not; if the reliability is higher than the reliability, deleting the second designated detection frames with low reliability in the two adjacent second designated detection frames; if not, judging whether the distance between the central abscissa of the two adjacent second designated detection frames is larger than the product of the designated width value and the designated numerical value; the appointed width value is an average value of the widths of the two adjacent second appointed detection frames, and the appointed numerical value is a positive integer; if yes, deleting the second designated detection frames with low confidence in the two adjacent second designated detection frames; if not, judging whether the distance between the central ordinate of the two adjacent second designated detection frames is smaller than the average value of the heights of the first detection frames, and whether the distance between the central abscissa of the two adjacent second designated detection frames is larger than the designated width value; if yes, deleting the second designated detection frames with low confidence in the two adjacent second designated detection frames to finish screening treatment of all the second designated detection frames. According to the embodiment, the corresponding screening rule is intelligently selected according to the second specified detection frame, and the corresponding filtering treatment is carried out on the second specified detection frame in the column dimension, so that the corresponding sixth detection frame can be accurately obtained, the subsequent rapid and intelligent generation of the required seventh detection frame according to the obtained sixth detection frame is facilitated, and the accuracy of the generated seventh detection frame is improved.

Further, in an embodiment of the present application, the step S23 includes:

s230: judging whether the number of fifth negative increasing detection frames corresponding to the negative increasing field contained in a sixth detection frame in a second block is smaller than the number of fifth metal amount detection frames corresponding to the amount field contained in the second block; wherein the second block is any block after block division;

s231: if not, traversing all the fifth increasing negative detection frames according to a first preset sequence to obtain the longitudinal distance between two adjacent fifth increasing negative detection frames;

s232: obtaining average longitudinal distances of all the fifth negative increasing detection frames;

s233: calculating a first difference absolute value of the longitudinal spacing and the average longitudinal spacing, and judging whether the first difference absolute value is larger than a preset numerical threshold;

s234: if yes, acquiring the abscissa of the two adjacent fifth increasing and decreasing detection frames, and acquiring the average height values of the second detection frames of all the fifth increasing and decreasing detection frames;

s235: and carrying out frame adding processing in the two adjacent fifth negative-increasing detection frames based on the abscissa and the average height value of the second detection frame to obtain an added fifth negative-increasing detection frame, carrying out matching processing between the longitudinal spacing and the average longitudinal spacing of the two adjacent fifth negative-increasing detection frames again on the obtained added fifth negative-increasing detection frame, and carrying out frame adding processing in the two fifth negative-increasing detection frames with the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing being greater than the preset numerical threshold until the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing of any two fifth negative-increasing detection frames is not greater than the preset numerical threshold, thereby obtaining the seventh detection frame.

As described in the foregoing steps S230 to S235, the step of performing a preset row correction process on the sixth detection frame based on the second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame may specifically include: first, it is determined whether the number of fifth negative-increasing detection frames corresponding to the negative-increasing field included in the sixth detection frame in the second block is smaller than the number of fifth amount detection frames corresponding to the amount field included in the second block. The second block is any block after the block division. And if the number of the detected frames is not smaller than the first preset number, traversing all the fifth negative increasing detection frames according to a first preset sequence to obtain the longitudinal distance between two adjacent fifth negative increasing detection frames. The first preset sequence may be a sequence from top to bottom. In addition, the longitudinal distance may refer to a data value obtained by subtracting a center ordinate with a smaller value from a center ordinate with a larger value in two adjacent fifth increasing negative detection frames. And then obtaining the average longitudinal spacing of all the fifth negative-increasing detection frames. The longitudinal spacing of two adjacent fifth negative-increasing detection frames can be sequentially obtained, and then the quotient between the sum value of all the obtained longitudinal spacing and the obtained number of all the longitudinal spacing is calculated to be used as the longitudinal spacing. And then calculating a first difference absolute value of the longitudinal spacing and the average longitudinal spacing, and judging whether the first difference absolute value is larger than a preset numerical threshold. The value of the preset numerical threshold is not particularly limited, and can be determined according to actual requirements. If yes, acquiring the abscissa of the two adjacent fifth increasing and decreasing detection frames, and acquiring the average height values of the second detection frames of all the fifth increasing and decreasing detection frames. And finally, carrying out frame adding treatment in the two adjacent fifth negative-increasing detection frames based on the abscissa and the average height value of the second detection frame to obtain an added fifth negative-increasing detection frame, carrying out matching treatment between the longitudinal spacing and the average longitudinal spacing of the two adjacent fifth negative-increasing detection frames again on the obtained added fifth negative-increasing detection frame, carrying out frame adding treatment in the two fifth negative-increasing detection frames with the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing being greater than the preset numerical threshold, and the like until the absolute value of the first difference between the longitudinal spacing and the average longitudinal spacing of any two fifth negative-increasing detection frames is not greater than the preset numerical threshold, and obtaining the seventh detection frame. The adding frame processing means adding a new detection frame in the two adjacent fifth negative-increasing detection frames. Specifically, the creating process of the new detection frame may include: firstly, acquiring an average value of left abscissas of the two adjacent fifth negative-increasing detection frames as a left abscissas of the new detection frame, and acquiring an average value of right abscissas of the two adjacent fifth negative-increasing detection frames as a right abscissas of the new detection frame; calculating the average value of the center point coordinates of the two adjacent fifth negative increasing detection frames to obtain a designated center point coordinate; then the ordinate of the specified center point coordinates is added with the average height value of the second detection frame by 1/2 times to be used as the upper ordinate of the new detection frame, and the average height value of the second detection frame which is subtracted from the ordinate by 1/2 times to be used as the lower ordinate of the new detection frame. According to the embodiment, the preset row correction processing can be performed on the sixth detection frame based on the second spatial position information corresponding to the sixth detection frame, so that the corresponding seventh detection frame can be accurately obtained, the subsequent rapid and intelligent generation of the required second detection frame according to the obtained seventh detection frame is facilitated, and the accuracy of the generated second detection frame is improved.

Further, in an embodiment of the present application, the step S3 includes:

s30: acquiring a preset average slope;

s31: traversing a sixth increasing and decreasing detection frame corresponding to the increasing and decreasing field, which is included in a second detection frame in a third block, according to a second preset sequence, and respectively calculating the slope between the sixth increasing and decreasing detection frame traversed currently and the sixth amount detection frame corresponding to each amount field for the sixth increasing and decreasing detection frame traversed currently; the sixth negative increasing detection frame traversed currently is marked as a designated negative increasing detection frame;

s32: acquiring the slope with the smallest value from all the slopes, and judging whether the absolute value of the second difference value between the smallest slope and the average slope is in a preset value range;

s33: if so, acquiring a specified amount detection frame corresponding to the slope with the smallest value, judging that the specified increase and decrease detection frame has a matching relationship with the specified amount detection frame, and establishing a matching relationship for the specified increase and decrease detection frame and the specified detection frame until the matching processing of all the fifth increase and decrease detection frames is completed to obtain the corresponding first matching frame.

As described in the above steps S30 to S33, the step of performing matching processing on the second increasing and decreasing detection frame and the second amount detection frame included in the second detection frame in each block according to the second preset rule to obtain a corresponding first matching frame may specifically include: firstly, a preset average slope is obtained. The average slope may be an average slope from the increment detection frame to the sum detection frame calculated according to the sequence from top to bottom. Specifically, for any one of the negative-increasing detection frames of the specified block, the slope between the negative-increasing detection frame and each of the amount detection frames in the specified block can be calculated, the specified slope with the smallest value can be obtained from the obtained slope, the quotient between the sum of the specified slopes corresponding to all the negative-increasing detection frames of the specified block and the number of all the negative-increasing detection frames of the specified block can be calculated, and the obtained resultAs said average slope. Then traversing the sixth increasing and decreasing detection frames corresponding to the increasing and decreasing fields, which are included in the second detection frames in the third block, according to a second preset sequence, and respectively calculating the slope between the sixth increasing and decreasing detection frames traversed currently and the sixth amount detection frames corresponding to each amount field for the sixth increasing and decreasing detection frames traversed currently; and marking the sixth negative increasing detection frame traversed currently as a designated negative increasing detection frame. The second predetermined sequence may be smooth from top to bottom. In addition, the formula can be passed

Calculating slope, wherein slope is the slope between the specified plus-minus detection frame and the sixth amount detection frame, x ₁ To specify the central abscissa of the plus-minus detection frame, y ₁ To specify the central ordinate, x, of the plus-minus detection frame ₂ The center abscissa, y, of the sixth amount detection frame ₂ The center ordinate of the sixth amount detection frame. And then acquiring the slope with the smallest value from all the slopes, and judging whether the absolute value of the second difference value between the smallest slope and the average slope is in a preset value range. The value of the preset numerical range is not particularly limited, and may be set according to actual requirements. If so, acquiring a specified amount detection frame corresponding to the slope with the smallest value, judging that the specified increase and decrease detection frame has a matching relationship with the specified amount detection frame, establishing a matching relationship for the specified increase and decrease detection frame and the specified detection frame, and the like until the matching processing of all the sixth increase and decrease detection frames is completed to obtain the corresponding first matching frame. According to the embodiment, the sixth increasing negative detection frame and the sixth amount detection frame contained in the second detection frame in each block are respectively matched by utilizing the preset rule, so that the corresponding first matching frame can be obtained quickly, the accuracy of the obtained first matching frame is ensured, the first matching frame can be filtered based on the general identification result of the preset project frame, and the required second matching frame can be obtained quickly.

Further, in an embodiment of the present application, the step S5 includes:

s50: acquiring a target data format corresponding to the target field;

s51: performing matching processing on all the first text information based on the target data format, and judging whether second text information which does not accord with the target data format exists in the first text information;

s52: if yes, an eleventh detection frame corresponding to the second text information is screened out from the second matching frame;

s53: and deleting the eleventh detection frame from the second matching frame to obtain the third detection frame.

As described in the above steps S50 to S53, the step of filtering all the detection frames included in the second matching frame based on the preset text filling rule to obtain a corresponding seventh detection frame may specifically include: first, a target data format corresponding to the target field is acquired. The target data format of the padding data corresponding to the negative-adding field may include: a percentage of less than 1, a fraction, several specific chinese, etc. The target data format of the amount field corresponding to the padding data may include: lower case numbers. And then carrying out matching processing on all the first text information based on the target data format, and judging whether second text information which does not accord with the target data format exists in the first text information. If yes, an eleventh detection frame corresponding to the second text information is screened out from the second matching frame. And finally deleting the eleventh detection frame from the second matching frame to obtain the third detection frame. According to the embodiment, all detection frames contained in the second matching frame are filtered based on a preset text filling rule, so that a corresponding third detection frame can be obtained quickly, a third matching frame required by the subsequent quick generation based on the third detection frame can be generated conveniently, and further the first numerical value of the self-charge amount field and the second numerical value of the partial self-charge amount field can be generated accurately based on the second text information in the third matching frame.

The invoice information identification method based on deep learning in the embodiment of the application can also be applied to the field of blockchains, for example, the first numerical value and other data are stored on the blockchain. By using the blockchain to store and manage the first value, the security and non-tamperability of the first value can be effectively ensured.

The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism, an encryption algorithm and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.

The blockchain underlying platform may include processing modules for user management, basic services, smart contracts, operation monitoring, and the like. The user management module is responsible for identity information management of all blockchain participants, including maintenance of public and private key generation (account management), key management, maintenance of corresponding relation between the real identity of the user and the blockchain address (authority management) and the like, and under the condition of authorization, supervision and audit of transaction conditions of certain real identities, and provision of rule configuration (wind control audit) of risk control; the basic service module is deployed on all block chain node devices, is used for verifying the validity of a service request, recording the service request on a storage after the effective request is identified, for a new service request, the basic service firstly analyzes interface adaptation and authenticates the interface adaptation, encrypts service information (identification management) through an identification algorithm, and transmits the encrypted service information to a shared account book (network communication) in a complete and consistent manner, and records and stores the service information; the intelligent contract module is responsible for registering and issuing contracts, triggering contracts and executing contracts, a developer can define contract logic through a certain programming language, issue the contract logic to a blockchain (contract registering), invoke keys or other event triggering execution according to the logic of contract clauses to complete the contract logic, and simultaneously provide a function of registering contract upgrading; the operation monitoring module is mainly responsible for deployment in the product release process, modification of configuration, contract setting, cloud adaptation and visual output of real-time states in product operation, for example: alarms, monitoring network conditions, monitoring node device health status, etc.

Referring to fig. 2, in an embodiment of the present application, there is further provided an invoice information recognition device based on deep learning, including:

the detection module 1 is used for detecting the medical invoice image based on a preset detection model to obtain a first detection frame corresponding to the target field; the target field comprises a negative increasing field and an amount field, and the first detection frame comprises a first negative increasing detection frame corresponding to the negative increasing field and a first amount detection frame corresponding to the amount field;

the first processing module 2 is configured to process the detection frame according to a first preset rule based on the spatial position information of the first detection frame, so as to obtain a second detection frame after block division;

the matching module 3 is configured to perform matching processing on a second increasing negative detection frame and a second amount detection frame included in the second detection frames in each block according to a second preset rule, so as to obtain a corresponding first matching frame; each first matching frame comprises a pair of second increasing negative detection frames and second amount detection frames which have a matching relationship;

the second processing module 4 is configured to obtain a general recognition result of a preset item frame in the medical invoice image, perform filtering processing on the first matching frame based on the general recognition result to obtain a corresponding second matching frame, and write corresponding first text information in the second matching frame based on the general recognition result;

The third processing module 5 is configured to perform filtering processing on all detection frames included in the second matching frame based on a preset text filling rule to obtain a corresponding third detection frame, and perform matching processing on the third detection frame according to the second preset rule to obtain a corresponding third matching frame;

the calculating module 6 is configured to calculate the second text information in the third matching frame according to a preset numerical calculation rule corresponding to all the self-payment amount fields and part of the self-payment amount fields, so as to generate a first numerical value of all the self-payment amount fields and a second numerical value of part of the self-payment amount fields; the total self-fee amount field is a field corresponding to the sum of amounts spent by all the self-fee items contained in the medical invoice image, and the partial self-fee amount field is a field corresponding to the sum of self-fee amounts in the amounts spent by all the partial self-fee items contained in the medical invoice image.

In this embodiment, the operations performed by the modules or units respectively correspond to the steps of the deep learning-based invoice information identification method in the foregoing embodiment one by one, and are not described herein again.

Referring to fig. 3, a computer device is further provided in the embodiment of the present application, where the computer device may be a server, and the internal structure of the computer device may be as shown in fig. 3. The computer device includes a processor, a memory, a network interface, a display screen, an input device, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a storage medium, an internal memory. The storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the storage media. The database of the computer device is used for storing the medical invoice image, the general recognition result, the first numerical value and the second numerical value. The network interface of the computer device is used for communicating with an external terminal through a network connection. The display screen of the computer equipment is an indispensable image-text output equipment in the computer and is used for converting digital signals into optical signals so that characters and graphics can be displayed on the screen of the display screen. The input device of the computer equipment is a main device for exchanging information between the computer and a user or other equipment, and is used for conveying data, instructions, certain sign information and the like into the computer. The computer program, when executed by a processor, implements a deep learning based invoice information recognition method.

The processor executes the steps of the invoice information identification method based on deep learning:

Those skilled in the art will appreciate that the structures shown in fig. 3 are only block diagrams of portions of structures that may be associated with the aspects of the present application and are not intended to limit the scope of the apparatus, or computer devices on which the aspects of the present application may be implemented.

An embodiment of the present application further provides a computer readable storage medium, on which a computer program is stored, where the computer program when executed by a processor implements a deep learning-based invoice information identification method, specifically:

Those skilled in the art will appreciate that implementing all or part of the above-described embodiment methods may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed, may comprise the steps of the above-described embodiment methods. Any reference to memory, storage, database, or other medium provided herein and used in embodiments may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, apparatus, article or method that comprises the element.

The foregoing description is only of the preferred embodiments of the present application, and is not intended to limit the scope of the claims, and all equivalent structures or equivalent processes using the descriptions and drawings of the present application, or direct or indirect application in other related technical fields are included in the scope of the claims of the present application.

Claims

1. An invoice information identification method based on deep learning is characterized by comprising the following steps:

Based on the spatial position information of the first detection frame, performing block division on the detection frame according to a first preset rule to obtain a second detection frame;

2. The method for identifying invoice information based on a model according to claim 1, wherein the step of processing the detection frames according to a first preset rule based on the spatial position information of the first detection frames to obtain second detection frames after block division comprises the steps of:

and taking the eighth detection frame as the second detection frame.

3. The method for identifying invoice information based on deep learning according to claim 2, wherein the step of filtering the first detection frame by a non-maximum suppression algorithm based on the coordinate information of the first detection frame to obtain a corresponding fourth detection frame includes:

4. The method for identifying invoice information based on deep learning according to claim 2, wherein the step of performing block division processing on the fourth detection frame based on the abscissa of the center point of the fourth detection frame to obtain a fifth detection frame after block division comprises the steps of:

5. The method for identifying invoice information based on deep learning according to claim 2, wherein the step of performing preset column correction processing on the fifth detection frame of each block based on the first spatial position information corresponding to the third detection frame to obtain a corresponding sixth detection frame includes:

acquiring the number of all the second specified detection frames;

6. The method for identifying invoice information based on deep learning according to claim 2, wherein the step of performing preset line correction processing on the sixth detection frame based on the second spatial position information corresponding to the sixth detection frame to obtain a corresponding seventh detection frame includes:

7. The method for identifying invoice information based on deep learning according to claim 1, wherein the step of matching a second increasing negative detection frame and a second amount detection frame included in a second detection frame in each block according to a second preset rule to obtain a corresponding first matching frame comprises the steps of:

acquiring a preset average slope;

8. An invoice information recognition device based on deep learning, which is characterized by comprising:

9. A computer device comprising a memory and a processor, the memory having stored therein a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 7.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.