A method for applying OCR to structured extraction of archival data

Abstract

A method for applying OCR to structured extraction of archival data, comprising an archival data preprocessing module, an ultra-clear reconstruction module, an image segmentation module, a character recognition module, an archival management module, an information retrieval module, and an information tracing module; the method for applying OCR to structured extraction of archival data proposed by the present invention processes archival files into image files, improves the quality of archival image files through the proposed image reconstruction algorithm, performs tabular cutting on the images, and then recognizes the text in the archival images through character recognition, establishes a document slice library, and users can directly display search information in archival images through text retrieval.

Description

Technical Field

[0001] The present invention relates to the field of archive management, and in particular to a method for applying OCR to structured extraction of archive data. Background Art

[0002] With the progress and development of society, digital technology has become an indispensable part of all walks of life. In this digital age, a large amount of archival data needs to be effectively managed, utilized and protected to meet the needs of information development. However, traditional archival management methods often rely on paper documents, which are inefficient, costly and error-prone. With the continuous increase in the amount of information, traditional manual processing methods can no longer meet the needs of fast, accurate and efficient processing of archival data. Therefore, it is particularly important to use advanced technical means to automate the processing and structured extraction of archival data. As an important digital tool, optical character recognition (OCR) technology has the ability to convert paper documents into editable and searchable electronic texts, bringing new development opportunities for archival management. In the past, archival management mainly relied on paper documents, which required a lot of manpower and material resources to organize, classify and store. However, this traditional management method has many problems. First, paper archives are easily affected by natural disasters, human damage and other factors, resulting in the loss and damage of archival information. Secondly, the traditional paper archive management method is inefficient and requires a lot of time and labor costs. Especially in the case of large-scale data management, manual processing often fails to meet the needs and is prone to errors and omissions. In addition, the retrieval and sharing of paper archives also face many difficulties and cannot meet the needs of rapid information exchange and sharing. With the continuous development and popularization of information technology, digital archive management has gradually become an inevitable trend. Digital archive management converts paper archives into electronic documents, realizes the digital storage, retrieval and sharing of archive information, and greatly improves the efficiency and quality of archive management. However, traditional digital archive management methods often rely on manual input and editing, which has problems such as low efficiency and high cost. Therefore, how to use advanced technical means to achieve automatic processing and structured extraction of archive data has become an important challenge facing current digital archive management. Optical character recognition (OCR) technology, as an important digital tool, has the ability to convert paper documents into editable and searchable electronic texts, bringing new development opportunities for archive management. By adopting OCR technology, automatic processing and structured extraction of archive data can be achieved, which greatly improves the efficiency and accuracy of archive management, reduces costs, and also increases the value of data utilization. Therefore, using OCR technology to automatically process and structure the archive data has become an important development direction of current digital archive management, which will greatly promote the development of digital archive management and provide strong support for the construction of an information society. Summary of the invention

[0003] The purpose of the present invention is to provide a method for applying OCR to structured extraction of archival data, aiming to solve the problems mentioned in the above background.

[0004] In order to achieve the above-mentioned purpose, the present invention proposes a method for applying OCR to structured extraction of archival data, comprising an archival data preprocessing module, an ultra-clear reconstruction module, an image segmentation module, a character recognition module, an archival management module, an information retrieval module, and an information tracing module; the archival data preprocessing module scans and identifies archival files and processes the archival files into image files; the ultra-clear reconstruction module proposes an image reconstruction algorithm and performs high-definition image reconstruction based on image pixel values; the image segmentation module performs tabular segmentation on archival images to form cells; the character recognition module proposes a high-performance character recognition algorithm to recognize archival image text; the archival management module classifies and manages the archival text data recognized by the character recognition module; in the information retrieval module, the user searches by text input to view the required archival information; in the information tracing module, the information retrieved by the user is segmented and highlighted to display the location in the archival image.

[0005] Furthermore, the archive data preprocessing module performs high-definition scanning on the archive files through a high-definition scanner to generate high-definition archive image files, and marks the archive image serial numbers of the generated high-definition archive image files according to the scanning order.

[0006] Furthermore, the ultra-high definition reconstruction module proposes an image reconstruction algorithm, optimizes the archival image based on the pixel shift characteristics of the pixel values ​​of the high-definition archival image file, and reconstructs the archival image in high definition.

[0007] Furthermore, the pixel shift feature is specifically as follows: for high-definition archive pictures, pixel value analysis is first performed, and the high-definition archive pictures are represented by a pixel matrix, which is represented as follows:

[0008]

[0009] Among them, B represents the pixel matrix of the high-definition archive picture, b1,1, b1,n, bn,1, bi,j, bn,n represent the pixel value of the 1st row and 1st column of the pixel matrix of the high-definition archive picture, the pixel value of the 1st row and nth column of the pixel matrix, the pixel value of the nth row and 1st column of the pixel matrix, the pixel value of the i-th row and j-th column of the pixel matrix, and the pixel value of the n-th row and n-th column of the pixel matrix, respectively. For the pixel matrix, an offset operation is performed to obtain the pixel offset feature, which represents the feature of each pixel point. The calculation formula is as follows;

[0010]

[0011] Where Δb i,j represents the pixel offset feature of pixel point (i, j), b i+l,kRepresents the pixel value of the i+lth row and kth column of the pixel matrix, calculates the pixel offset feature in the neighborhood of the pixel point (i, j), and then based on the pixel offset feature, performs diffusion processing on the pixel matrix of the high-definition archive image. The diffusion processing process is as follows:

[0012]

[0013] Where φ(·) represents the diffusion function, (β i ,β j ) represents the coordinate value of the pixel point (i, j), (α i ,α j ) represents the coordinate value of a point in the neighborhood of pixel point (i, j), Δb i,k The pixel offset feature of the pixel point (i, k) is represented. Through the diffusion function, each pixel point is used as the base point to diffuse the pixel matrix to obtain the basic archive image. Based on the basic archive image, the optimized archive image is obtained through iterative correction. The iterative correction formula is as follows:

[0014] g t =g t-1 +H t ×(g t-1 -g 0 ),

[0015] Among them, g 0 represents the basic archive image, g t-1 represents the optimized archive image at stage t-1, g t represents the optimized archive image at stage t, H t Represents the iterative optimization matrix of stage t. Based on all optimized archive images, the optimized archive image sequence G is constructed, which is expressed as follows:

[0016] G=[g 0 ,g 1 ,…,g i ,…,g t ],

[0017] Among them, g 1 , g i They represent the optimized archive image of the first stage and the optimized archive image of the i-th stage respectively. Then, a reconstruction probability method is proposed. Based on the optimized archive image sequence, the reconstructed archive image is obtained. The formula is as follows:

[0018]

[0019] Wherein, Y represents the reconstructed archive image, L represents the final output of the archive image reconstruction, p(·) represents the probability of the image, and the final reconstructed archive image is obtained by optimizing the archive image sequence. The image reconstruction algorithm proposed in the present invention performs pixel analysis on the archive image file, constructs the archive image file into a pixel matrix, and then obtains the pixel offset characteristics of the pixel point through an offset operation, and then performs diffusion processing through a diffusion function, performs a diffusion operation on the pixel matrix, and then iteratively optimizes the archive image through an iterative correction formula to obtain an optimized archive image sequence. Based on the optimized archive image sequence, the archive image is reconstructed in high definition through a probability formula. The image reconstruction algorithm proposed in the present invention improves the quality of the image and further improves the efficiency and accuracy of character recognition.

[0020] Furthermore, the image segmentation module performs tabular segmentation on the ultra-high-resolution reconstructed image file based on the UNet segmentation model, and divides the archive image into cells, with all the archive text contained in the cells.

[0021] Furthermore, the character recognition module proposes a high-performance character recognition algorithm and performs character recognition on archival image files by constructing a character recognition model.

[0022] Furthermore, the character recognition model has a detailed process as follows: the character recognition model includes an input layer, a hidden layer, and a prediction layer. The reconstructed archive image output by the ultra-clear reconstruction module is input into the input layer of the character recognition model. First, feature extraction is performed to obtain the archive image feature vector. The archive image feature vector is used to construct a feature sequence D, which is expressed as:

[0023] D = [χ 1 ,x 2 ,…,χ t ,…,χ n ],

[0024] Among them, χ 1 , χ 2 , χ t , χ n Respectively represent the first archival image feature vector, the second archival image feature vector, the tth archival image feature vector, and the nth archival image feature vector. Based on the archival image feature vector, a mapping transformation matrix is ​​constructed to transform the archival image feature vector χ t Perform mapping transformation to obtain the change feature vector χ t ′, the formula is as follows:

[0025]

[0026] Among them, S represents the feature network step size, θ x,1 ,θ y,1 ,θx,t ,θ y,t ,θ x,n ,θ y,n They represent the archival image feature vector χ 1 The horizontal and vertical coordinates of the position of the corresponding regional center in the feature map, the feature vector χ t The horizontal and vertical coordinates of the position of the corresponding regional center in the feature map, the feature vector χ n The horizontal and vertical coordinates of the position of the corresponding area center in the feature map are transformed into feature vectors as the output of the input layer of the character recognition model, which are input to the hidden layer for iteration. When the iteration reaches the k stage, the hidden layer encoding weight is calculated as follows:

[0027]

[0028] Among them, χ′ t,k represents the k-stage χ t The changing characteristic vector, r t,k represents the hidden layer state at time k, represents the iterative weight matrix, Represents the k-stage change feature vector χ′ t,k The hidden layer encoding weights are used to output the predicted character vector in the output layer, and the loss function is constructed to optimize the prediction training results. The loss function F formula is as follows:

[0029]

[0030] Among them, y k represents the character recognition output of the k stage, ||·||2 represents the bi-norm, and the character recognition prediction result is gradually optimized through the loss function. The high-performance character recognition algorithm proposed by the present invention constructs a character recognition model, extracts features of the high-definition archive image in the input layer, obtains the archive image feature vector, and then constructs a mapping transformation matrix based on the archive image feature vector, maps and changes the archive image feature vector to obtain the change feature vector, and then calculates the coding weight at each moment through hidden layer iteration, and optimizes the character recognition output result by constructing the loss function. The high-performance character recognition algorithm proposed by the present invention improves the efficiency of character recognition.

[0031] Furthermore, the archive management module automatically classifies and manages the archive text data recognized by the character recognition module according to information categories and segmented cells, wherein the information categories are names of information items in the archive, and an archive text slice library is established.

[0032] Furthermore, the information retrieval module provides an information retrieval box, in which the user enters the name of the information item to be retrieved, and after the retrieval, the information retrieval module displays the results obtained by the user's search for the information item name.

[0033] Furthermore, in the information tracing module, when the user searches for information, the high-definition archive image will be directly called out, and the text searched by the user will be directly highlighted in the corresponding high-definition archive image, locating the information searched by the user and displaying the location of the information.

[0034] Beneficial effects:

[0035] The present invention proposes a method for applying OCR to structured extraction of archival data, comprising an archival data preprocessing module, an ultra-clear reconstruction module, an image segmentation module, a character recognition module, an archival management module, an information retrieval module, and an information tracing module; the ultra-clear reconstruction module proposes an image reconstruction algorithm to perform high-definition reconstruction of the image based on the image pixel value. The image reconstruction algorithm proposed in the present invention performs pixel analysis on the archival image file to construct the archival image file into a pixel matrix, and then obtains the pixel offset feature of the pixel point through an offset operation, and then performs diffusion processing through a diffusion function to perform a diffusion operation on the pixel matrix, and then performs iterative optimization on the archival image through an iterative correction formula to obtain an optimized archival image sequence, and based on the optimized archival image sequence, performs high-definition reconstruction on the archival image through a probability formula, thereby improving the image quality and further improving the efficiency and accuracy of character recognition; the character recognition module proposes a high-performance character recognition algorithm to recognize the text in the archival image. The high-performance character recognition algorithm proposed in the present invention constructs a character recognition model, extracts features of the high-definition archival image in the input layer, and obtains the archival image feature vector. Based on the feature vector of the archive image, a mapping transformation matrix is ​​constructed, and the feature vector of the archive image is mapped and changed to obtain a changed feature vector. Then, through hidden layer iteration, the coding weight at each moment is calculated. By constructing a loss function, the character recognition output result is optimized, thereby improving the efficiency of character recognition. The image can be directly retrieved for information to obtain the required information, and the retrieved information is highlighted in the image, thereby facilitating information tracing and positioning. Compared with the traditional archive management method, which often relies on manual processing, is inefficient and prone to errors, the method of applying OCR to structured extraction of archive data proposed in the present invention realizes automated structured extraction of archive data through OCR technology, thereby greatly improving processing efficiency. Through automated preprocessing, reconstruction, character recognition and other steps, paper archives can be quickly and accurately converted into editable and searchable electronic texts, thereby greatly saving labor costs and time costs. Traditional manual entry methods are prone to errors and omissions, thereby affecting the accuracy of archive data. The present invention utilizes OCR technology to automatically process archive data, thereby reducing interference from human factors and greatly improving data accuracy. Through character recognition, errors can be effectively identified and corrected, ensuring the accuracy and completeness of extracted data; traditional archive management methods require a lot of manpower and material costs, while the automated processing method of the present invention can effectively reduce the cost of archive management, save manpower costs and time costs, and reduce the cost and risk of subsequent processing; in summary, the beneficial effects of the present invention have shown significant advantages and values ​​in improving archive data processing efficiency, improving data accuracy, reducing management costs, improving data utilization value, strengthening data security and promoting the process of modernization of management, providing important technical support and guarantee for digital archive management. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 It is a schematic diagram of the structure of the present invention;

[0038] Figure 2 It is a schematic diagram of the character recognition model of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] In order to achieve the above-mentioned purpose, the present invention proposes a method for applying OCR to structured extraction of archival data, comprising an archival data preprocessing module, an ultra-clear reconstruction module, an image segmentation module, a character recognition module, an archival management module, an information retrieval module, and an information tracing module; the archival data preprocessing module scans and identifies archival files and processes the archival files into image files; the ultra-clear reconstruction module proposes an image reconstruction algorithm and performs high-definition image reconstruction based on image pixel values; the image segmentation module performs tabular segmentation on archival images to form cells; the character recognition module proposes a high-performance character recognition algorithm to recognize archival image text; the archival management module classifies and manages the archival text data recognized by the character recognition module; in the information retrieval module, the user searches by text input to view the required archival information; in the information tracing module, the information retrieved by the user is segmented and highlighted to display the location in the archival image.

[0041] Specifically, the archive data preprocessing module performs high-definition scanning on the archive files through a high-definition scanner to generate high-definition archive image files, and marks the archive image serial numbers of the generated high-definition archive image files according to the scanning order.

[0042] Specifically, the ultra-high definition reconstruction module proposes an image reconstruction algorithm, optimizes the archival image based on the pixel shift characteristics of the pixel values ​​of the high-definition archival image file, and reconstructs the archival image in high definition.

[0043] Specifically, the pixel shift feature is as follows: for high-definition archive pictures, pixel value analysis is first performed, and the high-definition archive pictures are represented by a pixel matrix, as follows:

[0044]

[0045] Among them, B represents the pixel matrix of the high-definition archive picture, b1,1, b1,n, bn,1, bi,j, bn,n represent the pixel value of the 1st row and 1st column of the pixel matrix of the high-definition archive picture, the pixel value of the 1st row and nth column of the pixel matrix, the pixel value of the nth row and 1st column of the pixel matrix, the pixel value of the i-th row and j-th column of the pixel matrix, and the pixel value of the n-th row and n-th column of the pixel matrix, respectively. For the pixel matrix, an offset operation is performed to obtain the pixel offset feature, which represents the feature of each pixel point. The calculation formula is as follows;

[0046]

[0047] Where Δb i,j represents the pixel offset feature of pixel point (i, j), b i+l,k Represents the pixel value of the i+lth row and kth column of the pixel matrix, calculates the pixel offset feature in the neighborhood of the pixel point (i, j), and then based on the pixel offset feature, performs diffusion processing on the pixel matrix of the high-definition archive image. The diffusion processing process is as follows:

[0048]

[0049] Where φ(·) represents the diffusion function, (β i ,β j ) represents the coordinate value of the pixel point (i, j), (α i ,α j ) represents the coordinate value of a point in the neighborhood of pixel point (i, j), Δb i,k The pixel offset feature of the pixel point (i, k) is represented. Through the diffusion function, each pixel point is used as the base point to diffuse the pixel matrix to obtain the basic archive image. Based on the basic archive image, the optimized archive image is obtained through iterative correction. The iterative correction formula is as follows:

[0050] g t =g t-1 +H t ×(g t-1 -g 0 ),

[0051] Among them, g 0 represents the basic archive image, g t-1 represents the optimized archive image at stage t-1, g t represents the optimized archive image at stage t, Ht Represents the iterative optimization matrix of stage t. Based on all optimized archive images, the optimized archive image sequence G is constructed, which is expressed as follows:

[0052] G=[g 0 ,g 1 ,…,g i ,…,g t ],

[0053] Among them, g 1 , g i They represent the optimized archive image of the first stage and the optimized archive image of the i-th stage respectively. Then, a reconstruction probability method is proposed. Based on the optimized archive image sequence, the reconstructed archive image is obtained. The formula is as follows:

[0054]

[0055] Where Y represents the reconstructed archive image, L represents the final output of the archive image reconstruction, and p(·) represents the probability of the image. The final reconstructed archive image is obtained by optimizing the archive image sequence.

[0056] Furthermore, the image segmentation module performs tabular segmentation on the ultra-high-resolution reconstructed image file based on the UNet segmentation model, and divides the archive image into cells, with all the archive text contained in the cells.

[0057] Specifically, the character recognition module proposes a high-performance character recognition algorithm and performs character recognition on archival image files by constructing a character recognition model.

[0058] Specifically, the character recognition model has the following detailed process: the character recognition model includes an input layer, a hidden layer, and a prediction layer. The reconstructed archive image outputted by the ultra-clear reconstruction module is inputted into the input layer of the character recognition model. First, feature extraction is performed to obtain the archive image feature vector. The archive image feature vector is used to construct a feature sequence D, which is expressed as:

[0059] D = [χ 1 ,x 2 ,…,χ t ,…,χ n ],

[0060] Among them, χ 1 , χ 2 , χ t , χ n Respectively represent the first archival image feature vector, the second archival image feature vector, the tth archival image feature vector, and the nth archival image feature vector. Based on the archival image feature vector, a mapping transformation matrix is ​​constructed to transform the archival image feature vector χ tPerform mapping transformation to obtain the change feature vector χ t ′, the formula is as follows:

[0061]

[0062] Among them, S represents the feature network step size, θ x,1 ,θ y,1 ,θ x,t ,θ y,t ,θ x,n ,θ y,n They represent the archival image feature vector χ 1 The horizontal and vertical coordinates of the position of the corresponding regional center in the feature map, the feature vector χ t The horizontal and vertical coordinates of the position of the corresponding regional center in the feature map, the feature vector χ n The horizontal and vertical coordinates of the position of the corresponding area center in the feature map are transformed into feature vectors as the output of the input layer of the character recognition model, which are input to the hidden layer for iteration. When the iteration reaches the k stage, the hidden layer encoding weight is calculated as follows:

[0063]

[0064] Among them, χ′ t,k represents the k-stage χ t The changing characteristic vector, r t,k represents the hidden layer state at time k, represents the iterative weight matrix, Represents the k-stage change feature vector χ′ t,k The hidden layer encoding weights are used to output the predicted character vector in the output layer, and the loss function is constructed to optimize the prediction training results. The loss function F formula is as follows:

[0065]

[0066] Among them, y k represents the character recognition output of stage k, ||·|| 2 Represents the bi-norm, and gradually optimizes the character recognition prediction results through the loss function.

[0067] Specifically, the archive management module automatically classifies and manages the archive text data recognized by the character recognition module according to information categories and segmented cells. The information categories are the names of information items in the archives, and an archive text slice library is established.

[0068] Specifically, the information retrieval module provides an information retrieval box, in which the user enters the name of the information item to be retrieved. After the retrieval, the information retrieval module displays the results obtained by the user's search for the information item name.

[0069] Specifically, in the information tracing module, when the user searches for information, the high-definition archive image will be directly called out, and the text searched by the user will be directly highlighted in the corresponding high-definition archive image, locating the information searched by the user and displaying the location of the information.

Claims

1. A method for applying OCR to structured extraction of archival data, comprising an archival data preprocessing module, an ultra-clear reconstruction module, an image segmentation module, a character recognition module, an archival management module, an information retrieval module, and an information tracing module; the archival data preprocessing module scans and recognizes archival files and processes the archival files into image files; the ultra-clear reconstruction module proposes an image reconstruction algorithm and performs high-definition image reconstruction based on image pixel values; the image segmentation module performs tabular segmentation on archival images to form cells; the character recognition module proposes a high-performance character recognition algorithm to recognize archival image text; the archival management module classifies and manages archival text data recognized by the character recognition module; in the information retrieval module, users search by text input to view the required archival information; in the information tracing module, the information retrieved by the user is segmented and highlighted to display the location in the archival image; The archive data preprocessing module performs high-definition scanning on the archive files through a high-definition scanner to generate high-definition archive image files, and marks the archive image sequence numbers of the generated high-definition archive image files according to the scanning order; The ultra-high definition reconstruction module proposes an image reconstruction algorithm, optimizes the archive image based on the pixel shift characteristics of the pixel values ​​of the high-definition archive image file, and reconstructs the archive image in high definition; The pixel offset feature is specifically as follows: for high-definition archive pictures, pixel value analysis is first performed, and the high-definition archive pictures are represented by a pixel matrix. For the pixel matrix, an offset operation is performed to obtain a pixel offset feature, which represents the features of each pixel point. The calculation formula is as follows; in, Δb i,j represents the pixel offset feature of pixel point (i, j), b i,j represents the pixel value of the i-th row and j-th column of the pixel matrix, b i+l,k Represents the pixel value of the i+lth row and kth column of the pixel matrix, calculates the pixel offset feature in the neighborhood of the pixel point (i, j), and then based on the pixel offset feature, performs diffusion processing on the pixel matrix of the high-definition archive image. The diffusion processing process is as follows: Where φ(·) represents the diffusion function, (β i ,β j ) represents the coordinate value of the pixel point (i, j), (α i ,α j ) represents the coordinate value of a point in the neighborhood of pixel point (i, j), Δb i,k The pixel offset feature of the pixel point (i, k) is represented. Through the diffusion function, each pixel point is used as the base point to diffuse the pixel matrix to obtain the basic archive image. Based on the basic archive image, the optimized archive image is obtained through iterative correction. The iterative correction formula is as follows: g t =g t-1 +H t ×(g t-1 -g0), Among them, g0 represents the basic archive image, g t-1 represents the optimized archive image at stage t-1, g t represents the optimized archive image at stage t, H t Represents the iterative optimization matrix of stage t. Based on all optimized archival images, the optimized archival image sequence G is constructed. Then, through the reconstruction probability method, the reconstructed archival image is obtained based on the optimized archival image sequence. The formula is as follows: Among them, gi represents the optimized archival image of the i-th stage, Y represents the reconstructed archival image, L represents the final output of the archival image reconstruction, and p(·) represents the probability of the image. The final reconstructed archival image is obtained by optimizing the archival image sequence.

2. The method of applying OCR to structured extraction of archival data according to claim 1, characterized in that: The image segmentation module, based on the UNet segmentation model, performs tabular segmentation on the ultra-high-resolution reconstructed image file, and divides the archive image into cells, with all the archive text contained in the cells.

3. The method of applying OCR to structured extraction of archival data according to claim 1, characterized in that: The character recognition module proposes a high-performance character recognition algorithm and performs character recognition on archival image files by constructing a character recognition model.

4. The method of applying OCR to structured extraction of archival data according to claim 3, characterized in that: The character recognition model has the following detailed process: the character recognition model includes an input layer, a hidden layer, and a prediction layer. The reconstructed archive image outputted by the ultra-clear reconstruction module is inputted into the input layer of the character recognition model. First, feature extraction is performed to obtain the archive image feature vector. The archive image feature vector is used to construct a feature sequence D, which is expressed as: D = [χ1, χ2, ..., χ t ,…,χ n ], where χ1, χ2, χ t , χ n Respectively represent the first archival image feature vector, the second archival image feature vector, the tth archival image feature vector, and the nth archival image feature vector. Based on the archival image feature vector, a mapping transformation matrix is ​​constructed to transform the archival image feature vector χ t Perform mapping transformation to obtain the change feature vector χ′ t , the formula is as follows: Among them, S represents the feature network step size, θ x,1 ,θ y,1 ,θ x,t ,θ y,t ,θ x,n ,θ y,n The horizontal and vertical coordinates of the position of the regional center corresponding to the archive image feature vector χ1 in the feature map, the feature vector χ t The horizontal and vertical coordinates of the position of the corresponding regional center in the feature map, the feature vector χ n The horizontal and vertical coordinates of the position of the corresponding area center in the feature map are transformed into feature vectors as the output of the input layer of the character recognition model, which are input to the hidden layer for iteration. When the iteration reaches the k stage, the hidden layer encoding weight is calculated as follows: Among them, χ′ t,k represents the k-stage χ t The changing characteristic vector, r t,k represents the hidden layer state at time k, represents the iterative weight matrix, Represents the k-stage change feature vector χ′ t,k The hidden layer encoding weights are used to output the predicted character vector in the output layer, and the loss function is constructed to optimize the prediction training results. The loss function F formula is as follows: Among them, y k represents the character recognition output of stage k, ||·||2 represents the two-norm, and the character recognition prediction results are gradually optimized through the loss function.

5. The method of applying OCR to structured extraction of archival data according to claim 1, characterized in that: The file management module automatically classifies and manages the file text data recognized by the character recognition module according to the information category and the segmented cells. The information category is the name of the information item in the file, and an file text slice library is established.

6. The method of applying OCR to structured extraction of archival data according to claim 1, characterized in that: The information retrieval module provides an information retrieval box, in which the user enters the name of the information item to be retrieved. After the retrieval, the information retrieval module displays the results obtained by the user's search for the information item name.

7. The method of applying OCR to structured extraction of archival data according to claim 1, characterized in that: The information tracing module allows the user to directly call up a high-definition archive image when searching for information. The text searched by the user is directly highlighted in the segmentation table in the corresponding high-definition archive image, and the information searched by the user is located, and the segmentation table where the information is located is displayed.

Images