Writing examination intelligent evaluation device and method

Abstract

The invention provides an intelligent evaluation device and method for a writing test, and the device comprises an image obtaining module which obtains a collected image of a writing answer sheet through a photographing device; the detection and recognition module is used for detecting and recognizing the writing answer sheet image by adopting a deep learning model, positioning a written character in the image and marking the position of the written character; and the writing evaluation module is used for analyzing the detected written characters and the whole image, evaluating the writing accuracy score, the aesthetic degree score and the standard degree score, and calculating the total evaluation score. According to the method, the efficient target detection capability and the analysis capability of the deep learning model are combined, real-time and accurate evaluation of the written characters is realized, the method can be applied to scenes such as writing grade examination or writing training and the like, and the reading intelligence and the reading efficiency of the writing grade examination or the writing training are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and deep learning technology, and in particular to an intelligent evaluation device and method for writing tests. Background Art

[0002] In primary and secondary school calligraphy training, students typically practice through traditional methods such as copying calligraphy models. Teachers evaluate handwritten Chinese characters primarily based on their similarity to the standard calligraphy model. According to this criterion, the closer the handwriting is to the model, the higher the score. However, manual evaluation based on similarity has some limitations.

[0003] First, in terms of accuracy, the similarity standard is easily affected by the subjective factors of teachers, resulting in individual differences in scoring results. For example, different teachers may have inconsistent judgments on the same glyph, and teacher fatigue or distraction may also introduce errors in the scoring process. Secondly, in terms of diversity, this evaluation method fails to fully consider the individual differences in students' handwriting, ignores the expression of individual style and innovation, and thus limits the comprehensiveness and diversity of calligraphy evaluation. Thirdly, in terms of real-time performance, traditional methods cannot achieve real-time evaluation because they rely entirely on manual scoring. It is difficult to quickly and promptly discover and deal with writing problems. If multiple teachers conduct evaluations, it will significantly increase labor costs and operating costs. Therefore, the existing handwritten character evaluation methods have problems such as strong subjectivity, poor real-time performance, and high cost. There is an urgent need for a more efficient, real-time, and low-cost intelligent handwritten character evaluation method. Summary of the Invention

[0004] In response to the defects in the existing technology, the purpose of the present invention is to provide an intelligent assessment device and method for writing tests. The present invention combines the efficient target detection and analysis capabilities of deep learning models to achieve real-time and accurate assessment of written characters, thereby improving the intelligence and efficiency of marking writing grade tests or writing training.

[0005] In order to solve the above problems, the technical solution of the present invention is:

[0006] An intelligent evaluation device for a writing test, comprising:

[0007] An image acquisition module, which acquires images of the collected writing answer sheets through a photographing device;

[0008] The detection and recognition module uses a deep learning model to detect and recognize written answer sheet images, locate the written characters in the image, and mark their positions;

[0009] The handwriting evaluation module analyzes the detected handwriting and the entire image, evaluates the accuracy score, aesthetic score and standardization score of the handwriting, and calculates the total evaluation score.

[0010] Preferably, the detection and recognition module uses the EAST model to detect and locate the written answer sheet image, locate the written characters in the image and mark their positions.

[0011] Preferably, the accuracy scoring includes: comparing the recognition result of each written character with a reference standard character image; if the recognition result at the corresponding position is inconsistent with the standard image, the writing is considered to be incorrect, an error label is returned, and the number of error labels is counted to obtain a writing accuracy score;

[0012] The aesthetic evaluation includes: automatically evaluating the image to be evaluated using the discriminator of the trained CycleGAN model, and performing a weighted summation based on the image quality score and the stroke continuity score to obtain the writing aesthetic evaluation score;

[0013] The standardization score includes: obtaining a writing standardization score based on the component quality score and the character shape quality score of the written character;

[0014] The total evaluation score includes: if the writing accuracy score is less than a certain threshold, it is directly judged as unqualified writing; if the writing accuracy score is greater than or equal to a certain threshold, the single-character writing accuracy score, writing beauty score and writing standardization score are weighted and summed to obtain the single-character evaluation score, and the final score is obtained by averaging the scores of all Chinese characters on the writing answer sheet image.

[0015] An intelligent evaluation method for a written test comprises the following steps:

[0016] Capture images of written test answer sheets using a camera, detect and recognize them using a deep learning model, locate the written characters in the images, and mark their locations.

[0017] Compare the written characters with the reference standard character image and give a writing accuracy score;

[0018] The writing aesthetics score is given based on the image quality score and stroke continuity score of the written characters;

[0019] According to the component quality score and the shape quality score of the written characters, a writing standardization score is given;

[0020] A comprehensive score is given based on the writing accuracy score, writing beauty score and writing standardization score.

[0021] Preferably, the steps of acquiring an image of a written test answer sheet by a photographing device, detecting and recognizing the written answer sheet image using a deep learning model, locating the written characters in the image and marking their positions specifically include:

[0022] Acquire images of the collected written test answer sheets through a photographing device;

[0023] In order to meet the needs of handwritten Chinese character detection in the writing level test, a high-quality image dataset was constructed;

[0024] The preprocessed image is input into the EAST model, which performs feature extraction and target detection through its convolutional neural network layer. It combines multi-scale feature fusion with the channel attention mechanism and introduces multiple sets of convolution kernel combinations to enhance feature extraction capabilities, identify the written characters in the image, and give the location area of ​​each written character.

[0025] Preferably, the step of comparing the written characters with the referenced standard character image and giving a writing accuracy score specifically includes: comparing the recognition result of each written character with the referenced standard character image; if the recognition result at the corresponding position is inconsistent with the standard image, the writing is considered to be incorrect, an error label is returned, and the number of error labels is counted to give a writing accuracy score.

[0026] Preferably, the step of providing a writing aesthetic score based on the image quality score and the stroke continuity score of the written characters specifically includes:

[0027] Use the CycleGAN model, a deep learning model that includes a generator, a discriminator, and a loss function. After the model is trained, the discriminator is used to automatically evaluate the image to be evaluated.

[0028] The discriminator outputs the quality score of the image based on the learning results of the model;

[0029] Evaluate the coherence of the strokes of the written characters to obtain a stroke coherence score;

[0030] The obtained image quality score and stroke continuity score are weighted and summed to give a writing beauty score.

[0031] Preferably, the step of providing a writing standardization score based on the component quality score and the font quality score of the written character specifically includes:

[0032] The quality of written characters is evaluated by using the Canny operator for contour detection, extracting the contour information of each Chinese character component, and calculating the cosine similarity between the image and the standard Chinese character.

[0033] The convex hull is used to describe the overall shape and size of a character. The center of the convex hull is used as the center of gravity of the Chinese character. The offset between the convex hull centers of the test image and the standard Chinese character image is calculated. The structural similarity index is used to compare the overall skeleton features of the Chinese character. The two are combined to obtain a character shape quality score.

[0034] The obtained component quality score and glyph quality score are weighted and summed to give a writing standardization score.

[0035] Preferably, the step of performing comprehensive scoring based on the writing accuracy score, the writing beauty score, and the writing standardization score specifically includes:

[0036] If the obtained writing accuracy score is less than a certain threshold, it is directly judged as unqualified;

[0037] If the obtained writing accuracy score is greater than or equal to a certain threshold, the obtained single-character writing accuracy score, writing beauty score and writing standardization score will be weighted and summed, and the scores of all Chinese characters on the writing answer sheet image will be averaged to obtain the final score.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention uses computer vision and deep learning technologies to automatically detect, evaluate and analyze written characters, aiming to achieve real-time evaluation and scoring of written characters, and to improve the intelligence and efficiency of marking writing grade tests or writing training.

[0040] 2. The present invention can automatically learn and extract features from images to realize the detection, recognition and evaluation of Chinese characters. It can effectively reduce the subjective scoring differences between teachers and experts, and provide writers with timely and accurate writing feedback. It has high accuracy in evaluation results and significantly improves evaluation efficiency.

[0041] 3. The present invention does not require the purchase of expensive hardware equipment when collecting written answer sheet images. It only requires daily high-definition shooting settings. Automatic evaluation is achieved through software algorithms, which reduces dependence on physical equipment and avoids high maintenance costs, and has good economic benefits in terms of cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0043] Figure 1 Write a framework diagram of the intelligent test evaluation device for the present invention;

[0044] Figure 2 This is a schematic diagram of the scoring of the handwriting answer sheet of the present invention;

[0045] Figure 3 Schematic diagram of the improved CycleGAN model in the present invention;

[0046] Figure 4 Write a flowchart of the intelligent examination evaluation method for the present invention. DETAILED DESCRIPTION

[0047] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0048] Specifically, the present invention provides an intelligent evaluation device for writing examinations, such as Figures 1 to 3 As shown, the device includes the following modules:

[0049] Image acquisition module: acquires images of the collected written answer sheets through high-definition cameras and other shooting equipment;

[0050] In order to ensure that all written information in the answer sheet image can be clearly detected, the shooting equipment should be set at a high and fixed position to cover the entire answer sheet and prepare data for subsequent evaluation.

[0051] Detection and Recognition Module: This module uses a deep learning model to detect and recognize written answer sheet images, locate the written characters in the images, and mark their positions.

[0052] Specifically, the detection and recognition module uses the EAST model to detect and localize written words in the answer sheet image, locating the written words and marking their positions. The EAST model (Efficient and Accurate Scene Text Detector) is a simplified end-to-end text detection method that is particularly suitable for scene text detection tasks.

[0053] The Efficient Channel Attention mechanism is introduced in the feature extraction stage to refine the weight distribution between feature channels. For small component datasets (such as small-sized text areas or specific types of small object data), their features are first enhanced and refined through convolutional layers, and then cascaded with the feature pyramid to better retain fine-grained information in multi-scale feature aggregation.

[0054] Writing evaluation module: Utilizes traditional Chinese character features and computer vision-based image features to analyze the detected handwritten characters and the entire image, evaluates the writing accuracy, aesthetics, and standardization scores, and calculates the total evaluation score.

[0055] (1) Accuracy score: When recognizing written characters, an attention mechanism and parallel 3×3 and 5×5 convolution kernels are introduced into the bottleneck layer structure of the convolutional neural network architecture CNN to dynamically weight the importance of features and highlight the key features related to written characters. The model can decide how to allocate attention in the feature space based on the context information of each feature, thereby improving the accuracy and effect of feature extraction. The Squeeze-and-Excitation module is used to enhance the model's ability to capture key features. In the compression stage, the SE module aggregates the global spatial information of each channel of the input feature map through global average pooling. In the excitation stage, the SE module generates a weight value for each channel through an adaptive weight generation mechanism, thereby selectively enhancing or suppressing the channel to improve the overall recognition accuracy. Global average pooling is used to average all elements of each feature channel, thereby reducing the entire feature map to a scalar, avoiding the information loss that may be caused by traditional pooling methods and reducing the dimension of the feature map. The recognition result of each written character is compared with the reference standard character image. If the recognition result at the corresponding position is inconsistent with the standard image, the writing is considered incorrect, and an error label is returned. The number of error labels is counted as the basis for the writing accuracy score.

[0056] (2) Aesthetics scoring: The discriminator of the trained CycleGAN model is used to automatically evaluate the images to be evaluated. The CycleGAN model (Cycle-Consistent Generative Adversarial Network) is an image-to-image conversion model based on a generative adversarial network. The task of the generator is to convert high-quality writing test images in the self-built dataset into generated Chinese character images, while the other generator is to restore the generated Chinese character images to high-quality writing test images. The discriminator distinguishes between real images and generated images, and judges handwritten Chinese character images and printed images respectively. At the same time, the model adds a cycle consistency loss to ensure that the handwritten Chinese character images generated by the generator are as similar as possible to the reconstructed image after being converted into printed images by the inverse generator, thereby achieving cycle consistency. A multi-scale discriminator is introduced to comprehensively consider image features at multiple resolutions, thereby giving the image a more comprehensive "quality score". The structure of each discriminator is similar to that of a traditional PatchGAN, consisting of multiple convolutional layers with a kernel size of 4x4 and a stride of 2. This layer extracts higher-dimensional features by progressively reducing the size of the feature map. Each convolutional layer is followed by a LeakyReLU activation function and instance normalization (InstanceNorm). The final layer outputs a scalar to determine whether the input image is realistic. The generator portion of the model combines multiple ResNet building blocks, including an input convolutional layer, a downsampling layer, multiple ResNet blocks, a transposed convolution, and an output convolutional layer. The CycleGAN model's ability to perceive image details is enhanced by combining traditional image feature extraction methods, such as Gabor eight-directional features and histogram of oriented gradients. The coherence of the written characters is then evaluated. Stroke coherence is a key indicator of the aesthetic quality of Chinese characters. By examining the connectivity of strokes within a single character, it is possible to determine the naturalness and flow of writing and identify potential breaks. First, the character image is skeletonized to obtain a refined glyph outline. Next, the image's breakpoints and connectivity are analyzed to determine if any strokes are broken, further assessing the character's fluency and integrity. The image quality score and stroke continuity score are weighted and summed to produce a character's aesthetic score.

[0057] (3) Standardization scoring: It includes two parts: component evaluation and glyph evaluation. The Canny operator is used for contour detection to extract the contour information of each Chinese character component, and the cosine similarity between the component and the standard Chinese character image is calculated. Then, the symmetry of the glyph is judged based on the proportion of the component in the whole character, such as the left-right symmetry or the top-bottom symmetry of the glyph, so as to verify whether it conforms to the standard font structure and evaluate the accuracy and standardization of the Chinese character components. The convex hull center is introduced as a standard for glyph standardization evaluation. The convex hull can describe the overall shape and size of a character, and its center point represents the geometric center of gravity of the character, reflecting the symmetrical characteristics of the glyph. The standardization of the glyph to be tested is further quantified by calculating the offset between the convex hull center of the image to be tested and the standard Chinese character image. In the standardization evaluation of the whole character, the skeleton feature can effectively reflect the overall structure of the Chinese character. The structural similarity index (SSIM) is used to compare the overall skeleton features of the Chinese character image. SSIM is particularly suitable for comparing overall structure and global features and has higher effectiveness.

[0058] (4) Total evaluation score: If the writing accuracy score is less than a certain threshold, the writing is directly judged as unqualified; if the writing accuracy score is greater than or equal to a certain threshold, the single-character writing accuracy score, writing beauty score and writing standardization score are weighted and summed to obtain the single-character evaluation score. The final score is obtained by averaging the scores of all Chinese characters on the writing answer sheet image, or the score level is determined accordingly.

[0059] Furthermore, the present invention also provides an intelligent evaluation method for writing examinations, such as Figure 4 As shown, the method includes the following steps:

[0060] S1: Capture images of written test answer sheets using a camera, detect and recognize them using a deep learning model, locate the written characters in the images, and mark their locations.

[0061] Specifically, step S1 includes the following steps:

[0062] Step S11: Capture the collected written test answer sheet image using a high-definition camera or other shooting device. To ensure that all written characters in the answer sheet image can be clearly detected, the shooting device should be set at a high and fixed position to cover the entire answer sheet.

[0063] Step S12: In order to meet the requirements of handwritten Chinese character detection in the writing level test, a high-quality image dataset is constructed; the images are uniformly scaled to a fixed size of 600×600, covering more than 40,000 characters with different writing styles, aiming to simulate the actual differences between exam scenarios and daily writing.

[0064] Step S13: The preprocessed image is input into the EAST model. The model performs feature extraction and target detection through its convolutional neural network layer. It combines multi-scale feature fusion with the channel attention mechanism and introduces multiple sets of convolution kernel combinations to enhance feature extraction capabilities, identify the written characters in the image, and give the location area of ​​each written character.

[0065] S2: Compare the written characters with the reference standard character image and give a writing accuracy score;

[0066] Specifically, step S2 includes the following steps:

[0067] Step S21: Introduce the attention mechanism to dynamically weight the importance of features, highlight key features related to written characters, and let the model decide how to allocate attention in the feature space based on the contextual information of each feature, thereby improving the accuracy and effect of feature extraction; use the Squeeze-and-Excitation module to enhance the model's ability to capture key features to improve overall recognition accuracy; use Global Average Pooling to reduce the dimension of the feature map.

[0068] Step S22: Compare the recognition result of each written character with the reference standard character image. If the recognition result at the corresponding position is inconsistent with the standard image, it is considered that the writing is incorrect, and an error label is returned. The number of error labels is counted and a writing accuracy score is given.

[0069] S3: Give a writing aesthetic score based on the image quality score and stroke continuity score of the written characters;

[0070] Specifically, step S3 includes the following steps:

[0071] Step S31: Using a deep learning model called CycleGAN, which consists of a generator, a discriminator, and a loss function, after training, the discriminator automatically evaluates the test images. The generator is tasked with converting high-quality handwriting test images from a self-built dataset into generated Chinese character images. Another generator is responsible for restoring the generated Chinese character images back to high-quality writing test images. The discriminator distinguishes between real and generated images, making judgments on handwritten Chinese character images and printed characters, respectively.

[0072] Step S32: A multi-scale discriminator is introduced to comprehensively consider image features at multiple resolutions, thereby assigning a more comprehensive "quality score" to the image. The generator portion of the model combines multiple ResNet building blocks and uses traditional image feature extraction methods to enhance image detail and expressiveness. The discriminator outputs an image quality score based on the model's learning results.

[0073] Step S33: Evaluate the continuity of the strokes of the written character to obtain a stroke continuity score.

[0074] First, the character image is skeletonized to obtain a refined glyph outline. The image is then analyzed for breakpoints and connectivity to determine if there are any breaks in the strokes, further assessing the glyph's fluency and integrity.

[0075] Step S34: Perform a weighted sum of the image quality score obtained in step S32 and the stroke continuity score obtained in step S33 to obtain the aesthetic score of a Chinese character.

[0076] S4: Give a writing standardization score based on the component quality score and the font quality score of the written characters;

[0077] Specifically, step S4 includes the following steps:

[0078] Step S41: The quality of the written character components is evaluated. The Canny operator is used to perform contour detection, extracting the contour information of each Chinese character component and calculating the cosine similarity between the component and the standard Chinese character image. The symmetry of the glyph, such as left-right symmetry or top-bottom symmetry, is then determined based on the component's proportion within the entire character to verify compliance with the standard font structure.

[0079] Step S42: Use the convex hull to describe the overall shape and size of a character, use the convex hull center as the center of gravity of the Chinese character, calculate the offset between the convex hull centers of the test image and the standard Chinese character image; use the structural similarity index SSIM to compare the overall skeleton features of the Chinese character; combine the two to obtain a character quality score.

[0080] Step S43: Perform a weighted sum of the component quality score obtained in step S41 and the glyph quality score obtained in step S42 to obtain a standardization score for a certain written character.

[0081] S5: Comprehensive scoring is performed based on the writing accuracy score, writing beauty score and writing standardization score.

[0082] Specifically, step S5 includes the following steps:

[0083] Step S51: If the obtained writing accuracy score is less than a certain threshold, it is directly judged as unqualified writing;

[0084] Step S52: If the obtained writing accuracy score is greater than or equal to a certain threshold, the obtained single-character writing accuracy score, writing beauty score and writing standardization score are weighted and summed, and the scores of all Chinese characters on the writing answer sheet image are averaged to obtain the final score.

[0085] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. An intelligent evaluation device for writing examination, characterized in that: The device comprises: An image acquisition module, which acquires images of the collected writing answer sheets through a photographing device; The detection and recognition module uses a deep learning model to detect and recognize written answer sheet images, locate the written characters in the image, and mark their positions; The handwriting evaluation module analyzes the detected handwriting and the entire image, evaluates the accuracy score, aesthetic score and standardization score of the handwriting, and calculates the total evaluation score.

2. The intelligent evaluation device for writing test according to claim 1, characterized in that: The detection and recognition module uses the EAST model to detect and locate the written answer sheet image, locate the written characters in the image and mark their positions.

3. The intelligent writing test evaluation device according to claim 1, characterized in that: The accuracy scoring includes: comparing the recognition result of each written character with the reference standard character image; if the recognition result at the corresponding position is inconsistent with the standard image, the writing is considered to be incorrect, an error label is returned, and the number of error labels is counted to obtain a writing accuracy score; The aesthetic evaluation includes: automatically evaluating the image to be evaluated using the discriminator of the trained CycleGAN model, and performing a weighted summation based on the image quality score and the stroke continuity score to obtain the writing aesthetic evaluation score; The standardization score includes: obtaining a writing standardization score based on the component quality score and the character shape quality score of the written character; The total evaluation score includes: if the writing accuracy score is less than a certain threshold, it is directly judged as unqualified writing; if the writing accuracy score is greater than or equal to a certain threshold, the single-character writing accuracy score, writing beauty score and writing standardization score are weighted and summed to obtain the single-character evaluation score, and the final score is obtained by averaging the scores of all Chinese characters on the writing answer sheet image.

4. An intelligent evaluation method for writing examinations, characterized in that: The method comprises the following steps: Capture images of written test answer sheets using a camera, detect and recognize them using a deep learning model, locate the written characters in the images, and mark their locations. Compare the written characters with the reference standard character image and give a writing accuracy score; The writing aesthetics score is given based on the image quality score and stroke continuity score of the written characters; According to the component quality score and the shape quality score of the written characters, a writing standardization score is given; A comprehensive score is given based on the writing accuracy score, writing beauty score and writing standardization score.

5. The intelligent evaluation method for writing test according to claim 4, characterized in that: The steps of acquiring an image of a written test answer sheet by a photographing device, detecting and recognizing the written answer sheet image using a deep learning model, locating the written characters in the image and marking their positions specifically include: Acquire images of the collected written test answer sheets through a photographing device; In order to meet the needs of handwritten Chinese character detection in the writing level test, a high-quality image dataset was constructed; The preprocessed image is input into the EAST model, which performs feature extraction and target detection through its convolutional neural network layer. It combines multi-scale feature fusion with the channel attention mechanism and introduces multiple sets of convolution kernel combinations to enhance feature extraction capabilities, identify the written characters in the image, and give the location area of ​​each written character.

6. The intelligent evaluation method for writing test according to claim 4, characterized in that: The step of comparing the written characters with the reference standard character image and giving a writing accuracy score specifically includes: comparing the recognition result of each written character with the reference standard character image; if the recognition result at the corresponding position is inconsistent with the standard image, it is considered that the writing is incorrect, an error label is returned, and the number of error labels is counted to give a writing accuracy score.

7. The intelligent evaluation method for writing test according to claim 4, characterized in that: The step of providing a writing aesthetic score based on the image quality score and the stroke continuity score of the written characters specifically includes: Use the CycleGAN model, a deep learning model that includes a generator, a discriminator, and a loss function. After the model is trained, the discriminator is used to automatically evaluate the image to be evaluated. The discriminator outputs the quality score of the image based on the learning results of the model; Evaluate the coherence of the strokes of the written characters to obtain a stroke coherence score; The obtained image quality score and stroke continuity score are weighted and summed to give a writing beauty score.

8. The intelligent evaluation method for writing test according to claim 4, characterized in that: The step of providing a writing standardization score based on the component quality score and the font quality score of the written character specifically includes: The quality of written characters is evaluated by using the Canny operator for contour detection, extracting the contour information of each Chinese character component, and calculating the cosine similarity between the image and the standard Chinese character. The convex hull is used to describe the overall shape and size of a character. The center of the convex hull is used as the center of gravity of the Chinese character. The offset between the convex hull centers of the test image and the standard Chinese character image is calculated. The structural similarity index is used to compare the overall skeleton features of the Chinese character. The two are combined to obtain a character shape quality score. The obtained component quality score and glyph quality score are weighted and summed to give a writing standardization score.

9. The intelligent evaluation method for writing test according to claim 4, characterized in that: The step of performing comprehensive scoring based on the writing accuracy score, the writing beauty score, and the writing standardization score specifically includes: If the obtained writing accuracy score is less than a certain threshold, it is directly judged as unqualified; If the obtained writing accuracy score is greater than or equal to a certain threshold, the obtained single-character writing accuracy score, writing beauty score and writing standardization score will be weighted and summed, and the scores of all Chinese characters on the writing answer sheet image will be averaged to obtain the final score.

Images