Data input identification method and device, equipment and storage medium

Abstract

The invention relates to the field of computers, and discloses a data input identification method and device, equipment and a storage medium. The method comprises the following steps: acquiring recorded agricultural batch market data, wherein the agricultural batch market data is obtained by recording on paper with a dot matrix through a dot matrix pen and inputting and caching into the dot matrix pen; acquiring an image corresponding to the table data in the agricultural batch market data through a camera and a preset computer vision algorithm; through a Cycle-Center Net model, carrying out positioning to obtain row and column coordinates of cell contents in the image; obtaining row and column information in the image through a recurrent neural network, and performing identification according to row and column coordinates and the row and column information through a Transform model to obtain cell contents; and formatting the contents of the cells according to the row and column information and outputting the formatted contents to a terminal interface. According to the embodiment of the invention, intelligent identification of the agricultural batch market data handwritten by the user can be realized.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a data input recognition method, device, equipment and storage medium. Background Art

[0002] In daily life or work, in many occasions, users generally rely on stylus to write on paper documents, but paper documents have a certain storage period, which is not conducive to long-term storage and has great inconvenience in subsequent query processes.

[0003] Traditional agricultural wholesale market ERP systems rely on frequent user-system interactions to complete the management tasks of purchase, sales and inventory, while most merchants are accustomed to using paper and pen to record purchase and sales data in agricultural wholesale markets or agriculture-related fields. This creates a contradiction. Traditional ERP cannot automatically enter paper and pen data into the system to achieve intelligent recognition of user handwritten content. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem of how to realize intelligent recognition of user's handwritten agricultural wholesale market data.

[0005] A first aspect of the present invention provides a data input recognition method, the data input recognition method comprising:

[0006] Acquire recorded agricultural wholesale market data, the agricultural wholesale market data is obtained by recording on paper with a dot matrix by a dot matrix pen and inputting the data into a buffer in the dot matrix pen;

[0007] Obtaining images corresponding to the table data in the agricultural wholesale market data through a camera and a preset computer vision algorithm;

[0008] Using the Cycle-CenterNet model, the row and column coordinates of the cell content in the image are located;

[0009] Obtaining row and column information in the image through a recurrent neural network, and identifying cell content through a Transformer model based on the row and column coordinates and the row and column information;

[0010] According to the row and column information, the cell content is formatted and output to a terminal interface.

[0011] Optionally, in a first implementation of the first aspect of the present invention, locating the row and column coordinates of the cell content in the image by using the Cycle-CenterNet model includes:

[0012] Using the Cycle-CenterNet model, the center point of the cell in the image is located;

[0013] According to the center point, the row and column coordinates of the cell content in the image are located and obtained.

[0014] Optionally, in a second implementation of the first aspect of the present invention, locating the row and column coordinates of the cell content in the image according to the center point includes:

[0015] According to the center point, the vertices of the cell are calculated by regression;

[0016] Generate a bounding box corresponding to the cell according to the vertices of the cell;

[0017] The row and column coordinates of the cell content in the image are located and obtained according to the bounding box.

[0018] Optionally, in a third implementation of the first aspect of the present invention, obtaining row and column information in the image through a recurrent neural network, and identifying cell content through a Transformer model based on the row and column coordinates and the row and column information includes:

[0019] Obtaining row and column information in the image through a recurrent neural network;

[0020] According to the row and column coordinates, the cells in the image are clipped;

[0021] According to the clipped cells and the row and column information, the cell content is identified through the Transformer model.

[0022] Optionally, in a fourth implementation of the first aspect of the present invention, identifying the cell content through a Transformer model based on the clipped cell and the row and column information includes:

[0023] According to the trimmed cells and the row and column information, a query matrix, a key matrix and a value matrix are obtained through the Transformer model. The query matrix is ​​used to specify the words of interest, the key matrix is ​​used to measure the similarity between the cell content features input by the model and the query matrix, and the value matrix is ​​used to calculate the final output result of the model according to the similarity;

[0024] Calculating the similarity between the query matrix and the key matrix;

[0025] The cell content is identified based on the similarity between the query matrix and the key matrix, and the value matrix.

[0026] Optionally, in a fifth implementation of the first aspect of the present invention, identifying the cell content according to the similarity between the query matrix and the key matrix, and the value matrix includes:

[0027] The similarity is normalized by a softmax function to obtain an attention weight;

[0028] Calculate the output of the attention mechanism according to the attention weight and the value matrix;

[0029] According to the output of the attention mechanism, the cell content is identified.

[0030] Optionally, in a sixth implementation of the first aspect of the present invention, formatting the cell content and outputting it to a terminal interface according to the row and column information includes:

[0031] According to the row and column information, obtaining association information between the clipped cells;

[0032] Obtaining cell structure information according to the associated information, and formatting the cell content according to the cell structure information and outputting it to a terminal interface;

[0033] Through the terminal interface, determine whether the cell content is correct, and if not, obtain the modified cell content.

[0034] A second aspect of the present invention provides a data input identification device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the data input identification device executes the above-mentioned data input identification method.

[0035] A third aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned data input recognition method.

[0036] In an embodiment of the present invention, recorded agricultural wholesale market data is obtained, and the agricultural wholesale market data is obtained by recording on paper with a dot matrix with a dot matrix pen and inputting the data into the cache in the dot matrix pen; an image corresponding to the table data in the agricultural wholesale market data is obtained by a camera and a preset computer vision algorithm; the row and column coordinates of the cell content in the image are located by the Cycle-CenterNet model; the row and column information in the image is obtained by a recurrent neural network, and the cell content is identified by the Transformer model according to the row and column coordinates and the row and column information; the cell content is formatted and output to the terminal interface according to the row and column information. In the present invention, the dot matrix and the Transformer natural language processing model are used to realize the intelligent recognition of the user's handwritten content, which not only greatly improves the operating efficiency and reduces the errors of manual input, but also improves the user experience, making the purchase, sales and inventory management more intuitive and convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of an embodiment of a data input recognition method in an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of an embodiment of a data input recognition device in an embodiment of the present invention;

[0039] Figure 3 It is a schematic diagram of an embodiment of a data input recognition device in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] Embodiments of the present invention provide a data input recognition method, device, equipment and storage medium.

[0041] The embodiments disclosed in the present invention will be described in more detail below with reference to the accompanying drawings. Although certain embodiments disclosed in the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments disclosed in the present invention are only for exemplary purposes and are not intended to limit the scope of protection disclosed in the present invention.

[0042] In the description of the embodiments disclosed in the present invention, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0043] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the data input recognition method in the embodiment of the present invention includes:

[0044] S100, obtaining recorded agricultural wholesale market data.

[0045] In this embodiment, a dot matrix pen is used to hand-write agricultural wholesale market data on paper with a dot matrix, wherein the agricultural wholesale market data includes the purchase and sale data of the agricultural wholesale market or agricultural related fields, and the dot matrix pen can be any dot matrix pen device using dot matrix technology. A specific area to be written will be set on the paper with a dot matrix, and the position information of the area to be written and the corresponding paper information are encoded in a manner similar to binary encoding, and then the corresponding matrix points are drawn based on a specific algorithm to obtain special paper. All handwritten information will be cached in the dot matrix pen. When the dot matrix pen is connected to the terminal via Bluetooth, these data will be synchronized to the application on the terminal, and then the recorded agricultural wholesale market data will be obtained through the application, wherein the terminal can be but not limited to various personal computers, laptops, smart phones, tablet computers, etc. The dot matrix algorithm is a technology in computer graphics, which can convert continuous graphics, text, etc. into images composed of pixels according to certain rules. It is widely used in computer screens, printers, image processing and other fields. The dot matrix algorithm can accurately identify and convert handwritten content. The dot matrix calculation is mainly based on the geometric shape of graphics, text, etc. and the distance between pixels to determine the position of each pixel.

[0046] S200, obtains images corresponding to the table data in the agricultural wholesale market data through a camera and a preset computer vision algorithm.

[0047] In this embodiment, after the application obtains the synchronized agricultural wholesale market data, it uses a high-precision camera and computer vision algorithm to capture some clear dot matrix data that is invisible to the naked eye, analyze the dot matrix data, and faithfully restore the handwritten notes. Specifically, the high-precision camera clearly captures the images corresponding to the agricultural wholesale market data, and the computer vision technology can process and analyze these images to extract the images corresponding to the table data in the agricultural wholesale market data.

[0048] S300, through the Cycle-CenterNet model, locates the row and column coordinates of the cell content in the image.

[0049] In this embodiment, the Cycle-CenterNet model is a computer vision model that combines a recurrent neural network (RNN) and a CenterNet architecture, which is specifically used to detect and identify objects in images, especially tables and texts. The image corresponding to the tabular data in the agricultural wholesale market data is input into the Cycle-CenterNet model, which accurately locates and cuts out the table part containing the data by predicting the center point of the object. First, the input image is preprocessed, including operations such as denoising, contrast enhancement, and color correction to improve the clarity and readability of the table area; then, CenterNet is used to detect the table, that is, to locate the table area that may exist in the image. CenterNet is a network model based on single-point target detection, which can efficiently identify tables; then the key points in the table area are detected to locate the row and column coordinates of the cell content in the image.

[0050] S400, obtaining row and column information in the image through a recurrent neural network, and identifying the cell content through a Transformer model based on the row and column coordinates and the row and column information.

[0051] In this embodiment, after the row and column coordinates of the cell content in the image are located by the Cycle-CenterNet model, Cycle-CenterNet then uses a recurrent neural network (RNN) to process the spatiotemporal information of the object. RNN is particularly suitable for processing sequence data, so when processing objects with row and column structures such as tables, it can better understand and extract the contextual information of the table data. By combining RNN, Cycle-CenterNet can accurately detect and identify the table data in the image, providing accurate input for the subsequent handwriting recognition steps. Then, the row and column coordinates and row and column information are input into the Transformer model to identify the cell content, wherein the Transformer model is a deep learning model based on a self-attention mechanism, which can process sequence data and capture long-distance dependencies in the sequence, and the Transformer model ensures the fluency and accuracy of the recognition process. The present invention combines the Transformer model through a dot matrix algorithm, so that the system can understand and process complex agricultural wholesale market data, such as commodity names, quantities, prices and supplier information, thereby realizing intelligent purchase, sales and inventory management.

[0052] S500, formats the cell content according to the row and column information and outputs it to the terminal interface.

[0053] In this embodiment, the identified cell contents and row and column information are summarized to obtain a table, which is then output to a terminal interface.

[0054] In an optional implementation of the first aspect of the present invention, locating the row and column coordinates of the cell content in the image by using the Cycle-CenterNet model includes:

[0055] The Cycle-CenterNet model is used to locate the center point of the cell in the image; based on the center point, the row and column coordinates of the cell content in the image are located.

[0056] In this embodiment, when processing handwritten data, the Cycle-CenterNet model first uses CenterNet to locate the center point of the cell in the image, and then uses the center point to input the model into the image output image to obtain the physical coordinates of the concatenated table cells. Specifically, the model outputs the coordinates of the four vertices of the cell, and outputs the coordinates of each point in a clockwise order, which are (x1, y1)(x2, y2)(x3, y3)(x4, y4).

[0057] In an optional implementation of the first aspect of the present invention, locating the row and column coordinates of the cell content in the image according to the center point includes:

[0058] According to the center point, the vertices of the cell are calculated by regression; according to the vertices of the cell, the bounding box corresponding to the cell is generated; according to the bounding box, the row and column coordinates of the cell content in the image are located.

[0059] In this embodiment, CenterNet is used to locate the center point of the cell in the image, and then the four vertices of the cell are calculated by regression based on the center point, and a bounding box is generated for each cell object to locate the row and column coordinates of the content in the cell. This method is particularly effective for detecting tables of different sizes and shapes because it does not require pre-definition of a fixed bounding box shape. Specifically, the model is in a bottom-up manner: based on the cell center point, the distance to the four vertices is regressed and the cell bbox is decoded; at the same time, based on the cell vertex, the distance to the center point of the cell sharing the vertex is regressed, the gbox is decoded, and based on the gbox, the discrete bboxes are spliced ​​together to obtain an accurate and complete table; splicing changes the cell from "discrete" to "continuous", so the row and column information of the cell content is obtained by a post-processing algorithm.

[0060] In an optional implementation of the first aspect of the present invention, row and column information in the image is obtained through a recurrent neural network, and the cell content is identified through a Transformer model based on the row and column coordinates and the row and column information, including:

[0061] The row and column information in the image is obtained through a recurrent neural network; the cells in the image are cropped according to the row and column coordinates; and the cell content is identified and obtained through the Transformer model based on the cropped cells and row and column information.

[0062] In this embodiment, the row and column information in the table image is extracted through a recurrent neural network, and the cell position is located according to the row and column coordinates to crop the cells in the image, wherein the cell cropping is to identify the handwritten content in the cell, and the row and column information is to determine the association information between the cropped cells. The cropped cells and row and column information are input into the Transformer model, and the cell content is identified through the self-attention mechanism of the Transformer model. The model will perform self-learning based on the collected data, and will eventually obtain a recognition model dedicated to the agricultural wholesale market.

[0063] In an optional implementation of the first aspect of the present invention, according to the clipped cell and row and column information, the cell content is identified by the Transformer model, including:

[0064] According to the trimmed cell and row and column information, the query matrix, key matrix and value matrix are obtained through the Transformer model. The query matrix is ​​used to specify the words of interest, the key matrix is ​​used to measure the similarity between the cell content features of the model input and the query matrix, and the value matrix is ​​used to calculate the final output result of the model based on the similarity; calculate the similarity between the query matrix and the key matrix; identify the cell content based on the similarity between the query matrix and the key matrix, as well as the value matrix.

[0065] In this embodiment, the mathematical formula of the attention mechanism of the Transformer model is:

[0066]

[0067] In the expression, Q, K and V refer to the query, key and value matrices respectively; in these three matrices, each row corresponds to a word segment in the input text, and each column corresponds to a feature or dimension of this word segment. Among them, the data in the Query matrix represents the words of interest, and the Query can be regarded as a reference point that needs to be paid attention to and matched; the data in the Key matrix is ​​used to calculate the similarity between words (that is, the attention score), which provides the basis for similarity calculation with the Query. The similarity can be calculated by vector dot multiplication. The Key is used to measure the correlation between the Query and the content features or positions of the input cell; the Value matrix is ​​used to calculate the final output of the model based on the similarity. In short, the Query matrix is ​​used to specify the position or feature of interest, the Key matrix is ​​used to measure the correlation between the Query matrix and each position or feature in the input sequence, and the Value matrix contains the actual information or feature representation.

[0068] In an optional implementation of the first aspect of the present invention, according to the similarity between the query matrix and the key matrix, and the value matrix, identifying the cell content includes:

[0069] The similarity is normalized through the softmax function to obtain the attention weight; the output of the attention mechanism is calculated based on the attention weight and value matrix; and the cell content is identified based on the output of the attention mechanism.

[0070] In this embodiment, different weights are assigned to different parts of the input sequence through the self-attention mechanism, so as to more accurately recognize handwritten text. Specifically, by linearly transforming the input, the Query matrix, Key matrix and Value matrix are obtained respectively. These matrices calculate the similarity between Query and Key through dot product calculation and scaling operation, and normalize the similarity using the softmax function to obtain the attention weight. Then, the final output result is calculated based on the attention weight, where the attention weight can be multiplied by the Value matrix, and finally the weighted Value is summed to obtain the output of the attention mechanism.

[0071] In an optional implementation of the first aspect of the present invention, formatting the cell content and outputting it to the terminal interface according to the row and column information includes:

[0072] According to the row and column information, the association information between the trimmed cells is obtained; according to the association information, the cell structure information is obtained, and according to the cell structure information, the cell content is formatted and output to the terminal interface; through the terminal interface, it is determined whether the cell content is correct, and if not, the modified cell content is obtained.

[0073] In this embodiment, the association information between the clipped cells is determined, such as whether certain cells are located in the same row or the same column. When the results are summarized, they are formatted into structural information for display to the user and output to the terminal interface. The user will get the recognition result in the interface and manually modify it to get the final correct result. The user's modification process will be used as data to fine-tune the recognition model again.

[0074] See also Figure 2 According to a second aspect of the present invention, there is provided a data input recognition device, the data input recognition device comprising:

[0075] The data recording module 10 is used to obtain the recorded agricultural wholesale market data, which is obtained by recording on a paper with a dot matrix by a dot matrix pen and inputting the data into the cache of the dot matrix pen;

[0076] The image acquisition module 20 is used to acquire images corresponding to the table data in the agricultural wholesale market data through a camera and a preset computer vision algorithm;

[0077] The row and column coordinate positioning module 30 is used to locate the row and column coordinates of the cell content in the image through the Cycle-CenterNet model;

[0078] The cell recognition module 40 is used to obtain row and column information in the image through a recurrent neural network, and to identify the cell content through a Transformer model based on the row and column coordinates and the row and column information;

[0079] The data formatting module 50 is used to format the cell content according to the row and column information and output it to the terminal interface.

[0080] In an optional implementation of the second aspect of the present invention, the row and column coordinate positioning module 30 is also used to locate the center point of the cell in the image through the Cycle-CenterNet model; based on the center point, the row and column coordinates of the cell content in the image are located.

[0081] In an optional implementation of the second aspect of the present invention, the row and column coordinate positioning module 30 is also used to calculate the vertices of the cell by regression based on the center point; generate a bounding box corresponding to the cell based on the vertices of the cell; and locate the row and column coordinates of the cell content in the image based on the bounding box.

[0082] In an optional implementation of the second aspect of the present invention, the cell recognition module 40 is also used to obtain row and column information in the image through a recurrent neural network; crop the cells in the image according to the row and column coordinates; and identify the cell content through a Transformer model based on the cropped cells and row and column information.

[0083] In an optional implementation of the second aspect of the present invention, the cell identification module 40 is also used to obtain a query matrix, a key matrix and a value matrix through a Transformer model based on the trimmed cell and row and column information, the query matrix is ​​used to specify the words of interest, the key matrix is ​​used to measure the similarity between the cell content features of the model input and the query matrix, and the value matrix is ​​used to calculate the final output result of the model based on the similarity; calculate the similarity between the query matrix and the key matrix; and identify the cell content based on the similarity between the query matrix and the key matrix, as well as the value matrix.

[0084] In an optional implementation of the second aspect of the present invention, the cell identification module 40 is also used to normalize the similarity through a softmax function to obtain an attention weight; based on the attention weight and value matrix, the output of the attention mechanism is calculated; and based on the output of the attention mechanism, the cell content is identified.

[0085] In an optional implementation of the second aspect of the present invention, the data formatting module 50 is also used to obtain association information between the trimmed cells based on the row and column information; obtain cell structure information based on the association information, and format the cell content based on the cell structure information and output it to the terminal interface; determine whether the cell content is correct through the terminal interface, and if not, obtain the modified cell content.

[0086] Figure 3 1 is a schematic diagram of the structure of a data input recognition device provided by an embodiment of the present invention. The data input recognition device 500 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 (for example, one or more mass storage devices) storing application programs 533 or data 532. Among them, the memory 520 and the storage medium 530 can be short-term storage or permanent storage. The program stored in the storage medium 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the data input recognition device 500. Furthermore, the processor 510 can be configured to communicate with the storage medium 530 to execute a series of instruction operations in the storage medium 530 on the data input recognition device 500.

[0087] Based on the data input recognition device 500, one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input and output interfaces 560, and / or one or more operating systems 531, such as Windows Serve, Mac OS X, Unix, Linux, Free BSD, etc. It can be understood by those skilled in the art that Figure 3 The data input recognition device structure shown does not constitute a limitation on the data input recognition device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0088] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes the steps of the data input recognition method.

[0089] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0090] In addition, although each operation is described in a specific order, this should be understood as requiring such operation to be performed in the specific order shown or in a sequential order, or requiring that all illustrated operations should be performed to obtain desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination mode.

[0091] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A data input recognition method, characterized in that: The data input recognition method comprises: Acquire recorded agricultural wholesale market data, the agricultural wholesale market data is obtained by recording on paper with a dot matrix by a dot matrix pen and inputting the data into a buffer in the dot matrix pen; Obtaining images corresponding to the table data in the agricultural wholesale market data through a camera and a preset computer vision algorithm; Using the Cycle-CenterNet model, the row and column coordinates of the cell content in the image are located; Obtaining row and column information in the image through a recurrent neural network, and identifying cell content through a Transformer model based on the row and column coordinates and the row and column information; According to the row and column information, the cell content is formatted and output to a terminal interface.

2. The data input recognition method according to claim 1, characterized in that: The positioning of the row and column coordinates of the cell content in the image by the Cycle-CenterNet model includes: Using the Cycle-CenterNet model, the center point of the cell in the image is located; According to the center point, the row and column coordinates of the cell content in the image are located and obtained.

3. The data input recognition method according to claim 2, characterized in that: The step of locating the row and column coordinates of the cell content in the image according to the center point comprises: According to the center point, the vertices of the cell are calculated by regression; Generate a bounding box corresponding to the cell according to the vertices of the cell; The row and column coordinates of the cell content in the image are located and obtained according to the bounding box.

4. The data input recognition method according to claim 1, characterized in that: The row and column information in the image is obtained through the recurrent neural network, and the cell content is identified through the Transformer model according to the row and column coordinates and the row and column information, including: Obtaining row and column information in the image through a recurrent neural network; According to the row and column coordinates, the cells in the image are clipped; According to the clipped cells and the row and column information, the cell content is identified through the Transformer model.

5. The data input recognition method according to claim 4, characterized in that: The cell content is identified by the Transformer model according to the clipped cell and the row and column information, including: According to the trimmed cells and the row and column information, a query matrix, a key matrix and a value matrix are obtained through the Transformer model. The query matrix is ​​used to specify the words of interest, the key matrix is ​​used to measure the similarity between the cell content features input by the model and the query matrix, and the value matrix is ​​used to calculate the final output result of the model according to the similarity; Calculating the similarity between the query matrix and the key matrix; The cell content is identified based on the similarity between the query matrix and the key matrix, and the value matrix.

6. The data input recognition method according to claim 5, characterized in that: The identifying cell content according to the similarity between the query matrix and the key matrix, and the value matrix, comprises: The similarity is normalized by a softmax function to obtain an attention weight; Calculate the output of the attention mechanism according to the attention weight and the value matrix; According to the output of the attention mechanism, the cell content is identified.

7. The data input recognition method according to claim 1, characterized in that: The step of formatting the cell content and outputting the cell content to the terminal interface according to the row and column information includes: According to the row and column information, obtaining association information between the clipped cells; Obtaining cell structure information according to the associated information, and formatting the cell content according to the cell structure information and outputting it to a terminal interface; Through the terminal interface, determine whether the cell content is correct, and if not, obtain the modified cell content.

8. A data input recognition device, characterized in that: The data input recognition device comprises: A data recording module, used for obtaining recorded agricultural wholesale market data, wherein the agricultural wholesale market data is obtained by recording on paper with a dot matrix by a dot matrix pen and inputting the data into a buffer in the dot matrix pen; An image acquisition module, used to acquire images corresponding to the table data in the agricultural wholesale market data through a camera and a preset computer vision algorithm; A row and column coordinate positioning module, used to locate and obtain the row and column coordinates of the cell content in the image through a Cycle-CenterNet model; A cell recognition module is used to obtain row and column information in the image through a recurrent neural network, and to identify cell content through a Transformer model based on the row and column coordinates and the row and column information; The data formatting module is used to format the cell content according to the row and column information and output it to the terminal interface.

9. A data input recognition device, characterized in that: The data input recognition device comprises: a memory and at least one processor, the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor calls the instructions in the memory so that the data input identification device executes the data input identification method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the data input recognition method according to any one of claims 1 to 7 is implemented.

Images