Electronic bill storage method and device based on data multi-point association, equipment and medium

Abstract

The invention relates to the technical field of associated storage, and discloses an electronic bill storage method, device and equipment based on data multi-point association and a medium, and the method comprises the steps: extracting region features of a text region in a form picture set corresponding to an electronic data form set; identifying the text content of each form picture according to the region features to obtain an electronic form text; constructing a word co-occurrence matrix and an embedding matrix of the electronic single text, and performing feature learning on the word co-occurrence matrix and the embedding matrix to obtain learning features; performing feature fusion according to the learning features to obtain target features, and calculating an associated form picture of each form picture according to the target features; and constructing an association graph of each electronic data form according to the association form pictures, and storing the electronic data forms in a preset storage space according to the association graphs. According to the invention, the ordering of electronic bill storage can be improved, and the accuracy of electronic bill storage is further improved.

Description

Technical Field

[0001] The present invention relates to the field of associative storage technology, and in particular to an electronic single storage method, device, equipment and medium based on multi-point association of data. Background Art

[0002] An electronic data form is a semi-structured document that is widely used in many fields such as commercial transactions, administrative management, and logistics tracking. Electronic data forms can be HTML forms, PDF forms, etc. With the rapid popularization of the Internet, electronic transactions have become very common. Along with the wave of electronic transactions comes a huge number of electronic data forms. For the same customer or a group of customers, their transactions are related. Therefore, in order to achieve accurate control of customer data and subsequent query, how to store electronic forms in an orderly manner and improve the accuracy of electronic form storage has become an urgent problem to be solved. Summary of the invention

[0003] The present invention provides an electronic document storage method, device, equipment and medium based on multi-point data association, the main purpose of which is to solve the problem of poor accuracy of electronic document association storage.

[0004] To achieve the above-mentioned purpose, the present invention provides an electronic single storage method based on multi-point association of data, comprising:

[0005] Acquire a form image set corresponding to the electronic data form set, detect a text area of ​​each form image in the form image set, and extract regional features of the text area;

[0006] Identify the text content of each of the form images according to the regional features, and obtain the electronic single text of each of the form images;

[0007] Constructing a word co-occurrence matrix and an embedding matrix of the electronic single text, and performing feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix;

[0008] Perform feature fusion according to the learning features to obtain target features, and calculate the associated form images of each form image according to the target features;

[0009] A correlation map of each electronic data form in the electronic data form set is constructed according to the correlation form image, and each electronic data form is stored in a preset storage space according to the correlation map.

[0010] Optionally, detecting the text area of ​​each form image in the form image set includes:

[0011] Performing histogram equalization on each of the form images to obtain a standard form image;

[0012] Binarizing the standard form image using a plurality of preset thresholds to obtain a binary image set;

[0013] The maximum stable extreme value region corresponding to the form image is identified according to the binary image set to obtain the text region of each form image.

[0014] Optionally, extracting the regional features of the text region includes:

[0015] Binarize the text region using the maximum inter-class variance method to obtain a binary region;

[0016] Performing edge detection on the binary area to obtain edge width of the binary area, and calculating stroke features of the text area according to the edge width;

[0017] Calculating a gradient histogram of the text area, and generating edge gradient histogram features of the text area according to the gradient histogram;

[0018] The stroke features and the edge gradient histogram features are collected to obtain the region features of the text region.

[0019] Optionally, the identifying the text content of each of the form images according to the regional features to obtain the electronic single text of each of the form images includes:

[0020] Converting the stroke features in the region features into a stroke feature vector, and performing vector concatenation of the stroke feature vector and the edge gradient histogram feature in the region features to obtain a target feature;

[0021] Extracting a character region of each of the form images according to the target feature;

[0022] Character recognition is performed on the character area to obtain an electronic single text of each form image.

[0023] Optionally, constructing a word co-occurrence matrix and an embedding matrix of the electronic single text includes:

[0024] Performing text segmentation on the electronic single text to obtain a text segmentation set, and calculating the co-occurrence relationship between the text segmentations in the text segmentation set;

[0025] Generate a word co-occurrence matrix of the electronic single text according to the co-occurrence relationship;

[0026] Perform multi-head attention calculation on the text segmentation to obtain matrix elements of the embedding matrix, and construct the embedding matrix according to the matrix elements.

[0027] Optionally, calculating the co-occurrence relationship between text segmentations in the text segmentation set includes:

[0028] Generate a word segmentation sequence according to the text word segmentation set, and use a sliding window with a preset step size to perform sliding division on the word segmentation sequence to obtain window word segmentation of each sliding window;

[0029] Calculate the occurrence frequency of each text segmentation in the window segmentation and the co-occurrence frequency between text segmentations in the text segmentation set;

[0030] Calculate the co-occurrence relationship between the text segment words according to the occurrence frequency and the co-occurrence frequency;

[0031] The co-occurrence relationship between real-time text segmentations is calculated using the following formula:

[0032]

[0033] Among them, A i,j represents the co-occurrence relationship between the i-th text segmentation and the j-th text segmentation, ReLU represents the ReLU activation function, p(i,j) represents the co-occurrence frequency between the i-th text segmentation and the j-th text segmentation, p(i) represents the occurrence frequency of the i-th text segmentation, and p(j) represents the occurrence frequency of the j-th text segmentation.

[0034] Optionally, constructing a correlation map of each electronic data form in the electronic data form set according to the correlation form image includes:

[0035] Determining the associated electronic form of each of the electronic data forms according to the associated form image;

[0036] Constructing a relationship network of each of the electronic data forms according to the associated electronic forms;

[0037] A correlation graph of each electronic data form is generated according to the relationship network.

[0038] In order to solve the above problems, the present invention also provides an electronic single storage device based on multi-point association of data, the device comprising:

[0039] A region feature extraction module is used to obtain a form image set corresponding to the electronic data form set, detect the text region of each form image in the form image set, and extract the region feature of the text region;

[0040] An electronic form text recognition module, used to recognize the text content of each form image according to the regional features, and obtain the electronic form text of each form image;

[0041] A feature learning module, used to construct a word co-occurrence matrix and an embedding matrix of the electronic single text, and perform feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix;

[0042] An associated form image calculation module is used to perform feature fusion according to the learning features to obtain target features, and calculate an associated form image of each form image according to the target features;

[0043] The association storage module is used to construct an association map of each electronic data form in the electronic data form set according to the association form image, and store each electronic data form in a preset storage space according to the association map.

[0044] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0045] at least one processor; and,

[0046] a memory communicatively connected to the at least one processor; wherein,

[0047] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned electronic single storage method based on multi-point association of data.

[0048] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned electronic single storage method based on multi-point association of data.

[0049] The embodiment of the present invention detects the regional features of the text area in each form image, and can further identify the text and non-text in the text area according to the regional features, so as to more accurately identify the electronic single text of each form image; constructs the word co-occurrence matrix and embedding matrix of the electronic single text, and performs feature learning, so as to deeply mine the structural information and semantic information of the electronic single text, and improve the accuracy of subsequent calculation of associated form images; calculates the associated form image according to the target features of the learning feature fusion, and can integrate the feature information in the word co-occurrence matrix and the embedding matrix to more accurately calculate the associated form image; constructs an associated graph according to the associated form image, and stores each electronic data form according to the associated graph, so as to store the electronic data form set in an orderly manner, and improve the accuracy of electronic form storage. Therefore, the electronic form storage method, device, equipment and medium based on multi-point data association proposed by the present invention can solve the problem of poor accuracy of ordered associated storage of electronic forms. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic flow chart of an electronic single storage method based on multi-point data association provided by an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a process for detecting a text area of ​​each form image provided by an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of a process for identifying the text content of each form image according to regional features provided by an embodiment of the present invention;

[0053] Figure 4 A functional module diagram of an electronic single storage device based on multi-point association of data provided by an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of the structure of an electronic device for implementing the electronic single storage method based on multi-point association of data provided by an embodiment of the present invention.

[0055] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0056] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0057] The embodiment of the present application provides an electronic single storage method based on multi-point data association. The execution subject of the electronic single storage method based on multi-point data association includes but is not limited to at least one of the electronic devices such as the server, terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the electronic single storage method based on multi-point data association can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0058] Reference Figure 1 FIG. 1 is a flow chart of an electronic form storage method based on multi-point data association provided by an embodiment of the present invention. In this embodiment, the electronic form storage method based on multi-point data association includes:

[0059] S1. Obtain a form image set corresponding to an electronic data form set, detect a text region of each form image in the form image set, and extract region features of the text region.

[0060] In an embodiment of the present invention, an electronic data form set is a collection of electronic data forms generated by an enterprise in its work, for example, forms generated at each link in the enterprise supply chain, or electronic forms generated in the financial link, including but not limited to a form image set consisting of electronic forms such as purchase forms and payment slips. The electronic data form is a standard format for transmitting business documents between different computer systems, and can be an HTML form or a PDF form. Therefore, the format of electronic form text recognition can be unified according to the form images corresponding to electronic data forms in different forms, so as to more quickly identify the text content in each electronic data form.

[0061] In the embodiment of the present invention, refer to Figure 2 As shown, the detecting of the text area of ​​each form image in the form image set includes:

[0062] S21, performing histogram equalization on each of the form images to obtain a standard form image;

[0063] S22, binarizing the standard form image using a plurality of preset thresholds to obtain a binarized image set;

[0064] S23. Identify the maximum stable extreme value region corresponding to the form image according to the binary image set, and obtain the text region of each form image.

[0065] In the embodiment of the present invention, histogram equalization is performed by adjusting the histogram distribution of the form image so that the pixel intensity value distribution is more uniform, thereby enhancing the local contrast of the image. Specifically, an equalization mapping can be created according to the histogram of the form image, and the equalization mapping can be applied to each pixel of the form image to obtain a standard form image.

[0066] In an embodiment of the present invention, the standard form image is first grayscaled, and then the grayscaled standard form image is binarized using multiple preset thresholds that are incremented from 0 to 255 in sequence to obtain a binarized image corresponding to each threshold. As the threshold increases, the black portion of the binarized image gradually increases, and the black area eventually merges until the entire grayscaled standard form image becomes black. In the binarized image set, some connected areas change very little or even do not change, and this area is the maximum stable extreme value area. At the same time, since the grayscale value of the text area is consistent, and the grayscale value of the text area is quite different from that of the pixels surrounding the text, the text area will not be covered by black when the threshold continues to increase, and will not be covered by black until the threshold increases to the grayscale value of the text area itself. The maximum stable extreme value area is the text area in the form image.

[0067] In the embodiment of the present invention, the text in the electronic data form often has a fixed stroke width, which can be used to distinguish text from non-text in the text area, and at the same time extract the edge gradient histogram features of the text area to obtain the area features in the text area.

[0068] In the embodiment of the present invention, the step of extracting the regional features of the text region includes:

[0069] Binarize the text region using the maximum inter-class variance method to obtain a binary region;

[0070] Performing edge detection on the binary area to obtain edge width of the binary area, and calculating stroke features of the text area according to the edge width;

[0071] Calculating a gradient histogram of the text area, and generating edge gradient histogram features of the text area according to the gradient histogram;

[0072] The stroke features and the edge gradient histogram features are collected to obtain the region features of the text region.

[0073] In an embodiment of the present invention, the maximum inter-class variance method binarization is to select a threshold so that the pixels above the threshold and the pixels below the threshold have the largest intra-class variance. Edge detection by binarizing the area can avoid poor detection accuracy caused by poor contrast in the text area, thereby improving the accuracy of stroke feature extraction. Specifically, the canny detection operator can be used to detect text edges, wherein the ratio of the edge width variance to the square of the edge width mean can be used as a stroke feature.

[0074] In an embodiment of the present invention, the edge gradient histogram feature of the text area is to calculate the gradient direction and gradient amplitude of each edge point in the text area, divide the text area into several intervals (for example, 0 to 360 degrees, one interval every 30 degrees) according to the gradient direction, and then construct a gradient histogram for the gradient amplitude in each interval, and count the frequency of occurrence of the gradient amplitude in each interval. Finally, the gradient histogram can be merged into a feature vector to obtain the edge gradient histogram feature, which is used to represent the gradient feature of the entire text area.

[0075] In the embodiment of the present invention, the text and non-text in the text area can be further identified according to the area features, thereby improving the accuracy of subsequent electronic single text recognition.

[0076] S2. Identify the text content of each of the form images according to the regional features to obtain the electronic single text of each of the form images.

[0077] In the embodiment of the present invention, refer to Figure 3 As shown, the text content of each form image is identified according to the regional features to obtain the electronic single text of each form image, including:

[0078] S31, converting the stroke features in the region features into a stroke feature vector, and performing vector concatenation of the stroke feature vector and the edge gradient histogram feature in the region features to obtain a target feature;

[0079] S32, extracting a character region of each of the form images according to the target feature;

[0080] S33, performing character recognition on the character area to obtain an electronic single text of each form image.

[0081] In an embodiment of the present invention, the text area of ​​each form image can be further classified according to the target features to obtain a character area containing characters, and the character area where each character in the form image is located can be extracted, so that optical character recognition (OCR) can be performed on the characters in the character area to obtain the characters corresponding to each character area, and the characters are combined to obtain the electronic single text of the form image.

[0082] In the embodiment of the present invention, by identifying the electronic form text of each form image, the correlation between different electronic forms can be analyzed according to the content in the electronic form text, thereby improving the accuracy of electronic form storage.

[0083] S3. Construct a word co-occurrence matrix and an embedding matrix of the electronic single text, perform feature learning on the word co-occurrence matrix and the embedding matrix respectively, and obtain learning features of the word co-occurrence matrix and the embedding matrix.

[0084] In the embodiment of the present invention, the word co-occurrence matrix is ​​based on the structural information of the electronic text. The relationship matrix is ​​constructed using the co-occurrence information between words, which can capture the fixed structural information of the electronic text; the embedding matrix is ​​based on the semantic embedding information, and the self-attention method is used to capture the hidden dynamic relationship between words. The structural information and semantic information of the electronic text can be deeply mined through the word co-occurrence matrix and the embedding matrix, thereby improving the accuracy of subsequent association form image calculation.

[0085] In the embodiment of the present invention, the step of constructing the word co-occurrence matrix and embedding matrix of the electronic single text includes:

[0086] Performing text segmentation on the electronic single text to obtain a text segmentation set, and calculating the co-occurrence relationship between the text segmentations in the text segmentation set;

[0087] Generate a word co-occurrence matrix of the electronic single text according to the co-occurrence relationship;

[0088] Perform multi-head attention calculation on the text segmentation to obtain matrix elements of the embedding matrix, and construct the embedding matrix according to the matrix elements.

[0089] In the embodiment of the present invention, the calculating of the co-occurrence relationship between the text segmentations in the text segmentation set includes:

[0090] Generate a word segmentation sequence according to the text word segmentation set, and use a sliding window with a preset step size to perform sliding division on the word segmentation sequence to obtain window word segmentation of each sliding window;

[0091] Calculate the occurrence frequency of each text segmentation in the window segmentation and the co-occurrence frequency between text segmentations in the text segmentation set;

[0092] The co-occurrence relationship between the text segment words is calculated according to the occurrence frequency and the co-occurrence frequency.

[0093] In an embodiment of the present invention, a sliding window with a step size of 1 can be slid from left to right of the word segmentation sequence to obtain the window word segmentations contained in each sliding window, and the frequency of each text word segmentation appearing alone in the window word segmentation and the frequency of two text words appearing together in the window word segmentation are calculated to obtain the occurrence frequency and co-occurrence frequency.

[0094] Specifically, the co-occurrence relationship between real-time text segmentations is calculated using the following formula:

[0095]

[0096] Among them, A i,j represents the co-occurrence relationship between the i-th text segmentation and the j-th text segmentation, ReLU represents the ReLU activation function, p(i,j) represents the co-occurrence frequency between the i-th text segmentation and the j-th text segmentation, p(i) represents the occurrence frequency of the i-th text segmentation, and p(j) represents the occurrence frequency of the j-th text segmentation.

[0097] In an embodiment of the present invention, the co-occurrence relationship is used as a matrix element in a word co-occurrence matrix. For example, the co-occurrence relationship between the i-th text segmentation and the j-th text segmentation is the matrix element of the i-th row and j-th column in the word co-occurrence matrix, so that a word co-occurrence matrix can be generated based on the constructed co-occurrence relationship.

[0098] In the embodiment of the present invention, the multi-head attention calculation is performed on the text segmentation to obtain the matrix elements of the embedding matrix, including:

[0099] Calculate the embedding feature of each of the text words;

[0100] Calculate the attention weights between the text segmentations according to the embedding features;

[0101] A multi-head attention calculation is performed on the text segmentation according to the embedding features and the attention weights to obtain matrix elements of the embedding matrix.

[0102] In an embodiment of the present invention, the glove word embedding model can be used to calculate the embedding features, and the attention weight between two text segmentations can be calculated through the embedding features.

[0103] Specifically, the attention weights between the text segmentations are calculated using the following formula:

[0104]

[0105] Among them, B m,n represents the attention weight between the mth text segmentation and the nth text segmentation, softmax represents the softmax activation function, H m represents the embedding feature of the mth text segmentation, W q Represents the preset first weight matrix, H n represents the embedding feature of the nth text segmentation, W k represents the preset second weight matrix, T represents transposition, and d represents the vector distance between the embedding feature of the mth text segmentation and the embedding feature of the nth text segmentation.

[0106] In an embodiment of the present invention, attention calculation can be performed on the embedded features and attention weights through a pre-built multi-head attention mechanism. The calculation of each attention head in the multi-head attention calculation is independent, and the results of multiple heads are spliced ​​to obtain the attention calculation result between the mth text segmentation and the nth text segmentation, that is, the matrix element in the mth row and nth column in the embedding matrix.

[0107] In an embodiment of the present invention, by performing feature learning on the word co-occurrence matrix and the embedding matrix, the neighbor information of each matrix element in the word co-occurrence matrix and the embedding matrix can be aggregated to obtain information-enhanced learning features. Specifically, multi-scale feature convolution can be performed on the word co-occurrence matrix and the embedding matrix according to two independent convolution modules, so that multi-scale feature information can be processed to enrich the feature space and obtain learning features with richer features.

[0108] S4. Perform feature fusion on the learning features to obtain target features, and calculate the associated form images of each form image according to the target features.

[0109] In an embodiment of the present invention, feature fusion is to fuse the learning features corresponding to the word co-occurrence matrix and the embedding matrix into a target feature, so that the target feature includes the learning features corresponding to the word co-occurrence matrix and the embedding matrix at the same time, thereby more accurately classifying the form image and obtaining the associated form image.

[0110] In the embodiment of the present invention, the learning features can be pooled to obtain pooled features, the pooled features can be mapped to one-dimensional feature vectors, and the target features after feature fusion can be obtained by vector concatenation of the one-dimensional feature vectors.

[0111] In an embodiment of the present invention, the learning features obtained after feature learning of the word co-occurrence matrix and the embedding matrix are still graph-structured learning features. Therefore, it is necessary to pool and reduce the dimension of the learning features to obtain a one-dimensional feature vector, so that the one-dimensional feature vectors corresponding to the word co-occurrence matrix and the embedding matrix can be vector-concatenated to obtain the target features.

[0112] In the embodiment of the present invention, the associated form image is a form image corresponding to an electronic data form that has an associated relationship with the form image, for example, a form image corresponding to an electronic data form of the same customer, a form image corresponding to an electronic data form of the same business category, and the like.

[0113] In an embodiment of the present invention, the form image set can be classified according to the target features, wherein the classification result can be different association relationships. For example, the form image set can be classified according to association relationships such as the same customer, the same business, the same time, etc. Specifically, the pre-built fully connected layer can be used for classification, and other form images in the same category as the form image can be used as associated form images.

[0114] In another optional embodiment of the present invention, the step of calculating the associated form image of each form image according to the target feature includes:

[0115] Performing semantic extraction on the target features to obtain a form label for each of the form images;

[0116] The associated form image of each form image is determined according to the form tag.

[0117] In an embodiment of the present invention, the form label may be a form content label contained in each form image, for example, the form content contained in the electronic form such as the customer, business category, business link, etc. of the electronic form. The electronic order forms with the same label may be associated through the form label.

[0118] Specifically, semantic extraction is to analyze the semantic information contained in the target features. By activating the target features, the semantic information corresponding to the target features can be obtained, such as the key content in the electronic forms such as customers, business categories, and business links, and then the form labels of the form images corresponding to each electronic form can be obtained.

[0119] In the embodiment of the present invention, by calculating the associated form images of the form images through the target features, the semantic feature information in the word co-occurrence matrix and the embedding matrix can be integrated to more accurately calculate the associated form images.

[0120] S5. Construct a correlation map of each electronic data form in the electronic data form set according to the correlation form image, and store each electronic data form in a preset storage space according to the correlation map.

[0121] In the embodiment of the present invention, the association graph stores and represents the relationship between form images in the form of a graph, which can intuitively represent the association relationship between form images, store electronic data forms in an orderly manner, and achieve accurate control of customer data and facilitate later inquiries.

[0122] In an embodiment of the present invention, the step of constructing a correlation map of each electronic data form in the electronic data form set according to the correlation form image includes:

[0123] Determining the associated electronic form of each of the electronic data forms according to the associated form image;

[0124] Constructing a relationship network of each of the electronic data forms according to the associated electronic forms;

[0125] A correlation graph of each electronic data form is generated according to the relationship network.

[0126] In an embodiment of the present invention, a relationship network is constructed according to the relationship between each associated electronic form and the electronic data form, and the relationship between each electronic data form in the relationship network is marked according to the relationship, so as to obtain a relationship graph containing the relationship, wherein the electronic data form is the central node in the relationship graph, and the relationship between the associated electronic forms is connected to the central node as an edge, so as to obtain a relationship graph.

[0127] In an embodiment of the present invention, storing each of the electronic data forms in a preset storage space according to the association graph includes:

[0128] Creating a correlation index for each of the electronic data forms according to the correlation map;

[0129] Each of the electronic data forms is stored in a preset storage space according to the associated index.

[0130] In the embodiment of the present invention, the association index is to establish the relationship between each electronic data form and the associated electronic form through the association map to realize the connection and query between the electronic data form and the associated electronic form. The association index can be created by foreign key constraints and executing SQL commands. For example, a foreign key field pointing to the primary key of the associated electronic form can be created in the electronic data form to establish the association between the two electronic forms. For example, if the electronic data form is "Customer A" and the associated electronic form is "Customer B", a foreign key field "Customer B_id" can be created in the electronic data form of "Customer A" to point to the associated electronic form of the electronic data form "Customer A", so that the associated electronic data forms can be associated and stored, which is convenient for subsequent query and retrieval of the associated electronic data forms.

[0131] In the embodiment of the present invention, the associated electronic forms associated with each electronic data form are directly obtained through the associated index, so that the electronic data form set can be stored in an orderly manner, thereby improving the accuracy of electronic form storage.

[0132] like Figure 4 1 is a functional module diagram of an electronic single storage device based on multi-point data association provided by an embodiment of the present invention.

[0133] The electronic single storage device 100 based on multi-point data association of the present invention can be installed in an electronic device. According to the functions to be implemented, the electronic single storage device 100 based on multi-point data association can include a regional feature extraction module 101, an electronic single text recognition module 102, a feature learning module 103, an association form image calculation module 104 and an association storage module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0134] In this embodiment, the functions of each module / unit are as follows:

[0135] The region feature extraction module 101 is used to obtain a form image set corresponding to the electronic data form set, detect the text region of each form image in the form image set, and extract the region feature of the text region;

[0136] The electronic form text recognition module 102 is used to recognize the text content of each form image according to the regional features to obtain the electronic form text of each form image;

[0137] The feature learning module 103 is used to construct a word co-occurrence matrix and an embedding matrix of the electronic single text, and perform feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix;

[0138] The associated form image calculation module 104 is used to perform feature fusion according to the learning features to obtain target features, and calculate the associated form images of each form image according to the target features;

[0139] The association storage module 105 is used to construct an association map of each electronic data form in the electronic data form set according to the association form image, and store each electronic data form in a preset storage space according to the association map.

[0140] In detail, each module described in the electronic single storage device 100 based on multi-point association of data in the embodiment of the present invention is used in the same manner as described above. Figures 1 to 3 The electronic single storage method based on multi-point data association described in the invention has the same technical means and can produce the same technical effects, so it will not be repeated here.

[0141] like Figure 5 , is a schematic diagram of the structure of an electronic device for implementing an electronic single storage method based on multi-point association of data provided by an embodiment of the present invention.

[0142] The electronic device 1 may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as an electronic single storage program based on multi-point data association.

[0143] Among them, in some embodiments, the processor 10 can be composed of an integrated circuit. The processor 10 is the control core (Control Unit) of the electronic device, which uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions of the electronic device and processes data by running or executing programs or modules stored in the memory 11 (for example, executing electronic single storage programs based on multi-point data association, etc.), and calling data stored in the memory 11.

[0144] The memory 11 includes at least one type of readable storage medium, including flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 11 can be used not only to store application software and various data installed in the electronic device, such as the code of the electronic single storage program based on multi-point association of data, but also to temporarily store the data that has been output or is to be output.

[0145] The communication bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0146] The communication interface 13 is used for communication between the electronic device and other devices, and includes a network interface and a user interface.

[0147] The figure only shows an electronic device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0148] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0149] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0150] Acquire a form image set corresponding to the electronic data form set, detect a text area of ​​each form image in the form image set, and extract regional features of the text area;

[0151] Identify the text content of each of the form images according to the regional features, and obtain the electronic single text of each of the form images;

[0152] Constructing a word co-occurrence matrix and an embedding matrix of the electronic single text, and performing feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix;

[0153] Perform feature fusion according to the learning features to obtain target features, and calculate the associated form images of each form image according to the target features;

[0154] A correlation map of each electronic data form in the electronic data form set is constructed according to the correlation form image, and each electronic data form is stored in a preset storage space according to the correlation map.

[0155] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0156] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, which will not be repeated here.

[0157] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An electronic single storage method based on multi-point data association, characterized in that: The method comprises: Acquire a form image set corresponding to the electronic data form set, detect a text area of ​​each form image in the form image set, and extract regional features of the text area; Identify the text content of each of the form images according to the regional features, and obtain the electronic single text of each of the form images; Constructing a word co-occurrence matrix and an embedding matrix of the electronic single text, and performing feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix; Perform feature fusion according to the learning features to obtain target features, and calculate the associated form images of each form image according to the target features; A correlation map of each electronic data form in the electronic data form set is constructed according to the correlation form image, and each electronic data form is stored in a preset storage space according to the correlation map.

2. The electronic single storage method based on multi-point data association as claimed in claim 1, characterized in that: The detecting the text area of ​​each form image in the form image set includes: Performing histogram equalization on each of the form images to obtain a standard form image; Binarizing the standard form image using a plurality of preset thresholds to obtain a binary image set; The maximum stable extreme value region corresponding to the form image is identified according to the binary image set to obtain the text region of each form image.

3. The electronic single storage method based on multi-point data association as claimed in claim 1, characterized in that: The extracting the regional features of the text region comprises: Binarize the text region using the maximum inter-class variance method to obtain a binary region; Performing edge detection on the binary area to obtain edge width of the binary area, and calculating stroke features of the text area according to the edge width; Calculating a gradient histogram of the text area, and generating edge gradient histogram features of the text area according to the gradient histogram; The stroke features and the edge gradient histogram features are collected to obtain the region features of the text region.

4. The electronic single storage method based on multi-point data association as claimed in claim 1, characterized in that: The step of identifying the text content of each form image according to the regional features to obtain the electronic single text of each form image includes: Converting the stroke features in the region features into a stroke feature vector, and performing vector concatenation of the stroke feature vector and the edge gradient histogram feature in the region features to obtain a target feature; Extracting a character region of each of the form images according to the target feature; Character recognition is performed on the character area to obtain an electronic single text of each form image.

5. The electronic single storage method based on multi-point data association as claimed in claim 1, characterized in that: The step of constructing the word co-occurrence matrix and embedding matrix of the electronic single text includes: Performing text segmentation on the electronic single text to obtain a text segmentation set, and calculating the co-occurrence relationship between the text segmentations in the text segmentation set; Generate a word co-occurrence matrix of the electronic single text according to the co-occurrence relationship; Perform multi-head attention calculation on the text segmentation to obtain matrix elements of the embedding matrix, and construct the embedding matrix according to the matrix elements.

6. The electronic single storage method based on multi-point data association as claimed in claim 5, characterized in that: The calculating of the co-occurrence relationship between the text segmentations in the text segmentation set includes: Generate a word segmentation sequence according to the text word segmentation set, and use a sliding window with a preset step size to perform sliding division on the word segmentation sequence to obtain window word segmentation of each sliding window; Calculate the occurrence frequency of each text segmentation in the window segmentation and the co-occurrence frequency between text segmentations in the text segmentation set; Calculate the co-occurrence relationship between the text segment words according to the occurrence frequency and the co-occurrence frequency; The co-occurrence relationship between real-time text segmentations is calculated using the following formula: Among them, A i,j represents the co-occurrence relationship between the i-th text segmentation and the j-th text segmentation, ReLU represents the ReLU activation function, p(i,j) represents the co-occurrence frequency between the i-th text segmentation and the j-th text segmentation, p(i) represents the occurrence frequency of the i-th text segmentation, and p(j) represents the occurrence frequency of the j-th text segmentation.

7. The electronic single storage method based on multi-point data association as claimed in claim 1, characterized in that: The step of constructing a correlation map of each electronic data form in the electronic data form set according to the correlation form image includes: Determining the associated electronic form of each of the electronic data forms according to the associated form image; Constructing a relationship network of each electronic data form according to the associated electronic form; A correlation graph of each electronic data form is generated according to the relationship network.

8. An electronic single storage device based on multi-point data association, characterized in that: The device comprises: A region feature extraction module is used to obtain a form image set corresponding to the electronic data form set, detect the text region of each form image in the form image set, and extract the region feature of the text region; An electronic form text recognition module, used to recognize the text content of each form image according to the regional features, and obtain the electronic form text of each form image; A feature learning module, used to construct a word co-occurrence matrix and an embedding matrix of the electronic single text, and perform feature learning on the word co-occurrence matrix and the embedding matrix respectively to obtain learning features of the word co-occurrence matrix and the embedding matrix; An associated form image calculation module is used to perform feature fusion according to the learning features to obtain target features, and calculate an associated form image of each form image according to the target features; The association storage module is used to construct an association map of each electronic data form in the electronic data form set according to the association form image, and store each electronic data form in a preset storage space according to the association map.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the electronic single storage method based on multi-point association of data as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the electronic single storage method based on multi-point association of data as described in any one of claims 1 to 7 is implemented.

Images