Examination paper marking answer sheet filling identification method based on artificial intelligence and GAT

Abstract

The embodiment of the invention provides an examination paper marking answer sheet filling identification method based on artificial intelligence and GAT, and relates to the technical field of intelligent information identification, and the method comprises the steps: obtaining an answer sheet picture, and carrying out the preprocessing of the answer sheet picture; performing region segmentation on the preprocessed answer sheet picture to extract an option region; obtaining a full-filled area and a non-full-filled area of each option area; adopting a set template to match the option region to obtain a screened option region; constructing a graph attention model according to the node information; wherein the node information comprises a graph of a single option area, the number of options of the selected questions, the number of answers of the selected questions and center coordinates of the questions; predicting a set number of selection questions based on the model to obtain a prediction result; the calculation amount is reduced, calculation is accelerated, the model training speed and accuracy are improved, and the training cost is reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent information recognition technology, and in particular to a method for identifying test answer sheet filling based on artificial intelligence and GAT. Background Art

[0002] Test paper objective question filling recognition is a widely used image processing task used in educational assessment and automated marking. Existing test paper objective question filling recognition methods include: recognition schemes based on traditional image processing, recognition schemes based on deep learning, and hybrid schemes based on the combination of deep learning and traditional methods. However, existing methods have problems such as high training cost, low model interpretability, and complex error location and adjustment. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a method for identifying test answer sheet filling based on artificial intelligence and GAT, so as to solve the problems of high training cost, low model interpretability, and complex error location and adjustment in existing methods.

[0004] In a first aspect, an embodiment of the present application provides a method for identifying answer sheet filling based on artificial intelligence and GAT, the method comprising: Obtaining an answer sheet image and preprocessing the answer sheet image; Perform region segmentation on the preprocessed answer sheet image to extract the option area; Get the filled area and unfilled area of ​​each option area; The set template is used to match the option area to obtain the filtered option area; Build a graph attention model based on node information; the node information includes: a graph of a single option area, the number of multiple-choice options, the number of multiple-choice answers, and the center coordinates of the question; Based on the model, a set number of multiple-choice questions are predicted to obtain the prediction results.

[0005] In the above implementation process, the answer sheet image is obtained and preprocessed; the preprocessed answer sheet image is segmented to extract the option area; the filled area and the unfilled area of ​​each option area are obtained; the option area is matched with a set template to obtain the filtered option area; a graph attention model is constructed according to the node information; wherein the node information includes: a graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions and the center coordinates of the questions; a set number of multiple-choice questions are predicted based on the model to obtain the prediction results; the amount of calculation is reduced, the calculation is accelerated, the model training speed and accuracy are improved, and the training cost is reduced.

[0006] Furthermore, obtaining the filled area and the unfilled area of ​​each option area includes: Compare each option in each option area with the set option filling template; If the option is the same as the set option filling template, the option is judged to be a filling area; If the option is different from the set option filling template, the option is judged to be an unfilled area.

[0007] Furthermore, the setting template is used to match the option area to obtain the filtered option area, including: Use the set template to match the height and width of the option area and calculate the matching degree; If the matching degree reaches the set matching threshold, the option area is judged to match the set template; After matching all option areas, filter out all matching option areas; After re-extracting the unmatched option areas, re-perform the matching screening.

[0008] In the above implementation process, the template is set to match the option area to improve the matching degree of the option area.

[0009] Furthermore, constructing a graph attention model according to the node information includes: After resizing the graph of a single option area, a three-layer convolution operation is performed to obtain the information of the set dimension, and the information of the set dimension is used as one of the basic attributes of the node; Construct an edge matrix to represent the relationship between options; A graph attention model is constructed based on the graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions, the center coordinates of the question, and the relationship between the options.

[0010] In the above implementation process, a graph attention model is constructed to implement prediction training for subsequent answer sheets and obtain recognition results for multiple-choice questions.

[0011] Furthermore, constructing the graph attention model includes: The graph attention model is constructed through three convolutional layers and corresponding batch normalization layers, as well as a ReLU activation function.

[0012] In the above implementation process, the training effect of the graph attention model is improved.

[0013] Furthermore, constructing the edge matrix includes: Set the initial weight of the option to 1; Determine the set of all options and the set of all possible edges between options; Create a square matrix of size n×n, where n is the number of options; Initialize all elements in the matrix to 0, where 0 means there is no direct relationship between the two options; Traverse each edge and set the edge weight; Processing self-loops; Complete the matrix.

[0014] In the above implementation process, the options for the same question are weighted, with an initial weight of 1. The edge weight information between different questions is 0 and will not be calculated against each other, thus ensuring the accuracy of training.

[0015] In a second aspect, the embodiment of the present application further provides an examination answer sheet filling and recognition device based on artificial intelligence and GAT, the device comprising: Answer sheet acquisition module, used to acquire answer sheet images and pre-process the answer sheet images; The answer sheet segmentation module is used to segment the preprocessed answer sheet image to extract the option area; A filling judgment module is used to obtain the filled area and the unfilled area of ​​each option area; A region screening module is used to match the option region with a set template to obtain a screened option region; A model building module is used to build a model based on node information; wherein the node information includes: a diagram of a single option area, the number of multiple-choice options, the number of multiple-choice answers, and the center coordinates of the question; The result prediction module is used to predict a set number of multiple-choice questions based on the model to obtain prediction results.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, including: A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the test paper marking and answer sheet filling and recognition method based on artificial intelligence and GAT as described above.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a server, implements the test paper marking and answer sheet filling and recognition method based on artificial intelligence and GAT as described above.

[0018] In a fifth aspect, an embodiment of the present invention provides a computer program product, which includes instructions, and when the instructions are executed by a computer, the computer implements the test paper marking and answer sheet filling and recognition method based on artificial intelligence and GAT as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A flowchart of a method for identifying answer sheet filling based on artificial intelligence and GAT provided in an embodiment of the present application; Figure 2 A schematic diagram of an objective question of an answer sheet picture of a method for identifying answer sheet filling based on artificial intelligence and GAT provided in an embodiment of the present application; Figure 3 A schematic diagram of a single option obtained by cutting a method for identifying answer sheet filling based on artificial intelligence and GAT provided in an embodiment of the present application; Figure 4 A schematic diagram of a multiple-choice question on an untidy answer sheet provided in an examination answer sheet filling and recognition method based on artificial intelligence and GAT provided in an embodiment of the present application; Figure 5 A schematic diagram of a flow chart of an examination answer sheet filling and recognition device based on artificial intelligence and GAT provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0022] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0023] Examination paper objective question filling recognition is a widely used image processing task, used in educational assessment and automated examination marking. There are three major technical solutions, which are introduced below one by one: The recognition scheme based on traditional image processing has the following steps: Image preprocessing: grayscale, denoising, binarization; Region segmentation: extracting option areas based on template matching, edge detection or contour detection; Fill detection: calculating the black pixel ratio or projection histogram of each option area. Advantages: The algorithm is simple, easy to implement, and relies on fewer computing resources; it can run on low-cost hardware devices and is suitable for offline environments; it has high interpretability and can quickly locate and correct errors. Disadvantages: It has high requirements for image quality and is difficult to handle complex scenes such as stretching, erasure stains, noise, tilt, and uneven lighting caused by printing; it relies on templates, has poor adaptability, and cannot recognize flexible or highly variable test paper formats; it has limited processing capabilities for occluded and damaged areas; rapid location of errors will also cause the program to make various patches for different situations.

[0024] The recognition scheme based on deep learning has the following steps: Data annotation: Construct a training dataset containing test paper images and corresponding fill-in areas: Model training: Use a convolutional neural network (CNN) or target detection model (such as YOLO, Faster R-CNN) to train the fill-in recognition model: Prediction output: Detect the area of ​​each option and give whether the option area is filled in (expressed by category). Advantages: Strong adaptability, able to handle a variety of test paper formats and complex backgrounds; High accuracy, better able to cope with noise, tilt and lighting changes; Strong scalability, suitable for integration into large-scale automated marking systems. Disadvantages: Requires a large amount of labeled data and high training costs; Model training and reasoning require high computing resources, and school machines are generally not applicable; Model interpretability is low, and error location and adjustment are more complicated; Changes in the test paper style may cause the model to be unsuitable, and the model needs to be retrained.

[0025] A hybrid solution based on deep learning and traditional methods, steps: Image preprocessing: Use traditional methods for preliminary denoising, correction and template matching; Advanced detection: Apply deep learning models to specific areas for fill-in recognition; Decision fusion: Combine the results of the two methods to improve overall recognition accuracy. Advantages: Combining the high efficiency of traditional methods with the high accuracy of deep learning; Able to achieve higher performance under limited computing resources; More suitable for handling diverse and complex test paper scenarios. Disadvantages: High development complexity, requiring a reasonable fusion of the two methods; There are conflicts and accuracy bottlenecks in coordinating the two methods.

[0026] Based on this, an embodiment of the present application provides an examination answer sheet filling and recognition method based on artificial intelligence and GAT to solve the above problems.

[0027] Please see Figure 1 , Figure 1A flowchart of a method for identifying the filling and marking of an examination answer sheet based on artificial intelligence and GAT provided in an embodiment of the present application. The method for identifying the filling and marking of an examination answer sheet based on artificial intelligence and GAT includes: 100. Obtain the answer sheet image and pre-process the answer sheet image.

[0028] Optionally, if the answer sheet is submitted through an online questionnaire or examination platform, the answer sheet will be downloaded or screenshotted directly on the platform. If the answer sheet is in paper form, a scanner or camera will be used to convert it into a digital image and store it. Download all the answer sheet images and take a set number of answer sheet images of the same set of test papers. Optionally, a certain number of answer sheet images can be obtained according to demand. Preprocessing the answer sheet images includes loading images, resizing, graying, binarization, denoising, etc. Preprocessing the answer sheet images helps improve the image quality and facilitates subsequent processing and analysis.

[0029] 200. Perform region segmentation on the preprocessed answer sheet image to extract the option area.

[0030] It is understandable that there are various techniques for performing region segmentation on answer sheet images.

[0031] Exemplarily, an image segmentation model is trained using a training set, the answer sheet image is segmented into regions using the image segmentation model, the distribution of the segmented regions is labeled, and option regions in the segmented regions are extracted.

[0032] For example, Figure 2 The objective questions of the answer sheet sample are shown. Figure 3 The resulting single option is shown.

[0033] 300. Get the filled area and unfilled area of ​​each option area.

[0034] Optionally, each option in each option area is compared with a set option filling template; if the option is the same as the set option filling template, the option is judged to be a filled area; if the option is different from the set option filling template, the option is judged to be an unfilled area.

[0035] Optionally, count the white pixels in the binary image of the selected area to distinguish the filled area from the unfilled area. Specifically, traverse the selected area; count the number of all white pixels in the image; calculate the total number of pixels in the image, and then subtract the number of pixels in the filled area to obtain the number of pixels in the unfilled area; record the number of pixels or the ratio of the filled and unfilled areas for each selected area.

[0036] 400. Use the set template to match the option area to obtain a filtered option area.

[0037] 410. Use the set template to match the height and width of the option area and calculate the matching degree.

[0038] 420. If the matching degree reaches the set matching threshold, it is determined that the option area matches the set template.

[0039] For example, scaling the height and width of the unified option template to match the height and width of the option area can achieve a matching degree of more than 0.8, which proves that the height and width at this time are the height and width of the option. Especially for untidy test papers, it will be better than the traditional projection method, please refer to Figure 4 Shown is a picture of an untidy answer sheet with multiple-choice questions.

[0040] 430. After matching all option areas, filter out all matching option areas.

[0041] 440. After re-extracting the unmatched option area, re-perform the matching screening.

[0042] It can be understood that the unmatched option areas are re-extracted until all option areas are screened.

[0043] 500. Construct a graph attention model based on node information; wherein the node information includes: a graph of a single option area, the number of multiple-choice question options, the number of multiple-choice question answers, and the center coordinates of the question.

[0044] For example, the graph of a single option area has a size of 25*13, which is adjusted to 32*16. Since the graph has two-dimensional information characteristics, the 32*16 will not be expanded into one dimension as an attribute. Three layers of convolution operations are required to finally obtain 1024-dimensional information as one of the basic attributes of the node. Among them, this part is the same network and participates in gradient backpropagation.

[0045] 510. After the graph of a single option area is resized, a three-layer convolution operation is performed to obtain information of a set dimension, and the information of the set dimension is used as one of the basic attributes of the node.

[0046] 520. Construct an edge matrix to represent the relationship between options.

[0047] 530. Construct a graph attention model based on the graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions, the center coordinates of the question and the relationship between the options.

[0048] Furthermore, constructing the graph attention model includes: The graph attention model is constructed through three convolutional layers and corresponding batch normalization layers, as well as a ReLU activation function.

[0049] Exemplary, the specific process is as follows: 1.self.conv1 = nn.Conv2d(inplanes, planes, kernel_size = 1, bias = False): defines a 2D convolution layer conv1 with inplanes input channels, planes output channels, 1x1 convolution kernel size, and no bias (bias = False).

[0050] 2.self.bn1=nn.BatchNorm2d(planes): defines a 2D batch normalization layer bn1, which is used to normalize the output of conv1. The number of channels after normalization is the same as the output channels of conv1.

[0051] 3.self.conv2==nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False): defines the second 2D convolution layer conv2, with the number of input channels and output channels both being planes, the convolution kernel size being 3x3, the stride being stride (this parameter is not defined in the code and is usually used to control the downsampling of the convolution), the padding being 1 (padding=1), and no bias term being used.

[0052] 4.self.bn2 = nn.BatchNorm2d(planes): defines the second 2D batch normalization layer bn2, which is used to normalize the output of conv2.

[0053] 5.self.conv3 = nn.Conv2d(planes, planes*4, kernel_size=1, bias=False): defines the third 2D convolution layer conv3, with the number of input channels being planes, the number of output channels being planes*4, the convolution kernel size being 1x1, and no bias term being used.

[0054] 6.self.bn3=nn.BatchNorm2d(planes*4): defines the third 2D batch normalization layer bn3, which is used to normalize the output of conv3.

[0055] 7.self.relu=nn.ReLU(inplace=True): defines a ReLU activation function relu. inplace=True means performing calculations in situ, which can reduce memory usage.

[0056] The design of this residual block allows the network to learn a residual mapping, i.e. the difference between input and output, thus making it easier for the network to learn deeper features.

[0057] Furthermore, constructing the edge matrix includes: Set the initial weight of the option to 1; determine the set of all options and the set of all possible edges between options; create a square matrix of size n×n, where n is the number of options; initialize all elements in the matrix to 0, where 0 indicates that there is no direct relationship between the two options; traverse each edge and set the weight of the edge; process self-loops; and complete the matrix.

[0058] For example, if the options in the question are weighted and the initial weights are all 1, a weighted edge matrix can be constructed to represent the relationship between these options; the options here can be understood as vertices in the graph, and the relationship between the options can be understood as edges. Thus, the weight of the edge represents the strength or importance of a certain connection between the options.

[0059] Determine vertices and edges: Determine the set of all options (vertices); Determine the set of all possible relationships (edges) between options.

[0060] Initialize the matrix: Create a square matrix of size (n\timesn), where (n) is the number of options (vertices); initialize all elements in the matrix to 0, because 0 indicates that there is no direct relationship between two options.

[0061] Fill the matrix: Go through each edge ((u, v)) that represents the relationship between option (u) and option (v); for an undirected graph, set (A[u][v]) and (A[v][u]) to the edge weight (in this case 1); for a directed graph, set (A[u][v]) to the edge weight (in this case 1).

[0062] Handling self-loops: If there are no self-loops between options (i.e. an option cannot have a relationship with itself), then the diagonal elements ((A[i][i])) should remain 0.

[0063] Complete the matrix: After completing the above steps, you will get a weighted edge matrix representing the relationship between options, where the value of each non-diagonal element represents the strength of the relationship between the corresponding option pairs.

[0064] 600. Predict a set number of multiple-choice questions based on the model and obtain prediction results.

[0065] For example, in order to process multiple-choice questions in batches without slowing down the calculation due to too many nodes and too large a matrix, this application predicts 5 multiple-choice questions together, which can control the time and improve the overall speed. If there are less than 5 questions, the last question will be used to fill the gap.

[0066] As described above, the embodiment of the present application obtains an answer sheet image and preprocesses the answer sheet image; performs region segmentation on the preprocessed answer sheet image to extract option areas; obtains the filled-in area and the unfilled area of ​​each option area; uses a set template to match the option area to obtain the filtered option area; constructs a graph attention model based on node information; wherein the node information includes: a graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions, and the center coordinates of the questions; predicts a set number of multiple-choice questions based on the model to obtain prediction results; reduces the amount of calculation, speeds up calculations, improves model training speed and accuracy, and reduces training costs.

[0067] The above steps are not to be performed in a strict order as described in the numbers, but should be understood as an overall solution.

[0068] In the second aspect, based on the above embodiments, the present application also provides an examination paper marking answer sheet filling and recognition device based on artificial intelligence and GAT, referring to Figure 5 The test answer sheet filling and recognition device based on artificial intelligence and GAT provided in this embodiment specifically includes: an answer sheet acquisition module 201, an answer sheet segmentation module 202, a filling and judging module 203, an area screening module 204, a model building module 205 and a result prediction module 206.

[0069] Among them, the answer sheet acquisition module 201 is used to obtain the answer sheet image and preprocess the answer sheet image; the answer sheet segmentation module 202 is used to perform regional segmentation on the preprocessed answer sheet image to extract the option area; the filling judgment module 203 is used to obtain the filled area and the unfilled area of ​​each option area; the area screening module 204 is used to match the option area with a set template to obtain the screened option area; the model construction module 205 is used to build a model according to the node information; wherein the node information includes: a map of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions and the center coordinates of the questions; the result prediction module 206 is used to predict a set number of multiple-choice questions based on the model to obtain the prediction results.

[0070] As described above, the embodiment of the present application obtains an answer sheet image and preprocesses the answer sheet image; performs region segmentation on the preprocessed answer sheet image to extract option areas; obtains the filled-in area and the unfilled area of ​​each option area; uses a set template to match the option area to obtain the filtered option area; constructs a graph attention model based on node information; wherein the node information includes: a graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions, and the center coordinates of the questions; predicts a set number of multiple-choice questions based on the model to obtain prediction results; reduces the amount of calculation, speeds up calculations, improves model training speed and accuracy, and reduces training costs.

[0071] In a third aspect, an embodiment of the present application further provides an electronic device that can integrate an artificial intelligence and GAT-based examination answer sheet filling and recognition device with a user-mode polling mechanism provided in an embodiment of the present application. Figure 6 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 6 The electronic device includes: an input device 43, an output device 44, a memory 42 and one or more processors 41; the memory 42 is used to store one or more programs; when the one or more programs are executed by the one or more processors 41, the one or more processors 41 implement the test answer sheet filling recognition method based on artificial intelligence and GAT in the user-mode polling mechanism provided in the above embodiment. The input device 43, the output device 44, the memory 42 and the processor 41 can be connected by a bus or other means, Figure 6 The example of connecting through bus is taken in the following.

[0072] The processor 41 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 42, that is, the test answer sheet filling and recognition method based on artificial intelligence and GAT of the above-mentioned user-mode polling mechanism is realized.

[0073] The electronic device provided above can be used to execute the test paper marking answer sheet filling and recognition method based on artificial intelligence and GAT of the user-mode polling mechanism provided in the above embodiment, and has corresponding functions and beneficial effects.

[0074] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the test paper marking answer sheet filling and recognition method based on artificial intelligence and GAT as described above, and can achieve the same beneficial effects.

[0075] Of course, the storage medium containing computer executable instructions provided in an embodiment of the present application is not limited to the above-mentioned method for filling in and identifying test paper grading answer sheets based on artificial intelligence and GAT, and can also execute related operations in the method for filling in and identifying test paper grading answer sheets based on artificial intelligence and GAT provided in any embodiment of the present application.

[0076] In a fifth aspect, the embodiments of the present application also provide a computer program product, and the methods described in the various embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the processes or functions described in the various embodiments of the present application are executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, a core network device, an OAM (Open Application Model) or other programmable device.

[0077] The computer program or instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0078] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0079] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0080] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0081] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0082] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0083] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for identifying answer sheet filling based on artificial intelligence and GAT, characterized in that: The method comprises: Obtaining an answer sheet image and preprocessing the answer sheet image; Perform region segmentation on the preprocessed answer sheet image to extract the option area; Get the filled area and unfilled area of ​​each option area; The set template is used to match the option area to obtain the filtered option area; Build a graph attention model based on node information; the node information includes: a graph of a single option area, the number of multiple-choice options, the number of multiple-choice answers, and the center coordinates of the question; Based on the model, a set number of multiple-choice questions are predicted to obtain the prediction results.

2. The method for identifying test answer sheets based on artificial intelligence and GAT according to claim 1, characterized in that: The step of obtaining the filled area and the unfilled area of ​​each option area includes: Compare each option in each option area with the set option filling template; If the option is the same as the set option filling template, the option is judged to be a filling area; If the option is different from the set option filling template, the option is judged to be an unfilled area.

3. The method for identifying test answer sheets based on artificial intelligence and GAT according to claim 1, characterized in that: The setting template matching option area is adopted to obtain the filtered option area, including: Use the set template to match the height and width of the option area and calculate the matching degree; If the matching degree reaches the set matching threshold, the option area is judged to match the set template; After matching all option areas, filter out all matching option areas; After re-extracting the unmatched option areas, re-perform the matching screening.

4. The method for identifying test answer sheets based on artificial intelligence and GAT according to claim 1, characterized in that: The step of constructing a graph attention model according to the node information includes: After resizing the graph of a single option area, a three-layer convolution operation is performed to obtain the information of the set dimension, and the information of the set dimension is used as one of the basic attributes of the node; Construct an edge matrix to represent the relationship between options; A graph attention model is constructed based on the graph of a single option area, the number of multiple-choice options, the number of answers to the multiple-choice questions, the center coordinates of the question, and the relationship between the options.

5. The method for identifying test answer sheets based on artificial intelligence and GAT according to claim 4, characterized in that: The constructing of the graph attention model comprises: The graph attention model is constructed through three convolutional layers and corresponding batch normalization layers, as well as a ReLU activation function.

6. The method for identifying test answer sheets based on artificial intelligence and GAT according to claim 4, characterized in that: The constructing edge matrix comprises: Set the initial weight of the option to 1; Determine the set of all options and the set of all possible edges between options; Create a square matrix of size n×n, where n is the number of options; Initialize all elements in the matrix to 0, where 0 means there is no direct relationship between the two options; Traverse each edge and set the edge weight; Processing self-loops; Complete the matrix.

7. An examination answer sheet filling and recognition device based on artificial intelligence and GAT, characterized in that: The device comprises: Answer sheet acquisition module, used to acquire answer sheet images and pre-process the answer sheet images; The answer sheet segmentation module is used to segment the preprocessed answer sheet image to extract the option area; A filling judgment module is used to obtain the filled area and the unfilled area of ​​each option area; A region screening module is used to match the option region with a set template to obtain a screened option region; A model building module is used to build a model based on node information; wherein the node information includes: a diagram of a single option area, the number of multiple-choice options, the number of multiple-choice answers, and the center coordinates of the question; The result prediction module is used to predict a set number of multiple-choice questions based on the model and obtain the prediction results.

8. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the test paper marking answer sheet filling and recognition method based on artificial intelligence and GAT as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a server, implements the test paper marking and answer sheet filling and recognition method based on artificial intelligence and GAT as described in any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises instructions which, when executed by a computer, cause the computer to implement the method according to any one of claims 1-6.

Images