An automatic adaptation method and device for flowchart digitization

Abstract

The application discloses an automatic adaptation method and device for a flowchart digitization mode, and the method comprises the following steps: in step S1, an original drawing file is acquired, multi-dimensional features are extracted, and a drawing feature vector is established; in step S2, a multi-modal drawing classification model is constructed by taking the multi-modal features of the drawing as input and taking the preset category of the drawing as output, a data set is constructed based on historical drawing data, the multi-modal drawing classification model is trained by using the data set, and the original drawing file is classified by using the multi-modal drawing classification model; in step S3, a digitization mode library is established, a complete processing flow template, a parameter template and applicable parameters are configured for each digitization mode; in step S4, the parameters of each digitization mode are configured by using the drawing feature vector; the matching degree of the original drawing file and each digitization mode is calculated based on the classification result and the parameters; and the digitization mode is recommended based on the matching degree. The application has high digitization efficiency for different types of drawing files.

Description

Technical Field

[0001] This invention relates to the field of flowchart digitization technology, and in particular to an automatic adaptation method and apparatus for flowchart digitization. Background Technology

[0002] A nuclear power plant system flowchart is a detailed schematic diagram of the process system, showing the main equipment and auxiliary mechanical components required to perform the system functions, and explaining the connections between these equipment and components. The flowchart is an important working document for the management, operation, and maintenance of nuclear power plants.

[0003] Currently, flowcharts are stored in unstructured electronic file formats such as DWG and PDF, or in paper form. Unifying the digitization of flowcharts is a prerequisite for their better utilization and management. In the existing flowchart digitization process, when faced with different types of original drawings, such as high-definition vector PDFs, old scanned documents, CAD electronic files, and hand-drawn sketches, users need to manually judge and select the appropriate digitization method, such as fully automatic parsing, base map tracing, and AI-assisted recognition. This operation is cumbersome and highly dependent on experience. Digitization parameters, such as OCR threshold, recognition sensitivity, and image enhancement intensity, need to be repeatedly adjusted manually. Using the same set of parameters for drawings of different qualities leads to poor recognition results. At the same time, for batch processing of drawings, it is impossible to automatically select the optimal method for each drawing, resulting in low overall digitization efficiency. Summary of the Invention

[0004] In view of this, it is necessary to provide an automatic adaptation method and apparatus for flowchart digitization, so as to effectively solve the technical problem of low overall efficiency when digitizing different types of original drawings.

[0005] This invention provides an automatic adaptation method for digitized flowcharts, comprising the following steps: Step S1: Obtain the original drawing file and perform multi-dimensional feature extraction to establish the drawing feature vector; Step S2: Construct a multimodal drawing classification model using the multimodal features of the drawing as input and the preset category of the drawing as output; construct a dataset based on historical drawing data; train the multimodal drawing classification model using the dataset; and classify the original drawing file using the multimodal drawing classification model. Step S3: Establish a digitalization method library and configure a complete processing flow template, parameter template, and applicable parameters for each digitalization method; Step S4: Automatically configure the parameters of each digitization method using the drawing feature vector; calculate the matching degree between the original drawing file and each digitization method based on the classification results and the parameters; recommend the corresponding digitization method to the user based on the matching degree.

[0006] Preferably, step S1 specifically comprises: The system receives original drawing files uploaded by users, performs multi-dimensional feature extraction on the original drawing files, including format features, image quality features, content features, and metadata features, and establishes the drawing feature vector by combining the features of each dimension.

[0007] Preferably, step S2 specifically comprises: Drawings are classified according to two dimensions: drawing format and file format, and divided into multiple preset categories. Visual features are extracted using a convolutional neural network, and numerical features are processed using a fully connected network to obtain metadata features. The visual features and metadata features are concatenated and input into a classifier. The classifier outputs the preset categories and the confidence scores of each category, thus constructing the multimodal drawing classification model. The historical drawing data is preprocessed to extract multimodal features from the preprocessed historical drawing data, and the historical drawing data is classified according to preset categories to construct the dataset; the multimodal drawing classification model is trained using the dataset. The original drawing classification model is classified using a trained multimodal drawing classification model.

[0008] Preferably, step S3 specifically comprises: A digitization method library is established by sorting out different digitization methods. For each digitization method, a complete processing flow template, parameter template, and applicable parameters are configured. The processing flow template includes the processing steps of the corresponding digitization method, the parameter template includes drawing parameters related to digitization, and the applicable parameters include applicable conditions, historical success rate, and automation level score.

[0009] Preferably, step S3 further includes: Each digitization method in the digitization method library is assigned a version number and a version log to record updates and improvements to the digitization algorithm. New versions of digitization methods are compatible with the parameter configurations of older versions.

[0010] Preferably, in step S4, based on the classification results and various parameters, the matching degree between the original drawing file and each digitization method is calculated, specifically as follows: Based on the classification results and the applicable conditions of each of the digitization methods, the type confidence score is calculated, the efficiency coefficient is calculated based on the automation level score of the digitization method, the historical success rate coefficient is calculated based on the historical success rate of the digitization method, and the weighted sum of the type confidence score, efficiency coefficient, and historical success rate coefficient is calculated to obtain the matching degree.

[0011] Preferably, in step S4, recommending the corresponding digital method to the user based on the matching degree specifically involves: The system recommends appropriate digitization methods to users in descending order of matching degree. After the user accepts the recommendation or manually selects another digitization method, the system records the user's selection and automatically executes the corresponding digitization process template according to the user's selected digitization method and corresponding configuration parameters.

[0012] Preferably, the method further includes step S5: recording user feedback and adjustment behaviors during the digitization process to form a feedback dataset, and optimizing the multimodal drawing classification model, automatic parameter configuration of digitization methods, expansion of the digitization method library, and user personalized preference model based on the feedback dataset.

[0013] Preferred options also include: The system receives multiple drawing files uploaded by users in batches, selects an appropriate digitization method for each drawing file and adds it to the processing queue, and sets high priority for drawing files that need to be processed first; it uses multi-threading to process different drawing files in parallel and supports distributed processing; it automatically adjusts the number of parallel processing operations based on server load. The processing time for each drawing file is estimated based on the drawing feature vector and the corresponding digitization method; the processing progress of each drawing is monitored in real time, and the estimated remaining time is displayed. After multiple drawing files are processed in batches, the processing method, actual time consumption, automation rate, and number of graphic elements requiring manual intervention for each drawing file are statistically analyzed, and a summary report is generated. Drawings with an automation rate below the threshold are marked.

[0014] The present invention also provides an automatic adaptation device for flowchart digitization, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the automatic adaptation method for flowchart digitization.

[0015] Compared with existing technologies, the advantages of this invention are as follows: First, this invention extracts multi-dimensional features from drawing files, then constructs a multi-modal drawing classification model, dividing drawings into several categories for categorized processing. Next, a digitization method library is established, calculating the matching degree between each digitization method and the drawing file, and recommending highly matched digitization methods to the user. This invention's intelligent digitization method matching based on multi-dimensional features and a classification model avoids failures or inefficiencies caused by inappropriate method selection. Users do not need professional knowledge; they only need to upload drawings, and the system automatically completes the entire adaptation process, allowing even beginners to quickly get started. While automatically adapting digitization methods, adaptive parameter configuration is performed, significantly reducing manual debugging time and improving parameter setting accuracy. This invention allows for the processing of various types of drawings with a single system, eliminating the need to switch between multiple tools and improving management convenience. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of an embodiment of an automatic adaptation method for flowchart digitization provided by the present invention. Detailed Implementation

[0017] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0018] Example 1

[0019] Please see Figure 1 The automatic adaptation method for flowchart digitization in this embodiment specifically includes the following steps: Step S1: Obtain the original drawing file and perform multi-dimensional feature extraction to establish the drawing feature vector; Step S2: Construct a multimodal drawing classification model using the multimodal features of the drawing as input and the preset category of the drawing as output; construct a dataset based on historical drawing data; train the multimodal drawing classification model using the dataset; and classify the original drawing file using the multimodal drawing classification model. Step S3: Establish a digitalization method library and configure a complete processing flow template, parameter template, and applicable parameters for each digitalization method; Step S4: Automatically configure the parameters of each digitization method using the drawing feature vector; calculate the matching degree between the original drawing file and each digitization method based on the classification results and the parameters; recommend the corresponding digitization method to the user based on the matching degree.

[0020] The automatic adaptation method for flowchart digitization provided in this embodiment, hereinafter referred to as "this method", automatically selects the optimal digitization process and adaptively adjusts parameters by intelligently recognizing drawing features, achieving "one-click" intelligent adaptation and significantly improving the success rate and efficiency of digitization. The following provides a detailed explanation of each step.

[0021] Step S1: Extraction of multi-dimensional features from drawings It receives raw drawing files uploaded by users, supporting formats including PDF, DWG, DXF, JPG, PNG, TIFF, BMP, etc. It performs multi-dimensional feature extraction on the drawings, establishing a drawing feature vector for subsequent classification and parameter configuration. These multi-dimensional features specifically include format features, image quality features, content features, and metadata features. Format features include file type (e.g., PDF / DWG / image); whether it's vector (e.g., vector PDF or scanned PDF); whether it contains text layers (e.g., whether text is selectable in PDF); whether it contains layers (e.g., DWG / DXF); and whether it has multiple pages (i.e., the number of pages in the PDF). Image quality features include resolution (DPI), calculated by parsing file metadata or image size; sharpness, calculated using Laplacian variance to determine image blurriness (lower values ​​indicate greater blurriness); contrast, calculated by the variance of the image histogram (lower values ​​indicate poorer contrast); noise level, detected by wavelet transform or median filtering to determine noise density; tilt angle, detected by Hough transform to determine if the drawing is tilted; presence of wrinkles, stains, and creases, determined by texture analysis and anomaly detection; and fading degree, determined by analyzing pixel brightness distribution to determine if fading has occurred. Content features include primitive density (estimated by edge detection to determine the number of devices or unit area); line complexity (using Hough transform to detect the number of lines and calculate intersection density); text density (using OCR to pre-identify the number and distribution of text); presence of a title bar, legend, and explanatory area (detected by template matching); presence of hand-drawn traces (analyzing line smoothness and continuity); and legend type distribution (preliminarily identifying the proportion of common legends, such as valves, pumps, and instruments). Metadata features include system number, unit number, and version number that may be included in the filename; file creation time and modification time; and file size.

[0022] Step S2: Construction of Drawing Classification Model A training set was constructed based on a large amount of historical drawing data to train a multimodal drawing classification model. Drawings were divided into seven preset types, AG in this embodiment. Type A is a high-definition vector PDF, containing text layers, clear primitives, resolution >300 DPI, and no noise; Type B is a scanned PDF, image format, resolution 150-300 DPI, with slight noise; Type C is a DWG or DXF original design file, containing layer and block information; Type D is an old blueprint scan, resolution <150 DPI, faded, wrinkled, and with high noise; Type E is a hand-drawn sketch, with irregular lines, non-standard legends, and no standardized annotations; Type F is a mixed type, with some areas clear and some areas blurry, like a photograph of a drawing; Type G is an image exported from CAD, although in image format, with clear lines and standardized legends.

[0023] The multimodal drawing classification model employs a multimodal fusion algorithm: image feature extraction uses convolutional neural networks, such as ResNet50, to extract visual features; metadata features are processed using fully connected networks to handle numerical features; the visual and metadata features are concatenated and fused in a multimodal manner before being input into the classifier Softmax; the classifier outputs the drawing type and the confidence score for each category. The model supports incremental learning; new drawings can be added to the training set after their type has been manually confirmed, continuously optimizing the classification accuracy.

[0024] Step S3: Construction of the Digital Method Library Establish a digital processing method library, with each method corresponding to a complete processing flow and parameter template. Method 1: Fully Automatic Parsing, suitable for Type C; the processing flow includes parsing CAD layers, recognizing blocks, mapping to the legend library, extracting text, and generating digital drawings; the parameter template includes block mapping rules, layer filtering rules, text extraction threshold, and unit conversion coefficient. Method 2: Intelligent OCR + Semi-Automatic Tracing, suitable for Type A and some Type G; the processing flow includes extracting text layers, OCR recognition device encoding, loading the base map, aligning the drawing frame, and assisting in tracing; the parameter template includes OCR confidence threshold, base map transparency, drawing frame recognition sensitivity, and text position matching tolerance. Method 3: Image Enhancement + Base Map Tracing, suitable for Types B, D, and some Type F; the processing flow includes image denoising, contrast enhancement, tilt correction, super-resolution reconstruction, loading as a base map, and assisting in tracing; the parameter template includes filter strength, contrast stretching coefficient, correction angle range, and super-resolution multiplier. Method 4: AI-assisted recognition + manual completion, suitable for type E and some types F; the processing flow is AI recognition of identifiable elements, generation of initial drawings, manual completion of the remaining parts, and AI-assisted suggestions; the parameter template includes recognition confidence threshold, completion suggestion display method, and legend matching sensitivity. Method 5: Hybrid mode, suitable for complex drawings and multi-page drawings; the processing flow is drawing area segmentation, different methods for different areas, result merging, and cross-page splicing; the parameter template includes area segmentation rules such as partitioning by clarity, selection strategies for each area method, and splicing tolerance. Each method records its applicable conditions, processing steps, parameter suggestions, and historical success rate statistics.

[0025] Step S4: Adaptive parameter configuration, intelligent matching and recommendation Based on the feature vector of the drawing, the system automatically configures various parameters of the selected digitization method to achieve adaptive parameter optimization. Specifically, this includes adaptive image enhancement parameters, adaptive OCR parameters, adaptive base map parameters, and adaptive recognition parameters. Adaptive image enhancement parameters are as follows: Filter strength = base value × (1 + noise level coefficient), the higher the noise, the stronger the filter; Contrast stretching coefficient = base value × (1 + (1 - contrast)), the lower the contrast, the stronger the stretching; Correction angle = tilt angle detected by Hough transform; Super-resolution multiplier: automatically selected according to resolution, 4x for <100 DPI, 2x for 100-200 DPI, and disabled for >200 DPI. Adaptive OCR parameters are as follows: Recognition region division: adjusts the recognition window size according to text density, using a smaller window to improve accuracy for high density; Confidence threshold = base value × sharpness coefficient, setting a high threshold of 0.9 for high sharpness and a low threshold of 0.6 for blurry text; Language model: automatically selected based on filename and content, Chinese, English, or a mix of both; Deduplication strategy: strengthens deduplication when text density is high to avoid duplicate recognition. The base map parameters are adaptively set as follows: Base map transparency = base value × (1 - clarity coefficient). High transparency is set for clear drawings, making it easy to see drawn elements; low transparency is set for blurry drawings, requiring the base map to be clearly visible. Frame alignment sensitivity: High transparency for clear frames ensures strict alignment; low transparency for blurry frames allows for manual fine-tuning. Recognition parameters are adaptively set as follows: Minimum size for element recognition = base value × (1 / resolution coefficient). Lower resolution results in a larger size threshold, avoiding misidentification of small noise points. Connection point detection sensitivity is increased for lines with high complexity to ensure no connection points are missed. Legend matching tolerance: High tolerance is set for hand-drawn drawings; low tolerance is set for standardized drawings. Parameter configurations are saved in JSON format, forming a "parameter configuration package," associated with the drawing type, and support manual fine-tuning and saving as custom configurations.

[0026] After parameter configuration, intelligent matching and recommendation are performed. Based on the drawing classification results and parameter configuration, the system automatically selects the optimal digitization method and recommends it to the user. The matching degree between the drawing and each digitization method is calculated, considering factors such as: format compatibility (e.g., DWG cannot be used with method 2, so it is forcibly excluded); quality feasibility (e.g., low-resolution drawings are not suitable for automatic OCR using method 2); efficiency priority principle (choosing the method with the highest degree of automation while ensuring quality); and historical success rate (statistically calculating the success rate of various methods for this type of drawing, prioritizing the method with the highest success rate). The matching degree calculation formula is: Matching degree = Type confidence × 0.5 + Historical success rate × 0.3 + Efficiency coefficient × 0.2. The efficiency coefficient = Automation degree score, with method 1 being the highest and method 4 the lowest. The system automatically selects the method with the highest matching degree and displays the basis for the recommendation and expected results on the interface, such as "Recommended method 2, matching degree 95%, expected automation rate 85%, requires approximately 30 minutes of manual intervention." Alternative options, the method with the second highest matching degree, are also displayed for the user to choose from. Users can accept the recommendation or manually select other methods; the system records the user's selection for subsequent optimization.

[0027] Once the parameters and digitization methods are automatically configured, the process executes automatically. The digitization process is executed automatically according to the selected method and configured parameters. Method 1: The CAD parsing engine is invoked, layers and blocks are traversed, and converted into digital primitives according to mapping rules. An initial drawing is generated and automatically opened for user acceptance. Method 2: OCR is first run to extract text, and the recognition results are temporarily stored. Then, a base map is loaded, and the drawing frame is automatically aligned. Finally, the drawing tool is opened; the base map is loaded, and the OCR text is displayed in the corresponding position with a semi-transparent frame, which the user can click to confirm and fill in. Method 3: Image enhancement is performed according to the configured parameters, including noise reduction, contrast enhancement, and tilt correction. After processing, it is loaded as a base map, automatically aligned, and the drawing tool is opened. Method 4: The AI ​​recognition model is run, generating an initial drawing and displaying it with low transparency. The drawing tool is opened, and the user adds details based on this. The AI ​​provides real-time suggestions for adding details. Method 5: Region segmentation is performed first, and the corresponding methods are executed in parallel for each region. After completion, the results are automatically merged and spliced ​​across pages. A progress bar and intermediate results are displayed in real-time during execution. Users can pause, view logs, and intervene and adjust at any time.

[0028] Step S5: Human Feedback Learning Loop Record user feedback and adjustments throughout the digitization process to create a feedback dataset. This dataset includes records of method selection, parameter adjustment, correction workload statistics, user satisfaction ratings, and error type classifications. Method selection records: Record whether the user accepted the recommended method; if so, record the manually selected method. Parameter adjustment records: Record which parameters the user manually adjusted, such as adjusting the OCR threshold from 0.8 to 0.7 or the background transparency from 40% to 30%. Correction workload statistics: After digitization, count the number of manually corrected and manually redrawn elements, calculate the automation rate (number of automatically generated elements / total number of elements), and record the total time spent. User satisfaction ratings: After digitization, a rating pop-up appears (1-5 stars), with optional feedback. Error type classifications: Record the main types of recognition errors: legend errors, connection errors, and text errors.

[0029] Based on the feedback dataset, the classification model and parameter configuration are continuously optimized, specifically including classification model optimization, parameter configuration optimization, method library expansion, and personalized models. Classification model optimization: User-confirmed drawing types are used as new training data, and the classification model is retrained periodically. Parameter configuration optimization: Statistical analysis is performed on the optimal parameter combinations for different drawing types, and the default parameter template is updated; for example, for type B drawings, statistics show that an OCR threshold of 0.75 minimizes manual correction, so the default value is updated. Method library expansion: If a certain type of drawing frequently requires special processing, such as photographic drawings, new digitization methods or sub-methods are added. Personalized models: Personalized preference models are built for high-frequency users; for example, if a user prefers to manually fine-tune alignment, the alignment requirements are automatically relaxed.

[0030] Furthermore, this embodiment supports batch processing and queue management of multiple drawings uploaded simultaneously. It supports batch uploading of multiple drawings, with the system automatically selecting an adaptation method for each drawing and adding it to a processing queue for intelligent queuing. Users can set high priority for urgent drawings for priority processing. Resource allocation automatically adjusts the number of parallel processing operations based on server load. It supports multi-threaded parallel processing of different drawings to improve batch efficiency. It supports distributed processing with multiple servers collaborating. The system estimates the processing time for each drawing based on its complexity and processing method, monitors progress, and displays the processing progress of each drawing in real time, such as preprocessing, recognition, pending verification, and completed. The estimated remaining time is displayed next to each drawing, and filtering by status allows viewing only drawings awaiting verification. After batch processing is complete, a summary report is generated, including the processing method, time consumption, automation rate, and number of elements requiring manual intervention for each drawing. The summary report can be exported in Excel or PDF format. Drawings with an automation rate below a threshold are automatically marked as "requiring special attention."

[0031] Furthermore, this embodiment also implements version management for digitization methods. The digitization method library supports version management, evolving with technological advancements. Each digitization method has an independent version number, such as Method 2 v1.2, recording algorithm updates and improvements. The version log records the content of each update, such as "Optimized OCR recognition algorithm, text recognition rate improved by 5%". New versions of methods are compatible with the parameter configurations of old versions, ensuring the reproducibility of historical projects. It supports comparing the processing effects of new and old versions on the same drawing, displaying them side by side. It automatically calculates indicators such as recognition rate improvement and reduction in manual corrections. Users can choose whether to upgrade to the new version. A rollback mechanism is included; if the new version's performance is unsatisfactory, users can roll back to the old version with one click.

[0032] Compared to existing technologies, this method is accessible to all users. No specialized knowledge is required; users simply upload drawings, and the system automatically completes the entire adaptation process, allowing even beginners to quickly get started. Furthermore, intelligent matching based on multi-dimensional features and classification models ensures high accuracy in method selection, avoiding failures or inefficiencies caused by inappropriate method choices. Adaptive parameter configuration significantly reduces manual debugging time, improving parameter setting accuracy and recognition success rate. While maintaining quality, this method utilizes highly automated methods to improve batch processing efficiency. It also leverages a human feedback learning loop to make the system increasingly intelligent with use, continuously improving adaptation accuracy and creating a virtuous cycle. Queue management and priority settings ensure that urgent tasks are processed first, maximizing resource utilization.

[0033] The following example, using the digitization of different types of drawings from a nuclear power plant, illustrates the application of this method in a specific scenario.

[0034] Scenario 1: High-resolution vector PDF, RCV system diagram The user uploads "RCV_PID_Rev5.pdf". The system extracts features: Format: PDF vector, including text layer, resolution 600 DPI; Quality: High clarity, Laplacian variance 120, no noise, no skew; Content: Medium primitive density, complete frame and title bar; Metadata: Filename contains "RCV", file size 2.5MB. Classification model output: Type A, confidence 98%. System matching method 2: Intelligent OCR + semi-automatic tracing. Automatic configuration parameters: OCR threshold 0.9, due to high clarity; background image transparency 40%; high frame recognition sensitivity; The interface displays "Recommended method 2, matching degree 96%, estimated automation rate 85%, requires manual intervention for about 30 minutes". Alternative method 3 is also displayed, matching degree 75%. The user accepts the recommendation, and the system executes: extracting the text layer, OCR recognizing 85% of the device code, loading the background image and automatically aligning the frame. The user opens the drawing tool, traces the remaining part, and the OCR text is displayed as a semi-transparent frame. The user clicks to confirm the input. It took 28 minutes and had an automation rate of 88%.

[0035] Scenario 2: Old scanned document, RCP system diagram The user uploaded "RCP_system_scan.jpg". The system extracted features: JPEG image format; quality: 150 DPI resolution, low sharpness, Laplacian variance 35, noticeable noise, low contrast, 2-degree tilt; content: some legends faded, text blurred; classification model output: Type D, confidence 85%. The system used matching method 3: image enhancement + background tracing; automatic parameter configuration: medium filter intensity, denoising coefficient 1.2; contrast stretching coefficient 1.5; correction angle automatically calculated 2 degrees; super-resolution multiplier 2x, boosted to 300 DPI; background transparency 30%. The user accepted the recommendation, and the system performed: image enhancement, denoising, correction, super-resolution reconstruction, and loaded it as the background image. The user opened the drawing tool and traced the image; the system provided auxiliary lines. The process took 1.5 hours, with an automation rate of 60%.

[0036] Scenario 3: DWG original file, electrical primary diagram The user uploads "Electrical_OneLine.dwg". The system extracts features: format: DWG, containing layers and tiles; content: complete layer structure, standardized primitives; classification model output: type C, confidence level 99%. The system uses matching method 1, fully automatic parsing; automatic parameter configuration: tile mapping rules are based on a preset mapping table; layer filtering retains "Equipment" and "Pipeline" layers. The system executes the parsing: completed in 2 minutes, generating a digital drawing, which can be previewed. The user saves the file after verifying that it is correct.

[0037] Scene 4: Hand-drawn sketches The user uploaded a hand-drawn sketch "draft.png". The system extracted features: format: PNG; quality: 150 DPI resolution, irregular lines, traces of correction; content: no standard legend, hand-drawn symbols; classification model output: type E, confidence level 92%. The system used matching method 4: AI-assisted recognition + manual completion, with automatically configured parameters: recognition confidence threshold 0.6; completion suggestion display as a dotted line; high legend matching tolerance. The system's AI recognition identified approximately 40% of the recognizable elements, generating an initial drawing and displaying it semi-transparently. The user then opened the drawing tool and completed the drawing, with the AI ​​providing real-time suggestions, such as "This may be a valve." The process took 45 minutes.

[0038] Scenario 5: Batch Processing The administrator needed to digitize 50 drawings from a certain system. After batch uploading, the system automatically selected an adaptation method for each drawing: 35 drawings of type A, matching method 2; 10 drawings of type D, matching method 3; 3 drawings of type C, matching method 1; and 2 drawings of type E, matching method 4. The system processed them in parallel according to priority, with the 5 highest priority drawings processed first, and a progress bar was displayed in real time. The entire process was completed after 3 hours. A summary report was generated: total time 3.2 hours, average automation rate 82%, and 5 drawings with an automation rate below 60% were marked. The administrator exported the report and arranged subsequent processing.

[0039] Scenario 6: Feedback Learning In Scenario 2, the user manually lowered the OCR threshold from 0.9 to 0.8 to obtain more recognition results. The system recorded this adjustment. One month later, the system's statistical analysis found that for Type D drawings, setting the OCR threshold to 0.75 resulted in the least average manual correction. The system automatically updated the default parameter template for Method 3.

[0040] In the batch processing of scenario 5, three drawings were processed using methods that differed from the recommended approach. The system records these choices and adjusts the recommendation algorithm the next time similar drawings are processed.

[0041] Example 2

[0042] This embodiment provides an automatic adaptation device for flowchart digitization, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the automatic adaptation method for flowchart digitization described in Embodiment 1.

[0043] The automatic adaptation device for flowchart digitization provided in this embodiment is used to implement the automatic adaptation method for flowchart digitization. Therefore, the automatic adaptation device for flowchart digitization also has the same technical effects as the automatic adaptation method for flowchart digitization, and will not be described again here.

[0044] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of the present invention.

Claims

1. An automatic adaptation method for flowchart digitization, characterized in that, Includes the following steps: Step S1: Obtain the original drawing file and perform multi-dimensional feature extraction to establish the drawing feature vector; Step S2: Construct a multimodal drawing classification model using the multimodal features of the drawing as input and the preset category of the drawing as output; construct a dataset based on historical drawing data; train the multimodal drawing classification model using the dataset; and classify the original drawing file using the multimodal drawing classification model. Step S3: Establish a digitalization method library and configure a complete processing flow template, parameter template, and applicable parameters for each digitalization method; Step S4: Automatically configure the parameters of each digitization method using the drawing feature vector; calculate the matching degree between the original drawing file and each digitization method based on the classification results and the parameters; recommend the corresponding digitization method to the user based on the matching degree.

2. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, Step S1 specifically involves: The system receives original drawing files uploaded by users, performs multi-dimensional feature extraction on the original drawing files, including format features, image quality features, content features, and metadata features, and establishes the drawing feature vector by combining the features of each dimension.

3. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, Step S2 specifically involves: Drawings are classified according to two dimensions: drawing format and file format, and divided into multiple preset categories. Visual features are extracted using a convolutional neural network, and numerical features are processed using a fully connected network to obtain metadata features. The visual features and metadata features are concatenated and input into a classifier. The classifier outputs the preset categories and the confidence scores of each category, thus constructing the multimodal drawing classification model. The historical drawing data is preprocessed, multimodal features of the preprocessed historical drawing data are extracted, and the historical drawing data is classified according to preset categories to construct the dataset; The multimodal drawing classification model is trained using the dataset. The original drawing classification model is classified using a trained multimodal drawing classification model.

4. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, Step S3 specifically involves: A digitization method library is established by sorting out different digitization methods. For each digitization method, a complete processing flow template, parameter template, and applicable parameters are configured. The processing flow template includes the processing steps of the corresponding digitization method, the parameter template includes drawing parameters related to digitization, and the applicable parameters include applicable conditions, historical success rate, and automation level score.

5. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, Also includes: Step S3 further includes: Each digitization method in the digitization method library is assigned a version number and a version log to record updates and improvements to the digitization algorithm. New versions of digitization methods are compatible with the parameter configurations of older versions.

6. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, In step S4, based on the classification results and various parameters, the matching degree between the original drawing file and each digitization method is calculated, specifically as follows: Based on the classification results and the applicable conditions of each of the digitization methods, the type confidence score is calculated, the efficiency coefficient is calculated based on the automation level score of the digitization method, the historical success rate coefficient is calculated based on the historical success rate of the digitization method, and the weighted sum of the type confidence score, efficiency coefficient, and historical success rate coefficient is calculated to obtain the matching degree.

7. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, In step S4, recommending corresponding digital methods to the user based on the matching degree specifically involves: The system recommends appropriate digitization methods to users in descending order of matching degree. After the user accepts the recommendation or manually selects another digitization method, the system records the user's selection and automatically executes the corresponding digitization method processing template according to the user's selected digitization method and corresponding configuration parameters.

8. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, It also includes step S5, recording user feedback and adjustment behaviors during the digitization process to form a feedback dataset, and optimizing the multimodal drawing classification model, automatic parameter configuration of digitization methods, expansion of the digitization method library, and user personalized preference model based on the feedback dataset.

9. The automatic adaptation method for flowchart digitization according to claim 1, characterized in that, Also includes: The system receives multiple drawing files uploaded by users in batches, selects an appropriate digitization method for each drawing file and adds it to the processing queue, and sets high priority for drawing files that need to be processed first; it uses multi-threading to process different drawing files in parallel and supports distributed processing; it automatically adjusts the number of parallel processing operations based on server load. The processing time for each drawing file is estimated based on the drawing feature vector and the corresponding digitization method; the processing progress of each drawing is monitored in real time, and the estimated remaining time is displayed. After multiple drawing files are processed in batches, the processing method, actual time consumption, automation rate, and number of graphic elements requiring manual intervention for each drawing file are statistically analyzed, and a summary report is generated. Drawings with an automation rate below the threshold are marked.

10. An automatic adaptation device for digitizing flowcharts, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the automatic adaptation method for flowchart digitization as described in any one of claims 1-9.

Images