A method and system for intelligent translation and information verification of DWG drawings

Abstract

This invention discloses an integrated method and system for intelligent translation and information verification of DWG drawings, belonging to the field of computer-aided design document processing technology. The method includes the following steps: generating a spatial fingerprint reflecting the layout hierarchy based on the coordinates, orientation, and font features of each text block in the DWG drawing; and establishing the layout hierarchy trajectory of the text blocks based on the spatial fingerprint to form continuous trajectory data. This invention constructs a stable hierarchical structure through spatial fingerprints, semantic islands, sequential chains, and local merging windows, ensuring paragraph splicing is performed within a controlled range, maintaining clear text hierarchy, and reducing semantic shifts caused by spatial aliasing. Simultaneously, it adjusts the merging threshold through time-slice driving and stabilizes key engineering information with a parameter-priority retention strategy, ensuring consistency in paragraph generation within a dynamic browsing environment and reducing the risk of semantic distortion.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design document processing technology, specifically to an integrated method and system for intelligent translation and information verification of DWG drawings. Background Technology

[0002] The integrated intelligent translation and information verification of DWG drawings refers to the simultaneous completion of multilingual intelligent translation of drawing text and automatic verification of drawing information in terms of accuracy, completeness, and location relevance within the same operational chain. This creates a closed loop in the translation process, from content extraction, semantic construction, and format restoration to result verification. The system directly parses the DWG source file, uniformly analyzing layer structure, entity types, and object attributes. It comprehensively identifies single-line text, multi-line text, attribute blocks, dynamic blocks, nested reference blocks, and translatable and recognizable elements such as engineering identifiers, numbering rules, dimension codes, and drawing name indexes. During the identifier parsing stage, it distinguishes and marks mixed alphanumeric identifiers, regularized encodings, version numbers, and professional symbols, filtering out non-translatable characters using regular expressions to prevent mistranslation of engineering identifiers. Based on this, the system reassembles fragmented statements, cross-block explanatory text, and descriptive content associated with identifiers according to the text's coordinate position, rotation angle, alignment, and geometric information such as text width and height, ensuring a complete and continuous semantic structure before entering the translation stage. During the translation phase, the system leverages industry terminology databases, abbreviation databases, identifier protection rules, and a self-developed AI translation engine to improve the accuracy and consistency of technical terms, long sentences, and descriptive text in engineering contexts. After the translation is backfilled, the system automatically verifies the correspondence between the source and translated texts, the consistency of identifiers, whether coordinate placement exceeds limits, whether rotation and scale are offset, whether formatting is consistent, whether reference blocks and nested structures are fully expanded, and whether there are any omissions or incorrect replacements. This mechanism of simultaneous translation and verification ensures that the accuracy of drawing language conversion and the reliability of drawing structure and engineering information are guaranteed at the same time, so that the final generated PDF and DWG translations maintain a high degree of consistency in readability, format fidelity, and engineering application security.

[0003] Existing technologies have the following shortcomings: In existing technologies, automatic paragraph merging of DWG drawings typically relies on spatial distance, alignment, and local geometric relationships between text blocks. However, in complex engineering drawing scenarios, text at different structural levels often exhibits extremely small spatial spacing due to compact layout, making it difficult for existing technologies to accurately identify the true semantic regions to which these texts belong. When the vertical or horizontal spacing of multiple text segments falls within the merging threshold range, the system may mistakenly classify texts originally intended for different equipment, components, or safety zones as belonging to the same semantic link and automatically splice them into continuous paragraphs. This type of mismerging can trigger a serious chain reaction during the translation stage, causing the translation engine to perform semantic inference in an incorrect context. This results in the incorrect replacement of technical terms and the incorrect rewriting of parameter descriptions, leading to semantic reversals or distortions in key engineering information such as equipment load levels, dimensional parameters, tolerance ranges, and safety zone levels.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an integrated method and system for intelligent translation and information verification of DWG drawings, so as to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an integrated method for intelligent translation and information verification of DWG drawings, comprising the following steps:

[0007] Based on the coordinates, orientation, and font features of each text block in the DWG drawing, a spatial fingerprint reflecting the layout hierarchy is generated, and the layout hierarchy trajectory of the text block is established according to the spatial fingerprint to form continuous trajectory data for subsequent structural attribution determination.

[0008] Based on the layout hierarchy trajectory, adjacent text blocks are divided into multiple semantic islands according to their structural affiliation, and a sequence chain that maintains the original reading order is generated within each semantic island to limit the splicing range of subsequent paragraph merging.

[0009] Within the concatenation range of the sequential chain, the beginning and end positions of the paragraph are marked according to the cross-level position changes, and a local merging window is generated based on the position markings. This allows the paragraph merging process to be executed in a restricted manner within the local merging window, thereby achieving spatial isolation of cross-level text.

[0010] Around the local merging window, candidate text blocks are grouped according to text block length, punctuation rhythm and engineering vocabulary density. Text blocks that are spatially adjacent but have different structural levels are excluded from the same local merging window to form a candidate text set with controllable boundaries.

[0011] By utilizing a time-slice-driven paragraph generation engine, the local merging window dynamically adjusts the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. Furthermore, a parameter-priority retention strategy is used to suppress cross-level splicing behavior, thereby reducing the risk of semantic distortion in engineering caused by paragraph mismerging.

[0012] Preferably, the steps for generating the layout hierarchy trajectory are as follows:

[0013] The text blocks are organized based on their coordinate information, orientation information, and font features, so that the text blocks form distinctive attribute descriptions on the drawing plane;

[0014] Spatial fingerprints are constructed based on the organized attribute descriptions, enabling each text block to form an independent identifier that can express layout relationships;

[0015] Based on the positional relationship, orientation synchronization relationship and font presentation trend between spatial fingerprints, a continuous association is made so that multiple spatial fingerprints can form a layout hierarchy trajectory that can express hierarchical relationships.

[0016] The sequential connection results based on the layout hierarchy trajectory are continuously integrated to form continuous trajectory data of the spatial relationship between text blocks, thereby enabling the overall layout structure in the drawing to have a trajectory expression basis for subsequent processing.

[0017] Preferably, the steps for dividing adjacent text blocks into semantic islands and forming sequential chains based on the layout hierarchy trajectory are as follows:

[0018] Based on the direction and orientation changes of the layout hierarchy trajectory, the distance trend between text blocks and the continuity of font presentation, the text blocks in the layout hierarchy trajectory are initially classified so that the text blocks that appear continuously and have coherent characteristics form basic aggregation units.

[0019] Based on the proximity, convergence, and reading extension trends of the basic aggregation units in the layout structure, multiple basic aggregation units are further aggregated to form semantic islands, so that the semantic islands have a stable reading direction and internal content concentration.

[0020] Based on the arrangement order, reading trend, font relationship and content logic of text blocks within the semantic island, the text blocks are sorted to form a sequence chain that maintains the original reading order, so that the sequence chain has a continuous reading path;

[0021] Based on the boundaries formed by the sequential chain, the scope of the sequential chain is used as the concatenation scope, so that paragraph merging is performed within the concatenation scope and cross-semantic island text blocks are excluded.

[0022] Preferably, based on the relative position and font size relationship of text blocks within the semantic island, the text blocks are sorted according to the reading trend from top to bottom or from left to right, and the spatial position and semantic structure of the text blocks are kept consistent during the sorting process to ensure that the reading path of the text blocks within the sequential chain has natural fluidity.

[0023] Preferably, the steps for marking the start and end positions of segments based on the concatenable range of the sequential chain and generating a local merge window are as follows:

[0024] Based on the positional relationship of text blocks in the sequential chain, the reading extension direction, the hierarchical change trend of the layout hierarchy trajectory, and the displacement gradient with adjacent text blocks, the regions where cross-level jumps occur between text blocks are identified, and the front and back ends of the abrupt change regions are marked as the beginning and end positions of the paragraph, respectively.

[0025] Based on the position of the beginning and end of the paragraph, a splicable range is formed, and the text blocks within the splicable range are ensured to have a stable structure and order;

[0026] Based on the spatial regions of the beginning and end of the segment, a local merging window is constructed, and the local merging window is made spatially consistent with the sequence chain to ensure that the merging process is limited to the window.

[0027] By using the local merge window as the sole merge execution area, the paragraph merging process is ensured to take place within this window. Furthermore, the window boundaries enable spatial isolation of text across different levels, thus providing each local merge window with a clear structural definition.

[0028] Preferably, when constructing a local merge window, the boundary of the local merge window is adjusted synchronously with the reading extension direction of the sequential chain, so that the window maintains continuity in the extension direction, and a restricted interval is formed in the vertical direction based on the distance between the beginning and end of the paragraph, so that the text blocks inside the window continue to maintain an independent structure in space, and cross-level text remains isolated outside the window.

[0029] Preferably, the steps for forming a candidate text set with controllable boundaries around the local merging window are as follows:

[0030] Based on the generated local merged window, all candidate text blocks inside the window are obtained, and the candidate text blocks are initially divided according to the length of the text blocks, so that the candidate text blocks form a set with a partition structure.

[0031] Based on the preliminary segmentation results, each candidate text block is classified according to the punctuation rhythm within it, so that the differences between the candidate text blocks in terms of reading style and semantic completeness can be further expressed.

[0032] Based on the punctuation rhythm classification results, the candidate text blocks are regrouped according to the engineering vocabulary density within them, so that candidate text blocks with different structural levels are separated at the semantic level.

[0033] Based on the results of length division, punctuation rhythm classification and engineering vocabulary density grouping, and combined with the spatial constraints of the local merging window, candidate text blocks that are spatially adjacent but have different structural levels are excluded from the window range, so that the candidate text set forms a set with controllable boundaries.

[0034] Preferably, when forming a candidate text set with controllable boundaries, candidate text blocks with continuous text block length and punctuation rhythm and high engineering vocabulary density are preferentially retained within the local merging window. This ensures that the candidate text set remains consistent in structural hierarchy and further excludes low-density text blocks that are spatially adjacent from the set, thereby forming a stable source of text blocks in terms of semantic expression.

[0035] Preferably, using a time-slice-driven paragraph generation engine to dynamically adjust the merging threshold and suppress cross-level splicing behavior includes the following steps:

[0036] Based on the formed local merging window and the candidate text set inside the local merging window, a continuous time slice is constructed for the paragraph generation engine, so that the paragraph generation engine reads the candidate text set at a fixed rhythm in each time slice.

[0037] Based on the rhythm changes of time slices, the system monitors drawing browsing behavior, zooming actions, and view dwell time, so that these behaviors trigger dynamic adjustment of the merging threshold under the action of time slices.

[0038] Based on the adjusted merging threshold, the candidate text set inside the local merging window is combined with the new threshold within each time slice, so that the paragraph generation engine decides whether the candidate text block enters the splicing process based on the latest threshold.

[0039] The paragraph generation engine implements a parameter-priority retention strategy for splicing behavior, which ensures that text blocks containing key engineering attributes are retained in multiple rounds of threshold adjustment, thereby suppressing cross-level splicing behavior and reducing the risk of engineering semantic distortion.

[0040] A DWG drawing intelligent translation and information verification integrated system includes a spatial fingerprint generation module, a semantic island construction module, a local merging window generation module, a candidate text grouping and filtering module, and a dynamic paragraph generation and control module.

[0041] The spatial fingerprint generation module generates a spatial fingerprint that reflects the layout hierarchy based on the coordinates, orientation, and font features of each text block in the DWG drawing. It also establishes the layout hierarchy trajectory of the text block based on the spatial fingerprint to form continuous trajectory data for subsequent structural attribution determination.

[0042] The semantic island construction module divides adjacent text blocks into multiple semantic islands according to their structural affiliation based on the layout hierarchy trajectory, and generates a sequence chain within each semantic island that maintains the original reading order in order to limit the splicing range of subsequent paragraph merging.

[0043] The local merge window generation module marks the beginning and end positions of paragraphs based on cross-level position changes within the splicable range of the sequential chain, and generates a local merge window based on these position marks. This restricts the paragraph merging process within the local merge window, achieving spatial isolation of cross-level text.

[0044] The candidate text grouping and filtering module groups candidate text blocks around the local merging window based on text block length, punctuation rhythm, and engineering vocabulary density. It excludes text blocks that are spatially adjacent but have different structural levels from the same local merging window, thus forming a candidate text set with controllable boundaries.

[0045] The dynamic paragraph generation and control module uses time slices to drive the paragraph generation engine, enabling the local merging window to dynamically adjust the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. It also suppresses cross-level splicing behavior through a parameter priority retention strategy, thereby reducing the risk of engineering semantic distortion caused by paragraph mismerging.

[0046] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0047] This invention constructs spatial fingerprints, semantic islands, sequential chains, and local merging windows to clearly present the hierarchical relationships between text blocks within the drawing structure. Paragraph generation is constrained within controlled boundaries, and cross-level text maintains stable separation even when spatially compact. This mechanism endows the paragraph splicing process with spatial self-organization capabilities, freeing the text structure in engineering drawings from geometric aliasing. This provides the translation engine with a clear reading path, stabilizes the semantic transmission link, and reduces the risk of semantic shifts caused by paragraph merging errors.

[0048] This invention introduces a time-slice-driven approach, enabling the local merging window to automatically adjust the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. It also employs a parameter-priority retention strategy to protect critical engineering text, ensuring robust paragraph generation in dynamic environments. This continuous adjustment allows the paragraph construction process to adapt to changes in the user's focus area, maintaining consistency in structural boundaries through multiple adjustments. It prevents critical engineering information from being absorbed into the splicing chain by irrelevant content, thereby reducing the impact of semantic distortion on engineering applications. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a flowchart of a method for integrating intelligent translation and information verification of DWG drawings according to the present invention.

[0051] Figure 2 This is a schematic diagram of a module of an integrated system for intelligent translation and information verification of DWG drawings according to the present invention. Detailed Implementation

[0052] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0053] This invention provides, for example Figure 1 The method for intelligent translation and information verification of DWG drawings, as shown, includes the following steps:

[0054] Based on the coordinates, orientation, and font features of each text block in the DWG drawing, a spatial fingerprint reflecting the layout hierarchy is generated, and the layout hierarchy trajectory of the text block is established according to the spatial fingerprint to form continuous trajectory data for subsequent structural attribution determination.

[0055] To generate a spatial fingerprint that reflects the layout hierarchy of text blocks in the DWG drawing, and to further establish a layout hierarchy trajectory for subsequent structural attribution determination, the specific implementation steps are as follows:

[0056] For each text block appearing in the DWG drawing, its coordinate information, orientation information, and font features are acquired and organized, and these three types of information are integrated to give each text block a distinctive attribute description on the drawing plane. In this process, coordinate information reflects the specific position of the text block in two-dimensional space, orientation information describes the angle the text block faces, and font features reflect the differences in the visual presentation of the text block, such as character shape, width, height, and spacing. By extracting and associating these features in a unified way, each text block has a basic geometric and visual foundation that facilitates subsequent associative processing. This step not only provides basic data for subsequent steps but also transforms text blocks from independent point objects into comparable structured objects.

[0057] After acquiring the coordinates, orientation, and font features of the text blocks, these features are constructed into spatial fingerprints according to preset combination rules. This allows each text block to form an independent identifier in the drawing space that can express layout relationships. This spatial fingerprint not only contains the original feature information but also extends to the relative relationship between the text block and its surrounding space, giving it the ability to describe the structure of a local area. During the construction of the spatial fingerprint, the elements within it are combined in a fixed order, allowing the similarities and differences between text blocks to be identified at a higher level. In this way, each text block is no longer a single point but possesses a visual symbol to express its spatial structure, enabling subsequent steps to further perform association judgments based on the spatial fingerprint. Because each text block has its own spatial fingerprint, the overall drawing gradually forms a structural set composed of multiple spatial fingerprints.

[0058] After constructing the spatial fingerprints, a layout hierarchy trajectory is generated by continuously associating the spatial fingerprints of adjacent text blocks, thus creating a layout hierarchy trajectory that reflects the hierarchical relationship of the text blocks within the layout. This step associates factors such as proximity, orientation synchronization, and font presentation trends between spatial fingerprints in a consistent manner, so that the text blocks are no longer distributed as isolated points, but rather connected in a linear or tree-like trajectory. By progressively connecting the relative distribution relationships between spatial fingerprints, the arrangement direction, reading direction, and structural hierarchy of the text blocks gradually become clear, thereby forming a complete layout hierarchy trajectory. In this process, each layout hierarchy trajectory not only reflects the relative position between text blocks but also embodies the natural order of the text blocks in the layout structure, giving the trajectory extensibility and structural expressiveness. Since the trajectory is formed by sequentially connecting multiple spatial fingerprints, each trajectory can characterize the hierarchical trend formed by the text blocks in the layout structure.

[0059] After the layout hierarchy trajectory is formed, it is continuously integrated to present the spatial relationships related to text blocks in the entire drawing as continuous trajectory data. This continuous trajectory data is formed by connecting multiple layout hierarchy trajectories sequentially, allowing the overall spatial distribution structure of text blocks in the entire drawing to be fully expressed. The continuous trajectory data not only expresses the associated positions between text blocks, but also their layout direction, visual presentation trend, and hierarchical relationship in local areas, enabling subsequent processing to make structural attribution judgments based on the trajectory. The formation process of continuous trajectory data transforms text blocks from a discrete set of points on a two-dimensional plane into a continuous structure that can be read from the perspective of the overall layout, providing a holistic spatial foundation for subsequent processing steps. Through the above process, a continuous transformation of text blocks from their original geometric and visual features to the overall structure of the layout trajectory can be achieved, allowing subsequent steps to be carried out with complete spatial semantic clues.

[0060] Based on the layout hierarchy trajectory, adjacent text blocks are divided into multiple semantic islands according to their structural affiliation, and a sequence chain that maintains the original reading order is generated within each semantic island to limit the splicing range of subsequent paragraph merging.

[0061] To divide adjacent text blocks into multiple semantic islands based on their structural affiliation according to the layout hierarchy trajectory, and to generate a sequence chain within each semantic island that maintains the original reading order to limit the scope of subsequent paragraph merging, a multi-step approach can be implemented using the resulting continuous layout hierarchy trajectory. This approach further strengthens and refines the inherent relationships between text blocks in terms of spatial structure, content association, and reading direction. The specific implementation steps are as follows:

[0062] After acquiring the resulting layout hierarchy trajectory, the text blocks within the trajectory are initially categorized based on their structural cohesion through comprehensive observation of their spatial orientation, changes in orientation, distance trends between text blocks, and the continuity of font appearance. This initial categorization is based on the arrangement of text blocks within the trajectory, forming basic aggregation units from those text blocks that appear continuously, maintain stable orientation, exhibit consistent font features, and show a unidirectional extension trend in their positional distribution. In this process, by extending the continuity information within the layout hierarchy trajectory to higher levels, each basic aggregation unit acquires a preliminary semantic convergence trend, providing a prerequisite for the formation of semantic islands. Once the basic aggregation units are formed, the relationship between text blocks is no longer a simple linear trajectory connection, but rather forms a higher-level aggregation foundation capable of expressing local structural stability.

[0063] After forming multiple basic aggregation units, based on their proximity, convergence of direction, and the reading extension trend of the text content structure, these units are further aggregated into semantic islands. Each semantic island represents a local semantic region in the drawing. In this process, the formation of semantic islands relies not only on the continuous characteristics of the layout hierarchy but also on the visual organization and content arrangement relationships between text blocks, making the resulting semantic islands more aligned with the reading habits and information construction methods in actual engineering drawings. In this way, each semantic island possesses a highly consistent reading direction, a stable arrangement trend, and a high degree of semantic concentration within its content, making it a basic structural unit that can be used for subsequent paragraph merging. After the formation of semantic islands, the previously ambiguous hierarchical relationships between text blocks are clearly divided into multiple blocks with well-defined boundaries, effectively expressing the structural hierarchical relationships of the text blocks.

[0064] After obtaining multiple semantic islands, to ensure that subsequent paragraph merging follows a natural reading order within each semantic island, the text blocks within each semantic island are comprehensively sorted according to their arrangement order in the layout hierarchy trajectory, the top-to-bottom or left-to-right reading trend of the text blocks in their local areas, font size relationships, and the arrangement logic of the text content, forming a sequence chain that maintains the original reading order. The generation of the sequence chain relies not only on the directional information of the layout hierarchy trajectory but also on the spatial structure within the semantic islands, enabling the text blocks to form a continuous reading path within the sequence chain. In this step, the arrangement of text blocks within the sequence chain maintains the natural reading extension trend within the semantic islands, giving the reading path structural continuity and content coherence, providing a stable operational foundation for subsequent paragraph merging. After the sequence chain is formed, each semantic island is no longer a static collection of text blocks but a structured sequence with an internal reading direction.

[0065] After the sequential chain is formed, by using the sequential chain as a boundary framework and its range as the splicable range for subsequent paragraph merging, subsequent processing can be strictly confined to within each semantic island, preventing text blocks that cross semantic islands from entering the merging process. In this process, the sequential chain not only provides a reading path but also spatial boundaries, giving paragraph merging a clear structural definition. In this way, text blocks will not be mistakenly classified as the same paragraph due to spatial proximity in subsequent processing, ensuring consistency between reading logic and structural boundaries. Once the sequential chain is complete, each semantic island possesses continuity characteristics extending from the layout hierarchy, forming a stable framework in both space and content that can be used for paragraph integration, laying a solid foundation for subsequent operations.

[0066] Within the concatenation range of the sequential chain, the beginning and end positions of the paragraph are marked according to the cross-level position changes, and a local merging window is generated based on the position markings. This allows the paragraph merging process to be executed in a restricted manner within the local merging window, thereby achieving spatial isolation of cross-level text.

[0067] To mark the beginning and end positions of paragraphs based on cross-level positional changes within the combinable range of the sequential chain, and further generate local merging windows based on these positional marks, enabling the paragraph merging process to be executed within the local merging window to achieve spatial isolation of cross-level text, this can be accomplished through multiple continuously associated technical steps based on the formed semantic islands and sequential chains. The specific implementation process is as follows:

[0068] Based on the established sequential chain, for each text block within the sequential chain, its positional relationship within the sequential chain is determined according to its reading extension direction in the semantic island, the hierarchical change trend in the layout trajectory, and the displacement gradient with adjacent text blocks. This positional relationship is then used to identify regions where cross-hierarchical jumps may occur. When there are abrupt changes in the spatial arrangement trend between text blocks, such as a sudden increase in height difference, a sharp jump in lateral offset, discontinuous orientation, or an abnormal combination of font sizes, these changes can be considered potential signals of cross-hierarchical positional changes. Based on this, by precisely marking each potential change position, the position at the front of the abrupt change region is marked as the segment beginning position, and the position at the back of the abrupt change region is marked as the segment end position, thus dividing the reading path in the sequential chain into multiple segments with natural boundaries. In this way, the sequential chain is no longer considered a completely homogeneous continuous sequence, but rather forms a segmented structure with a clearer hierarchical structure.

[0069] After obtaining multiple markers for the beginning and end positions of paragraphs, these markers are used as boundaries to segment the text blocks within the sequential chain, creating concatenable regions for subsequent splicing within each segment defined by its beginning and end positions. In this step, the formation of concatenable regions relies not only on the linear arrangement within the sequential chain but also on the reading trends within semantic islands, the spatial matching relationships between text blocks, and the extension direction in the layout hierarchy trajectory. This ensures that each concatenable region possesses a stable internal structure, allowing subsequent operations to be performed within clearly defined areas. The formation of concatenable regions not only helps avoid cross-regional merging but also allows text blocks within each region to participate more naturally in the subsequent paragraph generation process. In this process, the markers for the beginning and end positions ensure the continuity of the text block order within the concatenable regions, providing each region with a complete semantic extension basis.

[0070] After the splicable range is formed, a local merging window is constructed to control paragraph merging behavior by using the spatial area between the paragraph beginning and end positions as the core boundary. This allows the local merging window to form a closed or semi-closed operational area around the sequence chain in a spatial dimension. In this step, the range covered by the local merging window depends not only on the distance between the paragraph beginning and end positions but also on the overall spatial layout of the semantic island where the sequence chain is located. This ensures that the local merging window maintains a consistent extension trend with the sequence chain in space and dynamically expands or contracts according to the shape of the semantic island when necessary. By using the paragraph beginning position as the starting boundary and the paragraph end position as the ending boundary of the local merging window, the local merging window has clear operational restrictions, preventing cross-level text from entering the current area beyond the local merging window. The core of this step is to enable the local merging window to accurately express the natural segmentation relationship in the sequence chain, making each window an independent paragraph merging area.

[0071] After the local merge window is formed, by using it as the sole execution area for paragraph merging, the paragraph merging process is spatially confined within that window. The window boundaries also achieve spatial isolation of text across different levels, preventing text blocks from other semantic islands or levels from entering the current window. In this way, merging occurs only between text blocks covered by the local merge window, maintaining structural stability in the splicing relationships between text blocks. Each local merge window can handle splicing operations within a specific semantic range, allowing subsequent processing to proceed within a stable structural framework. Based on this, the structure comprised of paragraph beginning and ending positions, the splicable range, and the local merge window ensures clear hierarchical division of text blocks in complex engineering diagrams, providing comprehensive support for paragraph construction in both spatial and reading structures.

[0072] Around the local merging window, candidate text blocks are grouped according to text block length, punctuation rhythm and engineering vocabulary density. Text blocks that are spatially adjacent but have different structural levels are excluded from the same local merging window to form a candidate text set with controllable boundaries.

[0073] To group candidate text blocks around a local merging window based on block length, punctuation rhythm, and engineering vocabulary density, and to exclude spatially adjacent but structurally different text blocks from the same local merging window, thus forming a candidate text set with controllable boundaries, multiple interconnected technical steps can be implemented based on the established local merging window. This allows candidate text blocks to form a stable set under multiple constraints of structural hierarchy, semantic features, and reading habits, thereby providing a more reliable text source for the subsequent paragraph merging process. The specific implementation steps are as follows:

[0074] Within the generated local merge window, all candidate text blocks are retrieved and preliminarily divided according to their length. The length of each candidate text block becomes a crucial factor in determining its reading role within the local area. In this step, text block length not only represents the number of characters but also the text's visual proportion, compactness, and reading weight within the drawing. Longer text blocks tend to function as the main body of a paragraph, while shorter text blocks often serve as supplementary explanations or parameter annotations. By dividing candidate text blocks by length within the local merge window, subsequent steps can further filter them based on content characteristics, laying a hierarchical foundation for building the subsequent candidate text set. After this step, the candidate text blocks in the local merge window no longer exist as a single set but as a set with a preliminary partitioned structure.

[0075] After initial segmentation of candidate text blocks according to length, further classification is performed based on the punctuation rhythm within each block. This makes the distribution pattern of punctuation marks a crucial basis for identifying the reading style and semantic completeness of text blocks. In this process, punctuation rhythm reflects the sentence structure, pause rhythm, and logical paragraph relationships within a text block. Text blocks with a steady punctuation rhythm are more likely to fulfill a complete descriptive function, while those with sparse punctuation rhythm are more likely to express structural numbers, engineering symbols, or parameter values. In-depth analysis of punctuation rhythm further reveals the inherent semantic relationships between candidate text blocks and allows for further refinement of text block classification based on length segmentation, gradually separating text blocks with different functions at the structural level. This step relies on the defined candidate range within the local merging window, enabling punctuation rhythm to play a classification role within a stable local structural environment.

[0076] After candidate text blocks are divided by length and punctuation rhythm, they are semantically regrouped based on the density of engineering vocabulary within each block. This engineering vocabulary density becomes a crucial basis for distinguishing the structural hierarchy of text blocks. In this process, engineering vocabulary density reflects the semantic depth of a text block within an engineering context. Higher-density text blocks often contain engineering parameters, equipment names, component attributes, or safety levels, while lower-density text blocks are mostly auxiliary or connecting content. By analyzing engineering vocabulary density, the differences in content complexity among candidate text blocks are expressed, and text blocks with different structural levels are gradually separated during the candidate set construction process. Particularly in candidate text blocks that are spatially close but belong to completely different structural regions, the difference in engineering vocabulary density separates them in this step, thus preventing these candidate text blocks from being incorrectly combined in subsequent paragraph generation.

[0077] After candidate text blocks are categorized by length, punctuation rhythm, and engineering vocabulary density, the classification results across these three dimensions are combined with the spatial constraints of the local merging window. This excludes candidate text blocks that are spatially close but structurally different from those in the same local merging window, thus giving the candidate text set a controllable boundary. In this step, the boundary of the candidate text set is determined not only by the spatial constraints of the local merging window but also by the differences in semantic structure and reading patterns of the text blocks. This ensures that the candidate text set only includes candidate text blocks that truly belong to the same level in terms of content, structure, and semantic trends. This step ensures that the text sources used for subsequent paragraph merging have a highly consistent structural foundation, excluding cross-level texts even when they are spatially close, and maintaining a stable boundary for the candidate text set in complex engineering diagram structures.

[0078] By using a time-slice-driven paragraph generation engine, the local merging window dynamically adjusts the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. Furthermore, a parameter-priority retention strategy is used to suppress cross-level splicing behavior, thereby reducing the risk of engineering semantic distortion caused by paragraph mismerging.

[0079] To leverage a time-slice-driven paragraph generation engine, enabling the local merging window to dynamically adjust the merging threshold based on drawing browsing behavior, zooming actions, and changes in view dwell time, and to suppress cross-level splicing behavior through a parameter priority retention strategy, thereby reducing the risk of semantic distortion in engineering caused by incorrect paragraph merging, a paragraph generation process combining time changes, user interaction behavior, and local area structural features can be further constructed based on the existing candidate text set. This ensures that the merging threshold no longer exists in a fixed form but changes in real time under the superposition of multiple external and internal conditions. The specific implementation steps are as follows:

[0080] After obtaining the formed local merging window and the candidate text set generated within it, multiple consecutive time slices are constructed for the paragraph generation engine. This allows the engine to read the candidate text set within the local merging window at a fixed rhythm within each time slice. In this step, the time slice does not perform computational functions but serves as the rhythmic unit driving the paragraph generation engine, enabling the paragraph generation process to continuously advance and allowing the candidate text set within the local merging window to be gradually read and organized within consecutive time slices. During this process, the time slice provides a time sequence for the entire paragraph generation behavior, allowing subsequent processing to dynamically observe the state changes of local areas within this time structure, thus providing a basis for subsequent threshold adjustment. Driven by time slices, the paragraph generation engine can maintain a stable progression rhythm while continuously sensing external conditions, enabling subsequent steps to adjust the paragraph generation strategy according to changing conditions.

[0081] Driven by the continuous flow of time slices, the engine monitors changes in drawing browsing behavior, zooming actions, and view dwell time. These external behaviors become key conditions driving changes in the internal structural characteristics of the local merging window, thereby triggering dynamic adjustments to the merging threshold. In this step, drawing browsing behavior reflects the user's movement trend within the drawing area, zooming actions reflect changes in the user's attention level, and view dwell time reflects the duration of continuous observation of a specific local area. By synchronizing these behavioral factors with the time slice, the paragraph generation engine can determine the user's depth of attention to a local area based on the rhythm of browsing operations. When the user continuously zooms in on a local area, the paragraph generation engine increases the priority of structural recognition for that area based on view dwell time, causing the merging threshold to decrease accordingly, thus more strictly limiting the splicing of cross-level text. When the user quickly pans the drawing or frequently zooms, the paragraph generation engine determines that the current area's structure is unstable based on these behaviors, causing the merging threshold to increase accordingly, thus avoiding momentary mis-splitting caused by drawing movement. In this way, the merging threshold maintains a dynamic linkage with user behavior within the time slice rhythm, ensuring that changes in structural hierarchy are continuously responded to during paragraph generation.

[0082] After the paragraph generation engine dynamically adjusts the merging threshold based on the time-slice rhythm and changes in user behavior, it combines the candidate text set within the local merging window with the adjusted merging threshold. This allows the engine to determine whether candidate text blocks should enter the splicing process within each time slice based on the latest threshold. In this step, the merging threshold determines not only the distance between candidate text blocks but also their matching degree in semantic structure, reading direction, and hierarchical attributes. This transforms the splicing conditions between text blocks from a fixed standard to a dynamic standard that adjusts with changes in time slices and user behavior. When the area within the local merging window is in a stable reading state, the paragraph generation engine uses a lower merging threshold for stricter splicing judgments, preventing text blocks with different structural levels from entering the splicing process. Conversely, when the local area is in a rapidly changing state, the engine uses a higher merging threshold to connect text blocks under more lenient conditions, thus maintaining the continuous progress of the paragraph generation process. Throughout this process, the local merging window acts as a boundary framework for regulating behavior, ensuring that the merging range remains within a stable area and preventing cross-window splicing behavior due to dynamic threshold changes.

[0083] When the paragraph generation engine performs splicing on the candidate text set based on dynamic thresholds, a parameter priority retention strategy is implemented. This strategy prioritizes key engineering information such as engineering parameters, load levels, size ranges, and safety zone levels during paragraph generation, preventing these text blocks from being attracted to cross-level splicing by low-relevance text blocks even during multiple rounds of threshold adjustment. This strategy ensures the integrity of engineering information in dynamically changing local environments, maintaining the coherence and stability of the engineering semantic structure during paragraph generation. Under the parameter priority retention strategy, text blocks containing key engineering attributes in the candidate text set are automatically retained in their original structural positions, effectively reducing the risk of engineering semantic distortion caused by paragraph merging errors. Ultimately, through the synergistic effect of time-slice driving, dynamic threshold adjustment, and the parameter priority retention strategy, the paragraph generation engine maintains the stability of the paragraph generation process in the variable environment of complex engineering diagrams, continuously suppressing cross-level splicing behavior under multiple constraints.

[0084] This invention constructs spatial fingerprints, semantic islands, sequential chains, and local merging windows to clearly present the hierarchical relationships between text blocks within the drawing structure. Paragraph generation is constrained within controlled boundaries, and cross-level text maintains stable separation even when spatially compact. This mechanism endows the paragraph splicing process with spatial self-organization capabilities, freeing the text structure in engineering drawings from geometric aliasing. This provides the translation engine with a clear reading path, stabilizes the semantic transmission link, and reduces the risk of semantic shifts caused by paragraph merging errors.

[0085] This invention introduces a time-slice-driven approach, enabling the local merging window to automatically adjust the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. It also employs a parameter-priority retention strategy to protect critical engineering text, ensuring robust paragraph generation in dynamic environments. This continuous adjustment allows the paragraph construction process to adapt to changes in the user's focus area, maintaining consistency in structural boundaries through multiple adjustments. It prevents critical engineering information from being absorbed into the splicing chain by irrelevant content, thereby reducing the impact of semantic distortion on engineering applications.

[0086] This invention provides, for example Figure 2 The DWG drawing intelligent translation and information verification integrated system shown includes a spatial fingerprint generation module, a semantic island construction module, a local merging window generation module, a candidate text grouping and filtering module, and a dynamic paragraph generation and control module.

[0087] The spatial fingerprint generation module generates a spatial fingerprint that reflects the layout hierarchy based on the coordinates, orientation, and font features of each text block in the DWG drawing. It also establishes the layout hierarchy trajectory of the text block based on the spatial fingerprint to form continuous trajectory data for subsequent structural attribution determination.

[0088] The semantic island construction module divides adjacent text blocks into multiple semantic islands according to their structural affiliation based on the layout hierarchy trajectory, and generates a sequence chain within each semantic island that maintains the original reading order in order to limit the splicing range of subsequent paragraph merging.

[0089] The local merge window generation module marks the beginning and end positions of paragraphs based on cross-level position changes within the splicable range of the sequential chain, and generates a local merge window based on these position marks. This restricts the paragraph merging process within the local merge window, achieving spatial isolation of cross-level text.

[0090] The candidate text grouping and filtering module groups candidate text blocks around the local merging window based on text block length, punctuation rhythm, and engineering vocabulary density. It excludes text blocks that are spatially adjacent but have different structural levels from the same local merging window, thus forming a candidate text set with controllable boundaries.

[0091] The dynamic paragraph generation and control module uses time slices to drive the paragraph generation engine, enabling the local merging window to dynamically adjust the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. It also suppresses cross-level splicing behavior through a parameter priority retention strategy, thereby reducing the risk of engineering semantic distortion caused by paragraph mismerging.

[0092] The present invention provides an integrated method for intelligent translation and information verification of DWG drawings, which is implemented through the aforementioned integrated system for intelligent translation and information verification of DWG drawings. For details of the specific method and process of the integrated system for intelligent translation and information verification of DWG drawings, please refer to the embodiment of the aforementioned integrated method for intelligent translation and information verification of DWG drawings, which will not be repeated here.

[0093] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for integrated intelligent translation and information verification of DWG drawings, characterized in that, Includes the following steps: Based on the coordinates, orientation, and font features of each text block in the DWG drawing, a spatial fingerprint reflecting the layout hierarchy is generated, and the layout hierarchy trajectory of the text block is established according to the spatial fingerprint to form continuous trajectory data. The steps for generating layout hierarchy trajectories are as follows: The text blocks are organized based on their coordinate information, orientation information, and font features, so that the text blocks form distinctive attribute descriptions on the drawing plane; Spatial fingerprints are constructed based on the organized attribute descriptions, enabling each text block to form an independent identifier that can express layout relationships; Based on the positional relationship, orientation synchronization relationship and font presentation trend between spatial fingerprints, a continuous association is made so that multiple spatial fingerprints can form a layout hierarchy trajectory that can express hierarchical relationships. The sequential connection results based on the layout hierarchy trajectory are continuously integrated to form continuous trajectory data of the spatial relationship between text blocks, so that the overall layout structure in the drawing has a trajectory expression basis for subsequent processing. Based on the layout hierarchy trajectory, adjacent text blocks are divided into multiple semantic islands according to their structural affiliation, and a sequence chain that maintains the original reading order is generated within each semantic island to limit the splicing range of subsequent paragraph merging. The steps for dividing adjacent text blocks into semantic islands and forming sequential chains based on layout hierarchy trajectories are as follows: Based on the direction and orientation changes of the layout hierarchy trajectory, the distance trend between text blocks and the continuity of font presentation, the text blocks in the layout hierarchy trajectory are initially classified so that the text blocks that appear continuously and have coherent characteristics form basic aggregation units. Based on the proximity, convergence, and reading extension trends of the basic aggregation units in the layout structure, multiple basic aggregation units are further aggregated to form semantic islands, so that the semantic islands have a stable reading direction and internal content concentration. Based on the arrangement order, reading trend, font relationship and content logic of text blocks within the semantic island, the text blocks are sorted to form a sequence chain that maintains the original reading order, so that the sequence chain has a continuous reading path; Based on the boundaries formed by the sequential chain, the scope of the sequential chain is used as the concatenation scope, so that paragraph merging is performed within the concatenation scope and cross-semantic island text blocks are excluded. Within the splicable range of the sequential chain, the beginning and end positions of the segment are marked according to the cross-level position changes, and a local merging window is generated based on the position marks, so that the segment merging process is executed in a restricted manner within the local merging window; Around the local merging window, candidate text blocks are grouped according to text block length, punctuation rhythm and engineering vocabulary density. Text blocks that are spatially adjacent but have different structural levels are excluded from the same local merging window to form a candidate text set with controllable boundaries. By utilizing a time-slice-driven paragraph generation engine, the local merging window dynamically adjusts the merging threshold based on changes in drawing browsing behavior, zooming actions, and view dwell time. Furthermore, a parameter-priority retention strategy is used to suppress cross-level splicing behavior.

2. The integrated method for intelligent translation and information verification of DWG drawings according to claim 1, characterized in that, Based on the relative positions and font sizes of text blocks within the semantic island, the text blocks are sorted according to the reading trend from top to bottom or from left to right. During the sorting process, the spatial position and semantic structure of the text blocks are kept consistent, ensuring that the reading path of the text blocks within the sequential chain has natural fluidity.

3. The integrated method for intelligent translation and information verification of DWG drawings according to claim 1, characterized in that, The steps for marking the start and end positions of segments based on the concatenable range of the sequential chain and generating a local merge window are as follows: Based on the positional relationship of text blocks in the sequential chain, the reading extension direction, the hierarchical change trend of the layout hierarchy trajectory, and the displacement gradient with adjacent text blocks, the regions where cross-level jumps occur between text blocks are identified, and the front and back ends of the abrupt change regions are marked as the beginning and end positions of the paragraph, respectively. Based on the position of the beginning and end of the paragraph, a splicable range is formed, and the text blocks within the splicable range are ensured to have a stable structure and order; Based on the spatial regions of the beginning and end of the segment, a local merging window is constructed, and the local merging window is made spatially consistent with the sequence chain to ensure that the merging process is limited to the window. By using the local merge window as the sole merge execution area, the paragraph merging process is ensured to take place within this window. Furthermore, the window boundaries enable spatial isolation of text across different levels, thus providing each local merge window with a clear structural definition.

4. The integrated method for intelligent translation and information verification of DWG drawings according to claim 3, characterized in that, When constructing a local merge window, the boundaries of the local merge window are adjusted synchronously with the reading extension direction of the sequential chain, so that the window maintains continuity in the extension direction and forms a restricted interval in the vertical direction based on the distance between the beginning and end of the paragraph, so that the text blocks inside the window continue to maintain an independent structure in space and that cross-level text remains isolated outside the window.

5. The integrated method for intelligent translation and information verification of DWG drawings according to claim 3, characterized in that, The steps for forming a candidate text set with controllable boundaries around a local merging window are as follows: Based on the generated local merged window, all candidate text blocks inside the window are obtained, and the candidate text blocks are initially divided according to the length of the text blocks, so that the candidate text blocks form a set with a partition structure. Based on the preliminary segmentation results, each candidate text block is classified according to the punctuation rhythm within it, so that the differences between the candidate text blocks in terms of reading style and semantic completeness can be further expressed. Based on the punctuation rhythm classification results, the candidate text blocks are regrouped according to the engineering vocabulary density within them, so that candidate text blocks with different structural levels are separated at the semantic level. Based on the results of length division, punctuation rhythm classification and engineering vocabulary density grouping, and combined with the spatial constraints of the local merging window, candidate text blocks that are spatially adjacent but have different structural levels are excluded from the window range, so that the candidate text set forms a set with controllable boundaries.

6. The integrated method for intelligent translation and information verification of DWG drawings according to claim 5, characterized in that, When forming a candidate text set with controllable boundaries, candidate text blocks with continuous text block length and punctuation rhythm and high engineering vocabulary density are preferentially retained within the local merging window. This ensures that the candidate text set remains consistent in structural hierarchy and further excludes low-density text blocks that are spatially adjacent from the set, thereby forming a stable source of text blocks in terms of semantic expression.

7. The integrated method for intelligent translation and information verification of DWG drawings according to claim 5, characterized in that, Dynamically adjusting the merging threshold and suppressing cross-level splicing behavior using a time-slice-driven paragraph generation engine includes the following steps: Based on the formed local merging window and the candidate text set inside the local merging window, a continuous time slice is constructed for the paragraph generation engine, so that the paragraph generation engine reads the candidate text set at a fixed rhythm in each time slice. Based on the rhythm changes of time slices, the system monitors drawing browsing behavior, zooming actions, and view dwell time, so that these behaviors trigger dynamic adjustment of the merging threshold under the action of time slices. Based on the adjusted merging threshold, the candidate text set inside the local merging window is combined with the new threshold within each time slice, so that the paragraph generation engine decides whether the candidate text block enters the splicing process based on the latest threshold. The paragraph generation engine implements a parameter-priority retention strategy for splicing behavior, which ensures that text blocks containing key engineering attributes are retained in multiple rounds of threshold adjustment, thereby suppressing cross-level splicing behavior and reducing the risk of engineering semantic distortion.

8. A DWG drawing intelligent translation and information verification integrated system, used to implement the DWG drawing intelligent translation and information verification integrated method according to any one of claims 1-7, characterized in that, It includes a spatial fingerprint generation module, a semantic island construction module, a local merging window generation module, a candidate text grouping and filtering module, and a dynamic paragraph generation and control module: The spatial fingerprint generation module generates a spatial fingerprint that reflects the layout hierarchy based on the coordinates, orientation, and font features of each text block in the DWG drawing, and establishes the layout hierarchy trajectory of the text block based on the spatial fingerprint to form continuous trajectory data. The semantic island construction module divides adjacent text blocks into multiple semantic islands according to their structural affiliation based on the layout hierarchy trajectory, and generates a sequence chain within each semantic island that maintains the original reading order in order to limit the splicing range of subsequent paragraph merging. The local merge window generation module marks the beginning and end positions of segments based on cross-level position changes within the splicable range of the sequential chain, and generates a local merge window based on these position marks, so that the segment merging process is executed within the local merge window. The candidate text grouping and filtering module groups candidate text blocks around the local merging window based on text block length, punctuation rhythm, and engineering vocabulary density. It excludes text blocks that are spatially adjacent but have different structural levels from the same local merging window, thus forming a candidate text set with controllable boundaries. The dynamic paragraph generation control module uses time slices to drive the paragraph generation engine, enabling the local merging window to dynamically adjust the merging threshold based on drawing browsing behavior, zooming actions, and changes in view dwell time. It also suppresses cross-level splicing behavior through a parameter priority retention strategy.

Images