An intelligent adaptive timing chart analysis and structuring method

Abstract

This invention discloses an intelligent adaptive time series graph parsing and structuring method. The method preprocesses the input time series graph; identifies and extracts basic structural elements from the graph; extracts text information using optical character recognition (OCR); analyzes the identified signal lines, employing a neural network to extract signal state changes and key event points; detects arrows representing time series constraints, using an object detection network for arrow detection and classification; establishes associations between graphical elements and text; verifies and corrects the associations and identified elements using the time series graph's syntax rules and global layout information, filtering out unreasonable results; converts the time series relationship model into a standardized structured data format; and continuously optimizes overall recognition performance through interactive correction and feedback based on confidence levels. This invention significantly reduces the workload of engineers manually converting time series graphs, improves the accuracy of time series constraint extraction, and supports multiple time series graph formats and styles.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and image processing technology, specifically to an intelligent adaptive temporal graph parsing and structuring method. Background Technology

[0002] Timing diagrams are an indispensable tool in digital system design and analysis. They clearly describe the timing relationships between signals in a visual way, including the triggering order, duration, and synchronous / asynchronous relationships. Therefore, timing diagrams are widely used in many key areas such as chip design, protocol specification development, and interface definition. In the chip design process, timing diagrams are used to ensure that the signal interactions between various modules meet design requirements, thereby guaranteeing the normal operation of the chip.

[0003] However, traditional time series diagrams are typically presented as images, PDF files, or paper documents, which has significant limitations. Due to the lack of a machine-readable structured representation, these time series diagrams cannot be directly recognized and processed by automated tools, thus severely limiting the ability to perform automated verification and analysis.

[0004] In this situation, engineers have to manually interpret timing diagrams and convert them into formal descriptions for subsequent automated processing. This process is not only time-consuming and labor-intensive but also prone to errors. Engineers need to carefully analyze every key element in the timing diagram, such as signal waveforms, timing arrows, and constraint parameters, and accurately convert them into a formal language or data model. Due to the complexity of timing diagrams and the uncertainty of manual operation, omissions, misunderstandings, or errors can easily occur during the conversion process, leading to inaccurate results in subsequent verification and analysis, and potentially even introducing new design problems.

[0005] While some image recognition methods exist for identifying basic elements in charts, most are designed for general chart types and are not specifically optimized for the structure and semantics of time series diagrams. Time series diagrams possess unique characteristics, such as the shape and variation patterns of signal waveforms, the direction and meaning of timing arrows, and the labeling of constraint parameters; these elements carry important timing information. However, existing image recognition methods typically fail to accurately identify these key elements or establish the logical relationships between them.

[0006] Therefore, in order to improve the efficiency and accuracy of digital system design and verification, there is an urgent need for a technical solution that can automatically parse timing diagrams and generate structured data models to support the automated process of digital system design and verification. Summary of the Invention

[0007] Purpose of the invention: This invention provides an intelligent adaptive time series graph parsing and structuring method that can automatically extract information such as signal waveforms, timing constraints, and parameter annotations from time series graph images, and construct a structured time series relationship model.

[0008] Technical solution: The intelligent adaptive time series graph parsing and structuring method of the present invention includes the following steps:

[0009] (1) Preprocess the input time series image;

[0010] (2) Identify and extract the basic structural elements in the timing diagram, including horizontal signal lines and vertical time axes, and determine the basic grid structure based on the identified signal lines and time axes;

[0011] (3) Use optical character recognition technology to extract text information from the time sequence diagram;

[0012] (4) Analyze the identified signal lines and use neural networks to extract signal state changes and key event points;

[0013] (5) Detect arrows representing temporal constraints, and use an object detection network to detect and classify arrows;

[0014] (6) Establish the relationship between graphic elements and text;

[0015] (7) Construct a complete temporal relationship graph model, and use the syntax rules and global layout information of the temporal graph to verify and correct the relationships and identified elements generated in step (6) and filter out unreasonable results;

[0016] (8) Convert the time series relationship model into a standardized structured data format;

[0017] (9) After outputting the structured data format in step (8), the confidence of the recognition results recorded in steps (2) to (6) of the time series diagram is evaluated; a confidence threshold is set. If the confidence of all steps (2) to (6) is higher than the threshold, the output result of the time series diagram is considered to be correct; if the confidence of any step (2) to (6) is lower than the threshold, the time series diagram and its preliminary recognition result are added to the "to be reviewed pool"; the data in the "to be reviewed pool" is manually corrected internally, and a user feedback interface is provided to allow external users to report and correct recognition errors; correction data from all sources is collected and fed back to the corresponding processing module whose initial recognition confidence is lower than the threshold, for updating or retraining its internal model to continuously optimize the overall recognition performance.

[0018] Furthermore, the preprocessing in step (1) includes grayscale conversion, binarization, noise removal, and tilt correction.

[0019] Furthermore, when performing the feature extraction tasks in steps (2) to (5), a multi-scale pyramid representation of the image is constructed, and the results are fused after feature extraction at different scales to improve the robustness of recognition of elements of different sizes.

[0020] Furthermore, the implementation process of step (2) is as follows:

[0021] Signal line identification: Pixel projection is performed on the preprocessed image in the vertical direction, and the cumulative pixel value of each row is counted; candidate signal line regions are determined based on the peak positions appearing on the projection curve; connected component analysis and filtering are performed on the candidate regions to extract accurate horizontal signal lines;

[0022] Time axis recognition: First, using image segmentation technology, the image is divided into multiple regions by analyzing the local or global features of pixels, so that the set of pixels that constitute the potential vertical time axis is classified into one or more independent candidate regions and separated from other image content; then, these candidate regions are post-processed, and their geometric properties are evaluated for screening and verification, and finally the main vertical line segments in the image that conform to the definition of the time axis are extracted.

[0023] The basic grid structure is determined: the intersection points of signal lines and time axes on the image and the regions they form constitute the basic grid structure of the timing diagram.

[0024] Furthermore, the implementation process of step (3) is as follows:

[0025] Text detection algorithms are applied to locate regions containing text in an image. For each detected text region image block, an OCR engine is used for text recognition. The OCR recognition results are processed, including: spell correction based on a predefined dictionary, recognition and standardization of special time sequence symbols, and parsing of numerical values ​​and units. Based on the content characteristics of the text and its position in the image, each recognized text block is classified into signal name, time sequence parameter name, parameter value, and time unit.

[0026] Furthermore, the implementation process of step (4) is as follows:

[0027] Traverse all identified signal lines and time axes, and calculate their geometric intersections; extract a local image region of a predetermined size as the analysis unit, centered on each intersection point;

[0028] Each extracted local image region is used as input and fed into a pre-trained state classifier. The classifier outputs the corresponding signal state or key event type for each local image region, including high level, low level, rising edge, falling edge, and indeterminate state.

[0029] Furthermore, the implementation process of step (5) is as follows:

[0030] The YOLOv9 object detection model is used to process the input time series image to directly identify and locate arrow objects in the image, and output the bounding box of each detected arrow and its preliminary morphological classification, which includes left-pointing arrows, right-pointing arrows, and bidirectional arrows.

[0031] For each detected arrow, further analysis is performed on the image region within its corresponding bounding box. Through image processing or morphological analysis methods, the precise geometric features of the arrow are extracted, including its precise start point, end point, main axis direction, length, and specific shape details of the arrowhead, for use in subsequent association and model construction.

[0032] Furthermore, the implementation process of step (6) is as follows:

[0033] Proximity calculation and filtering: Calculate the spatial distance between each text region and each graphic element, and based on the distance threshold, filter out several candidate related graphic elements that are spatially closest to each text.

[0034] Association Feature Evaluation: Evaluate the association features between the text and each candidate graphic element, including relative layout and semantic matching; where relative layout analyzes the orientation and alignment of the text; semantic matching determines whether the text type and graphic element type meet the expected association rules;

[0035] Optimal association determination: A weighted score is used to calculate the total association score for each text-candidate graphic element pair; the graphic element with the highest score that exceeds the threshold for each text is selected as its final association object, and the association is recorded.

[0036] Furthermore, the implementation process of step (7) is as follows:

[0037] Create a structured signal object for each identified signal line;

[0038] Create a structured timing constraint object for each identified arrow, which contains the constraint type, the signal object acting on it or the specific event point on the signal, and the constraint value;

[0039] Based on the signal objects or event points connected to the start and end points of the timing constraint arrows, clarify the timing dependencies between them;

[0040] Integrate all signal objects and timing constraint objects and their dependencies into a unified, graph-structured or object-oriented timing relationship graph model.

[0041] Furthermore, the calibration data from all sources mentioned in step (9) includes internal calibration and calibration data from user feedback.

[0042] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention realizes the automated parsing of timing diagrams, which greatly reduces the workload of engineers manually converting timing diagrams; improves the accuracy of timing constraint extraction and reduces human error; supports multiple timing diagram formats and styles, and has good adaptability; generates standardized structured data models, which are convenient for integration with downstream verification tools; accelerates the design and verification process of digital systems and improves work efficiency. Attached Figure Description

[0043] Figure 1 This is a flowchart of the present invention;

[0044] Figure 2 This is an example of a timing diagram from a chip datasheet. Detailed Implementation

[0045] The present invention will now be described in further detail with reference to the accompanying drawings:

[0046] like Figure 1 As shown, the present invention provides an intelligent adaptive timing diagram parsing and structuring method. Through a multi-stage processing flow, including image preprocessing, signal line recognition, waveform state extraction, arrow detection and classification, text recognition and association, and timing relationship diagram construction, it achieves complete parsing of timing diagrams. It can process timing diagrams of various formats, extract key timing parameters and constraint relationships, and generate structured output, providing automated support for digital system design and verification.

[0047] Figure 2 This is an example timing diagram from a typical chip datasheet; this type of image is the input object that this invention aims to process. Specifically, it includes the following steps:

[0048] Step 1: Preprocess the input time series image to improve the accuracy of subsequent recognition.

[0049] Grayscale conversion: If the input is a color image, it is first converted into a single-channel 8-bit grayscale image using a standard weighted average method, such as the brightness formula Y = 0.299R + 0.587G + 0.114B, to reduce computational complexity and unify the processing basis.

[0050] Adaptive binarization: For each pixel, a binarization threshold T(x, y) is dynamically calculated based on the mean and standard deviation of pixels within its neighborhood window. The pixel value is compared with T(x, y) to generate a binary image that clearly separates the foreground and background. The window size and algorithm-specific parameters need to be set based on experience or experimentation.

[0051] Noise Removal: Eliminate isolated pixels or small unstructured breaks that may occur after binarization by applying morphological opening operations, which are defined as first eroding the image and then dilating it.

[0052] Tilt Correction: Detect the tilt angle θ by using Hough transform to detect the tilt angle distribution of the dominant line in the image, and take its peak or average value. After obtaining the angle θ, apply the rotation matrix from the affine transformation, using the image center or a specified point as the rotation center, to perform a rotation operation of angle θ.

[0053] Step 2: Identify the basic structural elements in the timing diagram.

[0054] Horizontal signal line identification and extraction: The preprocessed binarized time series image is vertically projected to generate a one-dimensional vertical projection histogram to count the number of foreground pixels in each row. Peak regions in this histogram are analyzed to locate potential dense rows of signal lines, which are then selected as candidate signal line regions. Subsequently, connected component analysis is applied to each candidate region to calculate the geometric properties of each connected component. Based on the characteristics of long, thin, and horizontally extending signal lines, filtering conditions are set to eliminate noise or non-signal line connected components, thereby extracting accurate horizontal signal lines and merging collinear breakpoints.

[0055] Vertical timeline identification and extraction: Optional preprocessing is applied to the image, such as applying vertical morphological operations or edge detection to enhance the vertical structure. Then, pixel-level image segmentation techniques are applied to classify image pixels into "timeline" and "background / other," which can be done using a pre-trained deep learning semantic segmentation model. Connectivity analysis is performed on the segmentation results to extract independent candidate timeline regions. By calculating their aspect ratio, principal direction, length, and other geometric properties, vertical line segments that meet the characteristics of a timeline are selected as the identified timelines.

[0056] Basic mesh structure determination: After extracting all horizontal signal lines and vertical time axes, the geometric intersection coordinates of each pair of signal lines and time axes are calculated. All the identified signal line sets, time axis sets, and their intersection sets are collectively defined as the basic mesh structure of the time series diagram. This structure provides a spatial framework for subsequent analysis.

[0057] Step 3: Use optical character recognition technology to extract text information from the time sequence diagram.

[0058] Text region detection: First, text detection algorithms are applied to locate regions in the image that contain text. Deep learning-based scene text detection models (such as EAST, DBNet, CRAFT, etc.) are used, which can directly output the bounding boxes or polygonal regions of text lines.

[0059] Optical Character Recognition (OCR): For each detected text region image block, an OCR engine is applied to perform text recognition. Mature open-source OCR libraries or commercial OCR services can be used. When configuring the OCR engine, you can specify the language, page segmentation mode, and may need to perform specific training or fine-tuning for common fonts and symbols in the time series diagram.

[0060] Post-processing and classification: The original OCR recognition results are post-processed to improve accuracy, including: spell correction based on a predefined dictionary; identification and standardization of special timing symbols; and parsing of numerical values ​​and units. Finally, based on the content features of the text and its position in the image, each recognized text block is classified into semantic types such as signal name, timing parameter name, parameter value, and time unit.

[0061] Step 4: Use a pre-trained convolutional neural network to identify signal waveform patterns and extract signal state changes and key event points.

[0062] Signal line and time axis intersection location and local image region extraction: Based on the basic grid structure determined in step 2, each identified horizontal signal line is traversed. For each signal line, its geometric intersection with all identified vertical time axes is determined sequentially. Centered on the coordinates of each calculated intersection point, a fixed-size (e.g., N×M pixels) local image region is extracted from the preprocessed time series image. This local image region captures the visual features of the signal around that specific time point.

[0063] Local Image Region Feature Extraction and State / Event Classification: Each extracted local image region is fed into a pre-trained state / event classifier. This classifier can be a deep learning-based model capable of directly learning features to distinguish different states from the raw pixels. The classifier's task is to classify the input local image features into predefined signal states (e.g., high level, low level, rising edge, falling edge, indeterminate / high impedance state) or key event types (e.g., data validity point, specific clock edge trigger point, etc.).

[0064] Waveform state sequence construction and key event point recording: For each signal line, state or event labels obtained from classifying the corresponding local image region are collected according to the order of its intersections with the time axis (usually from left to right). These labels are arranged in chronological order to form a discrete waveform state sequence for the signal line. Simultaneously, the intersection positions classified as "key event points" and their corresponding event types are recorded. By analyzing the changes in adjacent states in the state sequence, state transition points (such as the rising edge from low to high level) can be further identified. Ultimately, each signal line will be associated with a waveform state sequence describing its dynamic behavior and a set of time points marked with key events.

[0065] Step 5: Detect and classify the arrows representing timing constraints.

[0066] Arrow target detection and preliminary morphology classification: The preprocessed time-series image is input into a pre-trained YOLOv9 object detection model, which can directly identify and locate arrow objects in the image. The output of the object detection model is usually the bounding box coordinates of each detected arrow and a preliminary morphology classification label, which indicates the basic geometric shape of the arrow, such as predefined categories like "left-pointing single arrow", "right-pointing single arrow", "double-headed arrow", "vertical upward arrow", and "vertical downward arrow".

[0067] Fine-grained geometric feature extraction of arrows: For each arrow identified and located by the object detection model, a more detailed analysis is performed on the image region within its corresponding bounding box to extract precise geometric features. This includes: applying image processing techniques to accurately determine the position of the arrow's central axis and the arrowhead; calculating the precise coordinates of the arrow's start and end points; determining the direction of the arrow's main axis; and measuring the arrow's length. These fine-grained features are crucial for subsequently understanding the specific meaning and connection relationships of the arrows.

[0068] To handle time series graphs with different resolutions and complexities, a multi-scale pyramid representation of the image is constructed during the feature extraction tasks in steps 2 to 5. The results are then fused after feature extraction at different scales to improve the robustness of recognition of elements of different sizes.

[0069] Step 6: Establish the relationship between graphic elements and text.

[0070] Spatial Relationship Assessment: For each text region and each graphic element identified through the steps, calculate the shortest distance between the edge of the text region bounding box and the pixel set of the graphic element. By calculating the vector direction from the center point of the graphic element to the center point of the text region, analyze the orientation of the center point of the text region relative to the center point of the graphic element and the degree of alignment between the two in the horizontal or vertical direction, thereby determining the most spatially relevant candidate graphic element.

[0071] Using semantic features to infer association types: Semantic analysis is performed on the extracted text content to infer its functional type and the types of graphic elements that may be associated with it. By using keyword lists or pattern matching, it is possible to identify whether the text content represents signal names, time series parameter names, or signal states. Based on the identified text type, the expected types of associated graphic elements are preliminarily determined.

[0072] The comprehensive decision-making process determines associations: Based on the previously calculated distances, analyzed relative positions and alignment relationships, and inferred semantic matching degrees, a score is assigned to the potential associations between each text region and its candidate graphic elements. This scoring process uses a weighted summation method, where smaller distances, good alignments, and high semantic type matching receive higher scores. Finally, the graphic element with the highest score exceeding a predetermined threshold is selected as the determined association for each text region.

[0073] Step 7: Construct a complete time sequence diagram model. Use the syntax rules and global layout information of the time sequence diagram to verify and correct the relationships and identified elements generated in Step 6, and filter out unreasonable results.

[0074] Create signal objects: Create a signal object for each identified signal line, storing the associated signal name and the waveform state sequence obtained from the analysis.

[0075] Create constraint objects: Create a constraint object for each identified arrow, storing its constraint type, associated parameter values, and geometric information.

[0076] Establishing temporal dependencies: By mapping the start / end coordinates of the constraint object to the waveform event points of the corresponding signal object, the source / target events are recorded in the constraint object, thus clarifying the dependencies between signals.

[0077] Model integration: Organize all signal objects and constraint objects containing dependency links into a unified data structure to form the final time series graph model.

[0078] Step 8: Convert the constructed time-series relationship model into a standardized structured data format.

[0079] Designed to represent a standardized data model for time series relationships, including elements such as signals, events, and constraints; converts the time series relationship graph model into a defined data model format; generates output in different formats, such as JSON, XML, and dedicated time series description languages, according to application requirements; and provides interfaces with downstream verification tools to support automated time series verification.

[0080] Step 9: Support interactive correction and adaptive learning.

[0081] Comprehensive Confidence Assessment and Review-Pending Marking: After outputting the structured data format in step 8, for each processed time series graph, the recognition confidence of each element recorded in the key steps of structure extraction (step 2), text recognition (step 3), waveform analysis (step 4), arrow detection and classification (step 5), and association analysis (step 6) is determined. An independent confidence threshold is set for each step. If the confidence of the time series graph is higher than its respective threshold in all steps, the analysis result of the graph is initially considered highly reliable. Conversely, if the confidence of any key step is lower than the corresponding threshold, the time series graph and its current analysis result are automatically marked and sent to the "review pool".

[0082] The system collects calibration data from all sources, including internal calibration and user feedback. For internal calibration, the system maintains a "review pool" containing all time series graphs that fail to meet the confidence level after evaluation. Internal operators or designated review teams can access this pool to conduct detailed reviews and manual calibration of the parsing results. For user feedback calibration, the system provides a feedback interface for end users. When using the parsing results, if users find any errors or inaccuracies, even if the result is not automatically marked as low confidence, they can submit error reports and suggested calibration information through this interface.

[0083] Unified Recording: Whether it is data from internal calibration or calibration information collected through the user feedback interface, the system will record it in detail along with the original recognition results to form a high-quality labeled dataset in a unified format and mark its source.

[0084] Targeted Feedback and Model Update Triggering: The system integrates correction data from all sources. For each correction record, the system compares it with the original low-confidence identification result and accurately feeds this correction data back to the corresponding processing module that initially generated the low-confidence result or was pointed out as an error by the user. This feedback data will be used to expand the training set of the corresponding module for subsequent model updates or retraining.

[0085] Based on a feedback mechanism, high-quality labeled samples are collected from both internal correction and user feedback channels. The system not only collects the corrected data but also records the original low-confidence recognition results for each sample and which (or which) deep learning models performed poorly on that sample. This constructs a dynamic training sample library tailored to specific weaknesses.

[0086] Based on the sample library, the system can periodically, or after collecting a sufficient number of calibration samples for a specific model, trigger a fine-tuning or incremental learning process for that model. For example, if a large number of calibration samples involve recognition errors of a specific type of arrow or complex waveform transformation, these samples will be prioritized for optimizing the corresponding deep learning model. In this way, the model can learn and improve its performance in scenarios where it previously performed poorly, thereby continuously improving the overall system's recognition accuracy and robustness.

[0087] The above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on this technical solution are considered to be part of the technical concept proposed in this invention and fall within the scope of protection of the claims of this invention.

Claims

1. A smart adaptive timing chart parsing and structuring method, characterized in that, Includes the following steps: (1) Preprocess the input time series image; (2) Identify and extract the basic structural elements in the timing diagram, including horizontal signal lines and vertical time axes, and determine the basic grid structure based on the identified signal lines and time axes; (3) Use optical character recognition technology to extract text information from the time sequence diagram; (4) Analyze the identified signal lines and use neural networks to extract signal state changes and key event points; (5) Detect arrows representing temporal constraints, and use an object detection network to detect and classify arrows; (6) Establish the relationship between graphic elements and text; (7) Construct a complete temporal relationship graph model, and use the syntax rules and global layout information of the temporal graph to verify and correct the relationships and identified elements generated in step (6) and filter out unreasonable results; (8) Convert the time series relationship model into a standardized structured data format; (9) After the structured data format is output in step (8), the confidence of the recognition results of the time series diagram in steps (2) to (6) is evaluated; a confidence threshold is set. If the confidence of all steps (2) to (6) is higher than the threshold, the output result of the time series diagram is considered to be correct; if the confidence of any step (2) to (6) is lower than the threshold, the time series diagram and its preliminary recognition result are added to the "to be reviewed pool". The data in the "Pending Review Pool" is manually corrected internally, and a user feedback interface is provided to allow external users to report and correct recognition errors. Correction data from all sources is collected and fed back to the corresponding processing modules whose initial recognition confidence is lower than the threshold, so as to update or retrain their internal models and continuously optimize the overall recognition performance.

2. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The preprocessing in step (1) includes grayscale conversion, binarization, noise removal, and tilt correction.

3. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, When performing the feature extraction tasks in steps (2) to (5), a multi-scale pyramid representation of the image is constructed, and the results are fused after feature extraction at different scales to improve the robustness of recognition of elements of different sizes.

4. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (2) is as follows: Signal line identification: Pixel projection is performed on the preprocessed image in the vertical direction, and the cumulative pixel value of each row is counted; candidate signal line regions are determined based on the peak positions appearing on the projection curve; connected component analysis and filtering are performed on the candidate regions to extract accurate horizontal signal lines; Time axis recognition: First, using image segmentation technology, the image is divided into multiple regions by analyzing the local or global features of pixels, so that the set of pixels that constitute the potential vertical time axis is classified into one or more independent candidate regions and separated from other image content; then, these candidate regions are post-processed, and their geometric properties are evaluated for screening and verification, and finally the main vertical line segments in the image that conform to the definition of the time axis are extracted. The basic grid structure is determined: the intersection points of signal lines and time axes on the image and the regions they form constitute the basic grid structure of the timing diagram.

5. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (3) is as follows: Text detection algorithms are applied to locate regions containing text in an image. For each detected text region image block, an OCR engine is used for text recognition. The OCR recognition results are processed, including: spell correction based on a predefined dictionary, recognition and standardization of special time sequence symbols, and parsing of numerical values ​​and units. Based on the content characteristics of the text and its position in the image, each recognized text block is classified into signal name, time sequence parameter name, parameter value, and time unit.

6. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (4) is as follows: Traverse all identified signal lines and time axes, and calculate their geometric intersections; extract a local image region of a predetermined size as the analysis unit, centered on each intersection point; Each extracted local image region is used as input and fed into a pre-trained state classifier. The classifier outputs the corresponding signal state or key event type for each local image region, including high level, low level, rising edge, falling edge, and indeterminate state.

7. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (5) is as follows: The YOLOv9 object detection model is used to process the input time series image to directly identify and locate arrow objects in the image, and output the bounding box of each detected arrow and its preliminary morphological classification, which includes left-pointing arrows, right-pointing arrows, and bidirectional arrows. For each detected arrow, the image region within its corresponding bounding box is further analyzed, and the precise geometric features of the arrow are extracted through image processing or morphological analysis methods. This includes its precise start point, end point, main axis direction, length, and specific shape details of the arrowhead, for use in subsequent association and model building.

8. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (6) is as follows: Proximity calculation and filtering: Calculate the spatial distance between each text region and each graphic element, and based on the distance threshold, filter out several candidate related graphic elements that are spatially closest to each text. Association Feature Evaluation: Evaluate the association features between the text and each candidate graphic element, including relative layout and semantic matching; where relative layout analyzes the orientation and alignment of the text; semantic matching determines whether the text type and graphic element type meet the expected association rules; Optimal association determination: A weighted score is used to calculate the total association score for each text-candidate graphic element pair; the graphic element with the highest score that exceeds the threshold for each text is selected as its final association object, and the association is recorded.

9. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The implementation process of step (7) is as follows: Create a structured signal object for each identified signal line; Create a structured timing constraint object for each identified arrow, which contains the constraint type, the signal object acting on it or the specific event point on the signal, and the constraint value; Based on the signal objects or event points connected to the start and end points of the timing constraint arrows, clarify the timing dependencies between them; Integrate all signal objects and timing constraint objects and their dependencies into a unified, graph-structured or object-oriented timing relationship graph model.

10. The intelligent adaptive time series graph parsing and structuring method according to claim 1, characterized in that, The calibration data from all sources mentioned in step (9) includes internal calibration and calibration data from user feedback.

Images