Paper ruler regulation drawing operation image batch processing and cloud collaborative scoring system

Abstract

The invention relates to the technical field of image processing, and discloses a paper ruler regulation drawing operation image batch processing and cloud collaborative scoring system, which comprises the following steps: acquiring operation images of a paper ruler regulation drawing through a multi-source image, carrying out resolution adaptation and intelligent edge detection, carrying out perspective correction and standardization processing, and carrying out cloud collaborative scoring on the operation images of the paper ruler regulation drawing; reconstructing a brightness component of the operation image and carrying out vectorization processing; constructing a geometric constraint network of the job image to perform complex topological relation automatic constraint verification, after verification is finished, constructing a symbol data set, detecting and classifying symbol data, and performing multi-scale detection on symbol feature values obtained through calculation; constructing a directed acyclic graph, accelerating a key operator by using CUDA, constructing a mapping specification knowledge graph, dynamically adjusting a scoring weight, establishing a dynamic scoring mechanism, and calculating to obtain a processing score; and establishing a distributed storage architecture, analyzing and calculating a final score, and generating personalized improvement suggestions and teaching resource recommendations through error pattern mining.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to paper ruler and compass drawing operation image batch processing and cloud collaborative scoring system. BACKGROUND

[0002] Existing patents generally rely on Gaussian filtering, histogram equalization traditional method for image denoising and light correction, but the effect is limited when processing irregular lines of hand-drawn figures; Most systems use edge detection combined with Hough transform to recognize straight lines and circles, but the analytical ability for complex geometric structures is insufficient; The existing cloud system lacks dynamic scoring standard adjustment mechanism and relies on artificial preset sample; Some patents rely on customized hardware or specific formats, which limits the application scenarios; In the prior art, the existing system mainly identifies numbers and simple symbols, and lacks support for ruler and compass drawing special symbols; Deep learning models perform excellently in number recognition, but have poor generalization ability for different handwriting forms of the same symbol; The existing system only focuses on the final result and cannot restore the drawing process of students; When recognizing geometric errors, the system cannot locate the specific operation link; The education field requires scoring to include figure neatness and labeling integrity dimensions, but existing patents only focus on geometric correctness; Therefore, it is necessary to provide a paper ruler and compass drawing operation image batch processing and cloud collaborative scoring system. SUMMARY

[0003] The purpose of the present application is to provide a paper ruler and compass drawing operation image batch processing and cloud collaborative scoring system, in order to solve the above-mentioned prior art problems, the present application realizes the following technical scheme: In a first aspect, the paper ruler and compass drawing operation image batch processing and cloud collaborative scoring system provided by the embodiments of the present application specifically includes the following modules: The acquisition and processing module: through multi-source image acquisition, the resolution of the paper ruler and compass drawing operation image is adapted and intelligent edge detection is performed, perspective correction and standardization processing are performed, the brightness component of the operation image is reconstructed and vectorization processing is performed; The verification and detection module: based on the completion of operation image processing, the geometric constraint network of the operation image is constructed to automatically verify the complex topological relationship, the symbol feature value is calculated after the verification is completed, the symbol data is detected and classified, and the symbol feature value is calculated for multi-scale detection; The adjustment and evaluation module: based on the completion of multi-scale detection, a directed acyclic graph is constructed, CUDA is used to accelerate key operators, a drawing specification knowledge graph is constructed, scoring weights are dynamically adjusted, a dynamic scoring mechanism is established, and processing scores are calculated; Optimize the recommendation module: Based on the obtained processing scores, establish a distributed storage architecture, analyze and calculate the final scores, and generate personalized improvement suggestions and teaching resource recommendations through error pattern mining.

[0004] Furthermore, the method for performing resolution adaptation is as follows: It can collect images of paper ruler and compass drafting operations, and supports multi-source image acquisition. It supports input from mobile phones, scanners and document cameras, and automatically identifies the device type through the device model field in the EXIF ​​information. Based on the collected images of paper ruler and compass drafting, resolution adaptation is performed. Images with resolutions lower than 1920×1080 are automatically super-resolution processed, and the image super-resolution ESRGAN model is used to magnify the magnification factor. Based on the acquisition of images from paper-based drafting work, intelligent edge detection is performed using the Canny algorithm, calculated through the following formula: Calculate the detection dynamic threshold and ,in, The maximum value of the image gradient; The Hough transform was used to detect the border of the workbook and extract the coordinates of the four corner points. , , , ; Furthermore, the method for performing perspective correction is as follows: Based on the acquired images of paper ruler and compass drafting, perspective correction and standardization processes are performed, and the perspective transformation matrix is ​​calculated. The transformation error is minimized by using the least squares method, as shown in the formula. The minimum error was calculated. Furthermore, the method for reconstructing the brightness components of the image is as follows: Input the standardized image, convert it to HSV space, and extract the luminance component V for subsequent processing; The luminance component V is decomposed using a 3-level Haar wavelet decomposition, and the reconstruction formula is as follows: The reconstructed luminance components were calculated. ,in, The third layer of low-frequency coefficients is used to preserve the main lines. These are the high-frequency coefficients of the k-th layer, used to suppress noise. and These are the weighting coefficients. To optimize the value through experiments, It is a soft thresholding function, and ,in ; By using morphological closing operations and filling the breakpoints with an elliptic kernel, the structuring element B satisfies: in, These are the pixel coordinates; Based on the obtained standardized operation image, the operation image is vectorized. The Douglas-Peucker algorithm is used to simplify polygons with a simplification threshold of 2 pixels, and a DXF file is output. Furthermore, the method for automatic constraint verification is as follows: Based on the completion of detection, key point orientation is assigned, gradient histogram is calculated, and peak values ​​within ±15° of the main direction are taken. Key points are classified and identified using a CNN model to classify point types; Based on the completion of classification and recognition, a task image constraint network is constructed, with the detected geometric elements as nodes and the constraint relationships between elements as edges; Based on the obtained task image constraint network, a graph neural network is trained on the task image constraint network to construct a loss function. ,in To establish realistic constraint relationships, construct a ruler and compass drafting sample with annotations. Constraint verification is performed based on the trained task image constraint network. Parallel constraint verification: ; Vertical constraint verification: ; Tangency constraint verification: Where k is the slope, Preset parallel constraint threshold , The preset vertical constraint threshold is used; The equation of the line is Ax + By + C = 0. Let the coordinates be the center of the circle. Let the radius be the circle. The preset constraint threshold; Furthermore, the method for obtaining the symbolic feature value is as follows: Construct a symbol dataset containing ruler and compass drafting symbols. Each symbol class contains variant samples. Enhance the symbol data by random rotation, scaling, and Gaussian blur. Based on the constructed symbol dataset, symbol data is detected and classified, and a contextual attention module is added, using the formula: The symbolic eigenvalues ​​are calculated. Where M is a mask matrix that filters out interference from irrelevant regions. As the current focus, For the candidate information available for query, The actual content corresponding to the available candidate information. Scaling factor for Dimensions The dot product of the target that needs to be focused on and the available candidate information; Multi-scale detection based on symbolic feature values; Furthermore, the method for constructing the directed acyclic graph is as follows: Construct a Directed Acyclic Graph (DAG): nodes represent processing steps, and edges represent dependencies; GPU resources are allocated based on preset weights according to the computational complexity of the nodes; Nodes are divided into high-complexity nodes and low-complexity nodes based on a pre-defined weight according to computational complexity. For high-complexity nodes, allocate ≥4GB of video memory; for low-complexity nodes, share GPU resources. Furthermore, the method for obtaining the processing score is as follows: Establish a dynamic scoring mechanism using the formula: The processing score is calculated, where, For the total number of rules, Let be the degree of compliance with the i-th rule. For compliance weighting, For rule indexing, The average error rate of the rule in historical assignments. The error rate threshold, This is the weight sensitivity coefficient; Furthermore, the method for calculating the final score is as follows: Using IPFS to store incremental data, smart contracts ensure the atomicity of the scoring. Based on the obtained processing score, a distributed storage architecture is established to convert the original image into IPFS hash storage and upload the metadata to the blockchain; A conflict resolution mechanism is established through smart contracts. When the same task receives two different scores within 10 minutes, consensus is triggered using the following formula: Receive final score ,in As a weight for the rater, Rate the ratings, Index for raters; Furthermore, the method for error pattern mining is as follows: By identifying error patterns, personalized improvement suggestions and teaching resource recommendations are generated. A visual feedback report is generated based on the final score. Error annotation: Color marks are overlaid on the original image. The quantitative indicators are radar charts of scores in each dimension and curves comparing scores with the class average. Use DBSCAN to identify common error patterns and obtain a typical error pattern library; Given the similarity between user errors and resource coverage errors and the user's knowledge level, the probability of recommending resource i is obtained. The similarity between user errors and resource coverage errors is fused with the recommendation probability of resource i under the condition of user knowledge mastery to obtain the learning recommendation value, and teaching resources are recommended based on the learning recommendation value; Based on the visual feedback report, a heatmap of class error distribution is generated, and optimization suggestions are made for the teaching links corresponding to high-frequency errors.

[0005] Secondly, the method for batch processing and cloud-based collaborative scoring of paper ruler and compass drawing images provided in this embodiment of the invention specifically includes the following steps: Step 1: Acquire images of paper ruler and compass drawings from multiple sources, perform resolution adaptation and intelligent edge detection, perform perspective correction and standardization, reconstruct the brightness components of the drawings, and perform vectorization. Step 2: Based on the completion of the task image processing, construct the geometric constraint network of the task image to automatically verify the complex topological relationships. After the verification is completed, construct the symbol dataset, detect and classify the symbol data, calculate the symbol feature values, and perform multi-scale detection. Step 3: Based on the completion of multi-scale detection, construct a directed acyclic graph, use CUDA to accelerate key operators, construct a knowledge graph of mapping specifications, dynamically adjust the scoring weights, establish a dynamic scoring mechanism, and calculate the processing score. Step 4: Based on the obtained processing scores, establish a distributed storage architecture, analyze and calculate the final scores, and generate personalized improvement suggestions and teaching resource recommendations through error pattern mining.

[0006] The beneficial effects of this invention are: 1. By combining ESRGAN super-resolution technology with ruler and compass drawing processing, the super-resolution effect is optimized for the texture details of hand-drawn lines, solving the problems of blurring and loss of breakpoints after low-resolution images are enlarged; compared with fixed threshold, the dynamic threshold Canny algorithm can adaptively adjust the edge detection sensitivity according to the image gradient characteristics, making the recognition of lines of uneven thickness in ruler and compass drawings more accurate. 2. GNN is introduced into the geometric verification of ruler and compass drafting. The network is constructed with geometric elements as nodes and constraints as edges to achieve global reasoning on complex topological relationships, which solves the problem of local correctness but global contradiction caused by traditional step-by-step verification. A knowledge graph of drafting standards is constructed and combined with a dynamic weight formula so that the scoring weight is dynamically adjusted with the historical error rate, which solves the problem of insufficient sensitivity of traditional fixed weights to error-prone points. Error patterns are clustered based on the DBSCAN algorithm and combined with knowledge graph to associate knowledge points, so as to achieve the tracing from the scoring result to the error root cause. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the structure of the paper ruler drawing operation image batch processing and cloud collaborative scoring system provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of the steps of the method for batch processing of paper ruler drawing images and cloud-based collaborative scoring provided in Embodiment 2 of the present invention. Detailed Implementation

[0009] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0010] Example 1: As Figure 1 As shown, the image batch processing and cloud-based collaborative scoring system for paper ruler and gauge drafting work provided in this embodiment of the invention specifically includes the following modules: Acquisition and processing module: Acquires work images of paper ruler and compass drafting from multiple sources, performs resolution adaptation and intelligent edge detection, performs perspective correction and standardization processing, reconstructs the brightness components of the work images and performs vectorization processing; It can collect images of paper ruler and compass drafting operations, and supports multi-source image acquisition. It supports input from mobile phones, scanners and document cameras, and automatically identifies the device type through the device model field in the EXIF ​​information. Based on the collected images of paper ruler and compass drafting, resolution adaptation is performed. Images with resolutions lower than 1920×1080 are automatically super-resolution processed, and the image super-resolution ESRGAN model is used to magnify the magnification factor. It should be noted that the ESRGAN model is a technique that uses algorithms to recover high-resolution images from low-resolution images. Its core goal is to solve the problems of blurring and loss of detail that occur after image magnification, making low-resolution images clear and retaining more texture details. Based on the acquisition of images from paper-based drafting work, intelligent edge detection is performed using the Canny algorithm, calculated through the following formula: Calculate the detection dynamic threshold , ,in, The maximum value of the image gradient; It should be noted that the Canny algorithm is a classic edge detection algorithm. Its core goal is to accurately extract meaningful edges from an image while suppressing noise interference to the greatest extent possible. The Hough transform was used to detect the border of the workbook and extract the coordinates of the four corner points. , , , ; Based on the acquired images of paper ruler and compass drafting, perspective correction and standardization processes are performed, and the perspective transformation matrix is ​​calculated. The transformation error is minimized by using the least squares method, as shown in the formula. The minimum error was calculated. The standardized output size is A4: 210mm × 297mm, with a resolution of 300dpi. By combining wavelet transform and adaptive morphological filtering, the problem of broken points in blurred lines can be solved. Specifically, the standardized image is input, converted to HSV space, and the luminance component V is extracted for subsequent processing. The luminance component V is decomposed using a 3-level Haar wavelet decomposition, and the reconstruction formula is as follows: The reconstructed luminance components were calculated. ,in, The third layer of low-frequency coefficients is used to preserve the main lines. These are the high-frequency coefficients of the k-th layer, used to suppress noise. and These are the weighting coefficients. To optimize the value through experiments, It is a soft thresholding function, and ,in ; By using morphological closing operations and filling the breakpoints with an elliptic kernel, the structuring element B satisfies: in, These are the pixel coordinates; Based on the obtained standardized operation image, the operation image is vectorized. The Douglas-Peucker algorithm is used to simplify polygons with a simplification threshold of 2 pixels. The output DXF file includes: preserved linetype, layers, and endpoint coordinates. Verification and detection module: Based on the completion of the task image processing, construct the geometric constraint network of the task image to automatically verify the complex topological relationships. After the verification is completed, construct the symbol dataset, detect and classify the symbol data, calculate the symbol feature values ​​and perform multi-scale detection. A geometric constraint network for the task image is constructed using a graph neural network to achieve automatic verification of complex topological relationships; The SIFT algorithm is used to detect key points, which include: center point, endpoints, and intersections. Scale-space extremum detection is performed based on key points, with a standard deviation of 1.6 and a Gaussian difference pyramid of 3 levels; Based on the completion of detection, key point orientation is assigned, gradient histogram is calculated, and peak values ​​within ±15° of the main direction are taken. Key points are classified and identified using a CNN model to classify point types; Based on the completion of classification and recognition, a task image constraint network is constructed, with the detected geometric elements as nodes and the constraint relationships between elements as edges; It should be noted that geometric elements include, but are not limited to: points, lines, and circles; and constraint relationships include, but are not limited to: parallel, perpendicular, and tangent. Based on the obtained task image constraint network, a graph neural network is trained on the task image constraint network to construct a loss function. ,in To establish realistic constraint relationships, construct a ruler and compass drafting sample with annotations. Constraint verification is performed based on the trained task image constraint network. Parallel constraint verification: ; Vertical constraint verification: ; Tangency constraint verification: Where k is the slope, Preset parallel constraint threshold , The preset vertical constraint threshold is used; The equation of the line is Ax + By + C = 0. Let the coordinates be the center of the circle. Let the radius be the circle. To preset the constraint threshold, preset ; By combining the Transformer model, we can achieve contextual understanding of multi-scale technical symbols; Construct a symbol dataset containing ruler and compass drafting symbols. Each symbol class contains variant samples. Enhance the symbol data by random rotation, scaling, and Gaussian blur. Based on the constructed symbol dataset, symbol data is detected and classified, and a contextual attention module is added, using the formula: The symbolic eigenvalues ​​are calculated. Where M is a mask matrix that filters out interference from irrelevant regions. As the current focus, For the candidate information available for query, The actual content corresponding to the available candidate information. Scaling factor for Dimensions The dot product of the target that needs to be focused on and the available candidate information; Multi-scale detection based on symbolic feature values; Construct a symbol-semantic mapping table, for example: roughness symbol → surface finish requirement; dimension annotation → tolerance range calculation; Check the association between symbols and geometric elements; for example, a diameter symbol must be associated with a circle. Adjust the evaluation module: Based on the completion of multi-scale detection, construct a directed acyclic graph, use CUDA to accelerate key operators, construct a knowledge graph of mapping specifications, dynamically adjust the scoring weights, establish a dynamic scoring mechanism, and calculate the processing score; Based on multi-scale detection technology, the key operators are accelerated by CUDA through the task of decomposing a directed acyclic graph (DAG). Construct a Directed Acyclic Graph (DAG): nodes represent processing steps, and edges represent dependencies; GPU resources are allocated based on preset weights according to the computational complexity of the nodes; Nodes are divided into high-complexity nodes and low-complexity nodes based on a pre-defined weight according to computational complexity. For high-complexity nodes, allocate ≥4GB of video memory; for low-complexity nodes, share GPU resources. Construct a knowledge graph of cartographic standards and dynamically adjust the scoring weights; Ontology design: It includes multiple core entities, including: rules, error types, knowledge points, difficulty, and tools; Relationship definition: Includes the associations between core entities; For example, the triplet is: "Dimensioning error", "belongs to", "Normative error"; "The lines are not parallel", "leading to", "geometric relationship error"; Establish a dynamic scoring mechanism using the formula: The processing score is calculated, where, For the total number of rules, Let be the degree of compliance with the i-th rule. For compliance weighting, For rule indexing, The average error rate of the rule in historical assignments. The error rate threshold, This is the weight sensitivity coefficient; Optimize the recommendation module: Based on the obtained processing scores, establish a distributed storage architecture, analyze and calculate the final scores, and generate personalized improvement suggestions and teaching resource recommendations through error pattern mining; Using IPFS to store incremental data, smart contracts ensure the atomicity of the scoring. Based on the obtained processing score, a distributed storage architecture is established to convert the original image into IPFS hash storage and upload the metadata to the blockchain; A conflict resolution mechanism is established through smart contracts. When the same task receives two different scores within 10 minutes, consensus is triggered using the following formula: Receive final score ,in As a weight for the rater, Rate the ratings, Index for raters; By identifying error patterns, personalized improvement suggestions and teaching resource recommendations are generated. A visual feedback report is generated based on the final score. Error annotation: Color marks are overlaid on the original image. For example, red indicates a critical error, and yellow indicates a general error; The quantitative indicators are radar charts of scores in each dimension and curves comparing scores with the class average. Use DBSCAN to identify common error patterns and obtain a typical error pattern library; Given the similarity between user errors and resource coverage errors and the user's knowledge level, the probability of recommending resource i is obtained. The similarity between user errors and resource coverage errors is fused with the recommendation probability of resource i under the condition of user knowledge mastery to obtain the learning recommendation value, and teaching resources are recommended based on the learning recommendation value; A heatmap of class error distribution is generated based on the visual feedback report, and optimization suggestions are made for the teaching links corresponding to high-frequency errors. For example, if 30% of students have "circular arc connection errors", it is recommended to increase relevant practical training; Example 2: Figure 2 As shown in the figure, the method for batch processing and cloud-based collaborative scoring of paper ruler and compass drawing operations provided in this embodiment of the invention specifically includes the following steps: Step 1: Acquire images of paper ruler and compass drawings from multiple sources, perform resolution adaptation and intelligent edge detection, perform perspective correction and standardization, reconstruct the brightness components of the drawings, and perform vectorization. Step 2: Based on the completion of the task image processing, construct the geometric constraint network of the task image to automatically verify the complex topological relationships. After the verification is completed, construct the symbol dataset, detect and classify the symbol data, calculate the symbol feature values, and perform multi-scale detection. Step 3: Based on the completion of multi-scale detection, construct a directed acyclic graph, use CUDA to accelerate key operators, construct a knowledge graph of mapping specifications, dynamically adjust the scoring weights, establish a dynamic scoring mechanism, and calculate the processing score. Step 4: Based on the obtained processing scores, establish a distributed storage architecture, analyze and calculate the final scores, and generate personalized improvement suggestions and teaching resource recommendations through error pattern mining.

[0011] The above provides a detailed description of one embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and should not be considered as limiting the scope of the present invention. The above formulas are all dimensionless numerical calculations, and the formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world situation. The preset parameters in the formulas are set by those skilled in the art based on actual conditions and historical experience, and can be adjusted according to actual conditions. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. All equivalent changes and improvements made in accordance with the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A system for batch processing and cloud-based collaborative scoring of images from paper-based drafting work, characterized in that: Includes the following steps: Acquisition and processing module: Acquires work images of paper ruler and compass drafting from multiple sources, performs resolution adaptation and intelligent edge detection, performs perspective correction and standardization processing, reconstructs the brightness components of the work images and performs vectorization processing; Verification and detection module: Based on the completion of the task image processing, construct the geometric constraint network of the task image to automatically verify the complex topological relationships. After the verification is completed, construct the symbol dataset to calculate the symbol feature values, detect and classify the symbol data, and calculate the symbol feature values ​​for multi-scale detection. Adjust the evaluation module: Based on the completion of multi-scale detection, construct a directed acyclic graph, use CUDA to accelerate key operators, construct a knowledge graph of mapping specifications, dynamically adjust the scoring weights, establish a dynamic scoring mechanism, and calculate the processing score; Optimize the recommendation module: Based on the obtained processing scores, establish a distributed storage architecture, analyze and calculate the final scores, and generate personalized improvement suggestions and teaching resource recommendations through error pattern mining.

2. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for performing resolution adaptation is as follows: It can collect images of paper ruler and compass drafting operations, and supports multi-source image acquisition. It supports input from mobile phones, scanners and document cameras, and automatically identifies the device type through the device model field in the EXIF ​​information. Based on the collected images of paper ruler and compass drafting, resolution adaptation is performed. Images with resolutions lower than 1920×1080 are automatically super-resolution processed, and the image super-resolution ESRGAN model is used to magnify the magnification factor. Based on the acquisition of images from paper-based drafting work, intelligent edge detection is performed using the Canny algorithm, calculated through the following formula: Calculate the detection dynamic threshold and ,in, The maximum value of the image gradient; The Hough transform was used to detect the border of the workbook and extract the coordinates of the four corner points. , , , .

3. The system for batch processing and cloud-based collaborative scoring of images for paper ruler and compass drafting operations according to claim 1, characterized in that, The method for performing perspective correction is as follows: Based on the acquired images of paper ruler and compass drafting, perspective correction and standardization processes are performed, and the perspective transformation matrix is ​​calculated. The transformation error is minimized by using the least squares method, as shown in the formula. The minimum error was calculated.

4. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for reconstructing the brightness components of the image is as follows: Input the standardized image, convert it to HSV space, and extract the luminance component V for subsequent processing; The luminance component V is decomposed using a 3-level Haar wavelet decomposition, and the reconstruction formula is as follows: The reconstructed luminance components were calculated. ,in, The third layer of low-frequency coefficients is used to preserve the main lines. These are the high-frequency coefficients of the k-th layer, used to suppress noise. and These are the weighting coefficients. To optimize the value through experiments, It is a soft thresholding function, and ,in ; By using morphological closing operations and filling the breakpoints with an elliptic kernel, the structuring element B satisfies: in, These are the pixel coordinates; Based on the obtained standardized operation image, the operation image is vectorized. The Douglas-Peucker algorithm is used to simplify polygons with a simplification threshold of 2 pixels, and a DXF file is output.

5. The system for batch image processing and cloud-based collaborative scoring of paper ruler and gauge drafting operations according to claim 1, characterized in that, The method for automatic constraint verification is as follows: Based on the completion of detection, key point orientation is assigned, gradient histogram is calculated, and peak values ​​within ±15° of the main direction are taken. Key points are classified and identified using a CNN model to classify point types; Based on the completion of classification and recognition, a task image constraint network is constructed, with the detected geometric elements as nodes and the constraint relationships between elements as edges; Based on the obtained task image constraint network, a graph neural network is trained on the task image constraint network to construct a loss function. ,in To establish realistic constraint relationships, construct a ruler and compass drafting sample with annotations. Constraint verification is performed based on the trained task image constraint network. Parallel constraint verification: ; Vertical constraint verification: ; Tangency constraint verification: Where k is the slope, Preset parallel constraint threshold , The preset vertical constraint threshold is used; The equation of the line is Ax + By + C = 0. Let the coordinates be the center of the circle. Let the radius be the circle. This is a preset constraint threshold.

6. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for obtaining the symbolic feature values ​​is as follows: Construct a symbol dataset containing ruler and compass drafting symbols. Each symbol class contains variant samples. Enhance the symbol data by random rotation, scaling, and Gaussian blur. Based on the constructed symbol dataset, symbol data is detected and classified, and a contextual attention module is added, using the formula: The symbolic eigenvalues ​​are calculated. Where M is a mask matrix that filters out interference from irrelevant regions. As the current focus, For the candidate information available for query, The actual content corresponding to the available candidate information. Scaling factor for Dimensions The dot product of the target that needs to be focused on and the available candidate information; Multi-scale detection based on symbolic feature values.

7. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for constructing the directed acyclic graph is as follows: Construct a Directed Acyclic Graph (DAG): nodes represent processing steps, and edges represent dependencies; GPU resources are allocated based on preset weights according to the computational complexity of the nodes; Nodes are divided into high-complexity nodes and low-complexity nodes based on a pre-defined weight according to computational complexity. For high-complexity nodes, allocate ≥4GB of video memory; for low-complexity nodes, share GPU resources.

8. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for obtaining the processing score is as follows: Establish a dynamic scoring mechanism using the formula: The processing score is calculated, where, For the total number of rules, Let be the degree of compliance with the i-th rule. For compliance weighting, For rule indexing, The average error rate of the rule in historical assignments. The error rate threshold, This is the weight sensitivity coefficient.

9. The system for batch image processing and cloud-based collaborative scoring of paper ruler and compass drafting operations according to claim 1, characterized in that, The method for calculating the final score is as follows: Using IPFS to store incremental data, smart contracts ensure the atomicity of the scoring. Based on the obtained processing score, a distributed storage architecture is established to convert the original image into IPFS hash storage and upload the metadata to the blockchain; A conflict resolution mechanism is established through smart contracts. When the same task receives two different scores within 10 minutes, consensus is triggered using the following formula: Receive final score ,in As a weight for the rater, Rate the ratings, Index for raters.

10. The system for batch image processing and cloud-based collaborative scoring of paper ruler and gauge drafting operations according to claim 1, characterized in that, The method for error pattern mining is as follows: By identifying error patterns, personalized improvement suggestions and teaching resource recommendations are generated. A visual feedback report is generated based on the final score. Error annotation: Color marks are overlaid on the original image. The quantitative indicators are radar charts of scores in each dimension and curves comparing scores with the class average. Use DBSCAN to identify common error patterns and obtain a typical error pattern library; Given the similarity between user errors and resource coverage errors and the user's knowledge level, the probability of recommending resource i is obtained. The similarity between user errors and resource coverage errors is fused with the recommendation probability of resource i under the condition of user knowledge mastery to obtain the learning recommendation value, and teaching resources are recommended based on the learning recommendation value; Based on the visual feedback report, a heatmap of class error distribution is generated, and optimization suggestions are made for the teaching links corresponding to high-frequency errors.

Images