Scanning map place name annotation automatic extraction method and system based on wavelet neural network algorithm

Through the wavelet neural network algorithm, U-NET neural network and convolutional neural network structures are used to remove map noise and decompose image features, solving the problem of inaccurate text recognition in complex map images, and achieving high-accurate text extraction .

CN119007235APending Publication Date: 2024-11-22Chinese People's Liberation Army Cyberspace Force Information Engineering University
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411120749.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

When existing OCR technology processes complex map images, it is difficult to effectively remove background noise and non-uniformly distributed text annotations, resulting in inaccurate recognition results.

Method used

Using a method based on wavelet neural network algorithm, a background filter model is constructed through U-NET neural network to remove noise, and the image is decomposed using wavelet transform, combined with convolutional neural network and encoder decoder structure to extract text features, and pre-trained neural network model Identify polymorphic text.

Benefits of technology

It significantly improves the recognition accuracy and reliability of place name annotations, can process map images of different resolutions and types, has wide adaptability and good application prospects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007235A_ABST
    Figure CN119007235A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of map digitization, in particular to a scanning map place name annotation automatic extraction method and system based on a wavelet neural network algorithm, and the method comprises the steps: inputting a to-be-processed map scanning image into a background filtering model, and removing background noise through connected region analysis; the background filtering model is constructed based on a U-NET neural network; a background-free map image is decomposed into high-frequency information and low-frequency information of different scales and different directions, the resolution of the map image is enhanced through wavelet transform, and character features in the map image are extracted; and inputting the character features into a neural network model to identify polymorphic characters in the map image, the neural network model comprising a convolutional neural network structure for extracting an input image static local feature sequence and an encoder network for encoding, identifying and outputting the image static local feature sequence. The method can be applied to automatic character extraction of a paper or scanned map with relatively large background noise, and is high in recognition precision and strong in practicability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of map digitization, and in particular to a method and system for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm. Background Art

[0002] The common method for extracting text from images is optical character recognition (OCR). OCR has good performance in extracting text from clear, noise-free pictures with neatly arranged text. However, the content of map images is complex. In addition to text annotations, a map image also has other ground features such as points, lines, and surfaces, which affect the clarity and contrast of the text. OCR technology cannot effectively handle the interference of these noises on the recognition results, resulting in unclear edges of the recognized text annotations and other mixed elements besides the annotations. In map images, affected by cartographic synthesis, map annotations are not neatly arranged, but are unevenly distributed, distorted, or tilted depending on the map spots. In order to highlight the important information in the map, the annotations on the same map image usually use different fonts and font sizes, which makes it difficult for OCR technology to fully extract the text features of map annotations in complex environments, ultimately resulting in missed or misidentified map annotations. Summary of the invention

[0003] To this end, the present invention provides a method and system for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm, which solves the problem of inaccurate automatic text extraction from paper or scanned maps with large background noise.

[0004] According to the design scheme provided by the present invention, on the one hand, a method for automatically extracting place name annotations from a scanned map based on a wavelet neural network algorithm is provided, comprising:

[0005] The map scan image to be processed is input into a background filtering model, and background noise in the map scan image is removed by using the background filtering model and connected region analysis to obtain a background-free map image, wherein the background filtering model is a target neural network constructed based on a U-NET neural network and obtained by training with sample data;

[0006] Decomposing the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhancing the resolution of the map image by wavelet transform, and extracting text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information;

[0007] The text features are input into a pre-trained neural network model, and the neural network model is used to identify polymorphic text in the map scan image to be processed. The neural network model includes a convolutional neural network structure for extracting the static local feature sequence of the input image and an encoder network for encoding and identifying the static local feature sequence of the image.

[0008] As the automatic extraction method of scanned map place name annotations based on the wavelet neural network algorithm of the present invention, further, the background filtering model includes an encoder and a decoder, in the encoder, a convolution layer is used to extract input image features, batch normalization and activation functions are added after each convolution layer, a pooling layer is used for downsampling, and the convolution layer and the pooling layer are stacked in each downsampling to expand the feature diversity by increasing the number of feature map channels; in the decoder, a transposed convolution layer is used to perform a transposed convolution operation on the feature map, and the number of feature map channels in the transposed convolution process is consistent with the number of feature map channels in the image feature extraction process, and the two feature map channel numbers are jump-connected to superimpose the feature maps.

[0009] As the automatic extraction method of place name annotations on scanned maps based on the wavelet neural network algorithm of the present invention, further, the training process of the background filter model includes the following contents:

[0010] Constructing a sample data set, wherein the sample data set includes image sample data with ground object annotations;

[0011] The U-NET neural network is trained and optimized using the sample data set by minimizing the loss function, and the trained and optimized U-NET neural network is used as the target neural network to obtain the background filtering model.

[0012] As the automatic extraction method of scanned map place name annotation based on the wavelet neural network algorithm of the present invention, further, the background noise in the map scanned image is removed by using the background filtering model and connected region analysis, comprising:

[0013] The map scanned image is segmented using the background filtering model, and the map scanned image is converted into a grayscale image by calculating the average value of the maximum channel and the minimum channel image pixels;

[0014] The image binarization segmentation threshold is determined using the big law method, and the grayscale image is binarized using the threshold to remove background noise in the map scan image.

[0015] As the automatic extraction method of place name annotations on scanned maps based on the wavelet neural network algorithm of the present invention, further, the map image resolution is enhanced by wavelet transform, comprising:

[0016] Decompose the map image into the wavelet domain to obtain the low-frequency sub-image of the map image;

[0017] The low-frequency sub-image is processed with histogram equalization to stretch the concentrated grayscale distribution interval in the histogram to a balanced distribution in the global range, and the pixel values ​​of the image are redistributed to enhance the overall contrast of the low-frequency sub-image in the map image.

[0018] As the automatic extraction method of place name annotations in scanned maps based on the wavelet neural network algorithm of the present invention, further, extracting text features in the map image includes:

[0019] Use the Sobel operator to calculate the gradient of the map image in the x and y directions, and obtain the gray level co-occurrence matrix of the map image;

[0020] The gray level co-occurrence matrix is ​​used to extract text features from map images.

[0021] As the automatic extraction method of scanned map place name annotation based on wavelet neural network algorithm of the present invention, further, using the neural network model to identify polymorphic text in the map scanned image to be processed, it also includes:

[0022] The text recognition results are matched with a custom dictionary and / or the recognition results are corrected using a language model to correct errors in the text recognition results in the map image.

[0023] On the other hand, the present invention also provides a system for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm, comprising: a denoising module, an enhancement module and a recognition module, wherein:

[0024] A denoising module is used to input the map scan image to be processed into a background filtering model, and remove the background noise in the map scan image by using the background filtering model and connected region analysis to obtain a background-free map image. The background filtering model is a target neural network constructed based on a U-NET neural network and trained using sample data;

[0025] An enhancement module is used to decompose the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhance the resolution of the map image through wavelet transform, and extract text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information;

[0026] The recognition module is used to input text features into a pre-trained neural network model, and use the neural network model to recognize polymorphic text in the map scan image to be processed. The neural network model includes a convolutional neural network structure for extracting a static local feature sequence of the input image and an encoder network for encoding and recognizing the static local feature sequence of the image.

[0027] Beneficial effects of the present invention:

[0028] The present invention establishes a convolutional neural network that is adapted to scanned map noise filtering, and can be adapted to the conversion of place name annotations in paper or scanned map files in high-precision and low-noise environments. By combining multi-angle parallel processing and signal decomposition technology, the accuracy and reliability of place name annotation recognition can be significantly improved, and the place name annotations can be extracted and converted from paper or scanned maps into digital information. The present invention has strong robustness, can process map images of different resolutions and types, has wide adaptability, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the automatic extraction process of place name annotations on scanned maps based on the wavelet neural network algorithm in the embodiment;

[0030] Figure 2 Schematic diagram of the map image text extraction result in the embodiment. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and technical solutions.

[0032] In view of the fact that the existing map text recognition is prone to misrecognition or missed recognition in complex background scenes, resulting in low accuracy of extraction results, high requirements for image quality and text annotation features, and difficulty in processing map data of different types and qualities, the embodiments of the present invention refer to Figure 1 As shown, a method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm is provided, comprising:

[0033] S101. Input the map scan image to be processed into a background filtering model, and use the background filtering model and connected region analysis to remove background noise in the map scan image to obtain a background-free map image. The background filtering model is a target neural network constructed based on a U-NET neural network and trained with sample data.

[0034] Among them, the background filtering model includes an encoder and a decoder. In the encoder, a convolution layer is used to extract input image features, batch normalization and activation functions are added after each convolution layer, and a pooling layer is used for downsampling. In each downsampling, the convolution layer and the pooling layer are stacked to expand the feature diversity by increasing the number of feature map channels. In the decoder, a transposed convolution layer is used to perform a transposed convolution operation on the feature map, and the number of feature map channels in the transposed convolution process is consistent with the number of feature map channels in the image feature extraction process, and the two feature map channels are jump-connected to superimpose the feature maps.

[0035] The U-NET neural network model is established using the U-NET neural network structure to eliminate the complex background noise of the paper scanned map image. Specifically, the encoder sets the convolution layer and the pooling layer. The convolution layer is responsible for extracting features from the processed samples, increasing the depth and receptive field of the model. Batch normalization and activation functions are added after each convolution layer to cope with nonlinear operations. The pooling layer is used for downsampling to reduce the size of the feature map. In each downsampling, the stacked convolution layer and the pooling layer are responsible for increasing the number of feature map channels to increase the diversity of features. The decoder sets the transposed convolution layer, the convolution kernel size is set to 3*3, and an LRelu activation function is used after each convolution. The convolution step is 1, the maximum pooling size is 2*2, and the transposed convolution kernel size is 2*2. After the pooling operation, the feature map size changes. After the transposed convolution, the feature map size becomes larger, and the number of channels will be expanded to twice that before the transposed convolution. The number of feature map channels is equal to the number of feature map channels in the feature extraction process symmetrical to it. At this time, a jump connection operation is performed to superimpose the feature maps and supplement the shallow information. Add a convolutional layer to the last layer of the decoder to output the single-task result. Use the labeled training set to train the model and adjust the model parameters by minimizing the loss function.

[0036] Among them, the training process of the background filter model includes the following contents:

[0037] Constructing a sample data set, wherein the sample data set includes image sample data with ground object annotations;

[0038] The U-NET neural network is trained and optimized using the sample data set by minimizing the loss function, and the trained and optimized U-NET neural network is used as the target neural network to obtain the background filtering model.

[0039] The trained model is used to process the data, combining threshold processing technology and connected area analysis technology to eliminate the map image background.

[0040] Specifically, using a background filtering model and removing background noise in a map scan image through connected region analysis may include:

[0041] The map scanned image is segmented using the background filtering model, and the map scanned image is converted into a grayscale image by calculating the average value of the maximum channel and the minimum channel image pixels;

[0042] The image binarization segmentation threshold is determined using the big law method, and the grayscale image is binarized using the threshold to remove background noise in the map scan image.

[0043] For map images with complex backgrounds and noise, the optimal network depth and width are adjusted through model training to adapt to the complex structures and details in the image, and the color image is converted into a grayscale image by calculating the average of the maximum and minimum channels. The Otus method is used to traverse all possible thresholds in the world, calculate the intra-class variance under each threshold, and select the threshold with the smallest intra-class variance as the final threshold for image binarization to separate the background from the content information in the image.

[0044] S102, decomposing the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhancing the resolution of the map image through wavelet transform, and extracting text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information.

[0045] Wavelet transform is applied to the image without background, and the image is decomposed into high-frequency and low-frequency information of different scales and directions. The low-frequency information contains most of the information in the image, and the high-frequency part includes the edge and contour information in the image. The low-frequency information is processed by the histogram equalization method, and the high-frequency subband coefficient is used to process the high-frequency information. φ(x) and τ(x) are used to analyze the horizontal direction, and f(x,y) is decomposed into two parts: smooth approximation and approximate details. Then, φ(x) and τ(x) are used to perform similar analysis in the other direction. After the above wavelet frequency division, four sub-frequency images are obtained, namely, low frequency in both horizontal and vertical directions, low frequency in horizontal directions and high frequency in vertical directions, high frequency in horizontal directions and low frequency in vertical directions, and high frequency in horizontal directions and high frequency in vertical directions. The low-frequency parts are iteratively subjected to wavelet transform to obtain fine sub-frequency images. After completing the multi-level wavelet decomposition, the low-frequency sub-image of the original image is obtained. The low-frequency sub-image is subjected to histogram equalization, and the relatively concentrated gray distribution interval in the gray histogram is stretched to a balanced distribution in the global range, and the pixel values ​​of the image are redistributed to enhance the overall contrast of the low-frequency sub-image.

[0046] The Sobel operator can be used to calculate the gradient of the map image in the x and y directions, and the gray level co-occurrence matrix of the map image can be obtained at the same time; the gray level co-occurrence matrix can be used to extract the text features in the map image.

[0047] S103, inputting text features into a pre-trained neural network model, and using the neural network model to identify polymorphic text in the map scan image to be processed, wherein the neural network model includes a convolutional neural network structure for extracting a static local feature sequence of an input image and an encoder network for encoding and identifying the static local feature sequence of the image.

[0048] Based on the image convolution features, through the input of multi-font size samples, and according to the inconsistent characteristics of the annotation direction of the paper image, the sample is rotated to increase the sample capacity, and then a new convolutional neural network model is formed to extract the static local features of the image, so as to obtain the image feature sequence and realize the extraction of polymorphic text. Among them, when the network structure can be set to 30 layers, the complete structure is divided into 5 groups. The pooling layers in the first three groups of convolution layers downsample the map image, and the length and height of the image are scaled to the same size. The fourth and fifth groups of convolution layers scale the image height to retain the sequence information in the length direction. The convolution kernel size of the last three layers of the fifth group of convolution layers is reduced to 2*2 to extract more detailed features. While extracting features, the map image is converted into a two-dimensional image feature sequence. The sequence length is the number of features in the length direction of the image, and the dimension is the number of channels of the image. Then, the input image feature sequence is encoded using stacked one-dimensional convolution and nonlinear gating units, and the encoded features are passed to the stacked one-dimensional convolution, which is then decoded and recognized by the nonlinear gating unit and the decoder module, and the recognition result is finally output.

[0049] The text recognition results can be matched with a custom dictionary, and / or the recognition results can be corrected using a language model to correct errors in the text recognition results in the map image to reduce misrecognition and missed recognition.

[0050] Furthermore, based on the above method, an embodiment of the present invention also provides a system for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm, comprising: a denoising module, an enhancement module and a recognition module, wherein:

[0051] A denoising module is used to input the map scan image to be processed into a background filtering model, and remove the background noise in the map scan image by using the background filtering model and connected region analysis to obtain a background-free map image. The background filtering model is a target neural network constructed based on a U-NET neural network and trained using sample data;

[0052] An enhancement module is used to decompose the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhance the resolution of the map image through wavelet transform, and extract text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information;

[0053] The recognition module is used to input text features into a pre-trained neural network model, and use the neural network model to recognize polymorphic text in the map scan image to be processed. The neural network model includes a convolutional neural network structure for extracting a static local feature sequence of the input image and an encoder network for encoding and recognizing the static local feature sequence of the image.

[0054] In order to verify the effectiveness of this solution, the following is a further explanation based on experimental data:

[0055] like Figure 2 (a) in the figure takes the scanned map image of a certain area as an example. The map includes multiple fonts and has a complex background. The background filtering model and Otus method are used to remove the background noise in the scanned map. The binary image is decomposed into the wavelet domain, the wavelet coefficients are thresholded, and the wavelet transform is selected to remove high-frequency noise and retain the details and edges of the image. The Sobel operator is used to calculate the gradient of the image in the x and y directions, and the grayscale co-occurrence matrix of the image is calculated to extract the text features of the image. Then, for the text features of the map annotations, the feature extraction network in the neural network model is used to extract the image features and reduce them to image sequences. The encoder network is used to encode the image sequence information and output the recognition results; and the results are corrected in combination with the custom dictionary and language model. The results are shown in the figure. Figure 2 As shown in (b) in the figure, it is possible to extract and recognize polymorphic text in a map.

[0056] The above experimental data show that this solution can be applied to the automatic extraction of text from paper or scanned maps with large background noise. It has high recognition accuracy, strong practicality and good application prospects.

[0057] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0058] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0059] The units and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person of ordinary skill in the art may use different methods to implement the described functions for each specific application, but such implementation is not considered to be beyond the scope of the present invention.

[0060] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits, and accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of software function modules. The present invention is not limited to any specific form of combination of hardware and software.

[0061] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm, characterized in that: Include: The map scan image to be processed is input into a background filtering model, and background noise in the map scan image is removed by using the background filtering model and connected region analysis to obtain a background-free map image, wherein the background filtering model is a target neural network constructed based on a U-NET neural network and obtained by training with sample data; Decomposing the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhancing the resolution of the map image by wavelet transform, and extracting text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information; The text features are input into a pre-trained neural network model, and the neural network model is used to identify polymorphic text in the map scan image to be processed. The neural network model includes a convolutional neural network structure for extracting the static local feature sequence of the input image and an encoder network for encoding and identifying the static local feature sequence of the image.

2. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1 is characterized in that: The background filtering model includes an encoder and a decoder, wherein a convolution layer is used in the encoder to extract input image features, batch normalization and activation functions are added after each convolution layer, a pooling layer is used for downsampling, and convolution layers and pooling layers are stacked in each downsampling to expand feature diversity by increasing the number of feature map channels; In the decoder, a transposed convolution layer is used to perform a transposed convolution operation on the feature map, and the number of feature map channels in the transposed convolution process is consistent with the number of feature map channels in the image feature extraction process, and the two feature map channels are jump-connected to superimpose the feature maps.

3. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1 or 2, characterized in that: The training process of the background filter model includes the following: Constructing a sample data set, wherein the sample data set includes image sample data with ground object annotations; The U-NET neural network is trained and optimized using the sample data set by minimizing the loss function, and the trained and optimized U-NET neural network is used as the target neural network to obtain the background filtering model.

4. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1, characterized in that: Using background filtering model and connected region analysis to remove background noise in map scanned images, including: The map scanned image is segmented using the background filtering model, and the map scanned image is converted into a grayscale image by calculating the average value of the maximum channel and the minimum channel image pixels; The image binarization segmentation threshold is determined using the big law method, and the grayscale image is binarized using the threshold to remove background noise in the map scan image.

5. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1, characterized in that: The resolution of map images is enhanced through wavelet transform, including: Decompose the map image into the wavelet domain to obtain the low-frequency sub-image of the map image; The low-frequency sub-image is processed with histogram equalization to stretch the concentrated grayscale distribution interval in the histogram to a balanced distribution in the global range, and the pixel values ​​of the image are redistributed to enhance the overall contrast of the low-frequency sub-image in the map image.

6. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1 or 5, characterized in that: Extract text features from map images, including: Use the Sobel operator to calculate the gradient of the map image in the x and y directions, and obtain the gray level co-occurrence matrix of the map image; The gray level co-occurrence matrix is ​​used to extract text features from map images.

7. The method for automatically extracting place name annotations from scanned maps based on a wavelet neural network algorithm according to claim 1, characterized in that: Using a neural network model to recognize polymorphic text in a map scan image to be processed also includes: The text recognition results are matched with a custom dictionary and / or the recognition results are corrected using a language model to correct errors in the text recognition results in the map image.

8. A system for automatically extracting place name annotations from scanned maps based on wavelet neural network algorithm, characterized in that: It includes: denoising module, enhancement module and recognition module, among which, A denoising module is used to input the map scan image to be processed into a background filtering model, and remove the background noise in the map scan image by using the background filtering model and connected region analysis to obtain a background-free map image. The background filtering model is a target neural network constructed based on a U-NET neural network and trained using sample data; An enhancement module is used to decompose the background-free map image into high-frequency information and low-frequency information of different scales and directions, enhance the resolution of the map image through wavelet transform, and extract text features in the map image, wherein the high-frequency information includes map image edge information and contour information, and the low-frequency information includes map image information; The recognition module is used to input text features into a pre-trained neural network model, and use the neural network model to recognize polymorphic text in the map scan image to be processed. The neural network model includes a convolutional neural network structure for extracting a static local feature sequence of the input image and an encoder network for encoding and recognizing the static local feature sequence of the image.

9. An electronic device, characterized in that: include: at least one processor, and a memory coupled to the at least one processor; The memory stores a computer program, and the computer program can be executed by the at least one processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 can be implemented.

Citation Information

Cited By

  • Financial bill information automatic input system based on artificial intelligence

    CN121074935A