A treatment system for processing multi-modal power engineering drawing data

Abstract

The present application relates to the field of electric power engineering data processing, and more particularly to a treatment system for processing multi-modal electric power engineering drawing data, the present application cooperatively works by setting the acquisition module, the induction analysis module, the induction labeling module, the preliminary labeling module, the sliding verification module and the deep labeling module, acquires drawing data based on the acquisition module, extracts contour features and generates dense reference frames by clustering from the induction analysis module; the induction labeling module matches the above reference frame with the sample legend, and labels the induction reference frame; the preliminary labeling module identifies the components in the non-induction area based on the model; the sliding verification module slides in the induction reference frame according to the sliding window, and identifies the framed image; finally, the deep labeling module comprehensively identifies the result, determines the separate labeling or the overall labeling of the drawing, thereby improving the accuracy and reliability of component identification and labeling.

Description

Technical Field

[0001] This invention relates to the field of power engineering data processing, and in particular to a processing system for processing multimodal power engineering drawing data. Background Technology

[0002] With the deepening of the digital and intelligent transformation of power systems, power engineering drawings, as the core carrier of design intent and engineering information, are not only closely related to power grid planning, equipment configuration, and engineering construction, but also closely linked to operation and maintenance management, fault diagnosis, and the integration of data throughout the entire lifecycle. Efficient and accurate drawing data processing capabilities have become an indispensable part of improving power system reliability and achieving integrated design and operation.

[0003] Chinese Patent Publication No. CN114139876A discloses a method for automatically managing power engineering drawings and data based on big data. The method includes five steps: step one: data entry; step two: automatic classification; step three: QR code association; step four: borrowing management; and step five: advanced search support. This application changes the traditional management model of power engineering drawings and data, achieving automated management, reducing the input of manpower and resources, and greatly improving work efficiency.

[0004] However, the following problems still exist in the existing technology.

[0005] In practice, due to the complex structure, numerous types of components, and significant differences in layout density in power engineering drawings, traditional identification methods often struggle to achieve accurate segmentation and identification when faced with areas of high density and small-sized component clusters. Frequent occurrences of overlapping component outlines, overlapping legends, and symbol occlusion can lead to multiple independent components in a local area being misidentified as a single composite graphic, thereby reducing the accuracy and reliability of component identification and labeling. Summary of the Invention

[0006] To address this issue, the present invention provides a real-time fire prevention and extinguishing system for belt conveyor systems, overcoming the difficulty in accurately segmenting and identifying high-density, small-sized component clusters in existing technologies. Frequent occurrences of overlapping component outlines, overlapping symbols, and occlusion can lead to multiple independent components in a local area being misidentified as a single composite graphic, thus reducing the accuracy and reliability of component identification and labeling.

[0007] To achieve the above objectives, the present invention provides a processing system for processing multimodal power engineering drawing data, comprising,

[0008] The data acquisition module is used to acquire power engineering drawings uploaded by the user.

[0009] The induced analysis module, which is connected to the acquisition module, is used to extract contour features from the power engineering drawing data and perform induced feature analysis on the power engineering drawing data based on the contour features. This includes dividing the power engineering drawing data into several regions, clustering based on the contour features within the regions, determining the region clusters, and determining dense reference boxes based on the distribution density of the contour features in the region clusters.

[0010] The induced annotation module, which is connected to the induced parsing module, is used to match each of the dense reference boxes with the sample legend and to annotate several induced reference boxes based on the matching results.

[0011] The preliminary annotation module, which is connected to the induced parsing module and the induced annotation module respectively, is used to identify the components of the non-induced reference frame in the power engineering drawing based on the preset model and to annotate them;

[0012] The sliding verification module, which is connected to the preliminary annotation module, is used to determine the sliding window size based on the size of the identified components, and to construct a sliding window of the corresponding size in the induced reference frame for sliding verification. This includes moving the sliding window, selecting several regions, and individually recognizing the images in the selected regions using the model.

[0013] The depth annotation module, which is connected to the sliding verification module, is used to determine whether to annotate the induced reference box individually or as a whole based on the sliding verification result.

[0014] Furthermore, the induced parsing module is used to perform clustering based on contour features within the region to determine the region clusters, wherein,

[0015] The clustering condition is that each region in the said region cluster is adjacent and all have contour features.

[0016] Furthermore, the induced parsing module is used to determine dense reference boxes based on the distribution density of contour features in region clusters, wherein,

[0017] If the contour density of the contour features in a region cluster is greater than or equal to a predetermined contour density threshold, then the edge of the region cluster is determined as a dense reference box.

[0018] Furthermore, the induced annotation module is used to match each of the dense reference boxes with the sample legend, including:

[0019] Used to extract image features of each of the dense reference boxes and sample image features of the sample legend;

[0020] Used to determine the similarity between the image features of the dense reference box and the corresponding sample image features.

[0021] Furthermore, the induced annotation module is used to annotate several induced reference boxes based on the matching results, including:

[0022] If the similarity is greater than or equal to the similarity threshold, then the dense reference box is labeled as an induced reference box;

[0023] If the similarity is less than the similarity threshold, then the dense reference boxes are labeled as non-induced reference boxes.

[0024] Furthermore, the sliding verification module, used to determine the sliding window size based on the dimensions of the identified components, includes:

[0025] Used to determine the average width of the minimum circumscribed rectangle for each component;

[0026] A square sliding window is constructed based on a predetermined multiple of the average width of the minimum bounding rectangle as the side length of the sliding window.

[0027] Furthermore, the sliding verification module is used to construct a sliding window of corresponding size in the induced reference frame for sliding verification, including:

[0028] Used to determine the width of the induced reference frame;

[0029] Used to set the step size of the sliding window to a predetermined proportion of the width;

[0030] Used to move the sliding window based on the step size, and select several regions of the image.

[0031] Furthermore, the sliding verification module is used to individually identify the image in the selected area using the model, including:

[0032] Used to determine the image within the area selected by the sliding window;

[0033] This is used to input the image into a preset recording model for individual recognition.

[0034] Furthermore, the depth annotation module is used to determine, based on the sliding verification result, whether to annotate the induced reference box individually or as a whole, including...

[0035] If the recording model identifies an independent and complete component, the induced reference frame is labeled separately;

[0036] If the recording model fails to identify an independent and complete component, the induced reference frame is labeled as a whole.

[0037] Furthermore, the individual annotation includes only annotating each of the independent complete components, while the overall annotation includes individually identifying and annotating the image within the induced reference box using a preset model.

[0038] Compared with existing technologies, this invention improves the accuracy and reliability of component identification and annotation by setting up a collaborative work of an acquisition module, an induced analysis module, an induced annotation module, a preliminary annotation module, a sliding verification module, and a depth annotation module. The acquisition module acquires drawing data; the induced analysis module extracts contour features and clusters them to generate dense reference boxes; the induced annotation module matches these reference boxes with sample illustrations to annotate the induced reference boxes; the preliminary annotation module identifies components in non-induced areas based on a model; the sliding verification module slides a sliding window within the induced reference boxes and identifies the selected images; finally, the depth annotation module synthesizes the identification results to determine whether to annotate the drawing individually or as a whole, thereby improving the accuracy and reliability of component identification and annotation.

[0039] In particular, this invention determines regional clusters by clustering based on contour features within a region, and then determines dense reference frames based on the distribution density of contour features within these clusters. In practice, due to the complexity of drawing structures, large differences in component layout density, and factors such as inconsistent historical drawing standards, varying image quality, scanning distortion, or manual drawing errors, component contours often overlap, stick together, or obscure. If a global fixed scale is directly used for identification, misjudgment is very likely to occur in high-density areas, misidentifying multiple independent components as a single composite graphic. For example, in substation layout diagrams or power distribution network diagrams, small components such as circuit breakers and relays are often densely arranged, with their legend contours intersecting or even partially missing. Traditional methods often incorrectly identify multiple independent components as a single composite component, or fail to identify valid components due to the loss of local features, which may lead to the failure of subsequent annotation and data extraction. Therefore, by clustering regions using contour features, it is possible to better distinguish regions with different density distributions and identify high-density regions as dense reference boxes. This provides a reliable basis for subsequently labeling several induced reference boxes and then identifying components through sliding verification, thereby improving the accuracy and reliability of component identification and labeling.

[0040] In particular, this invention performs sliding verification by constructing a sliding window of corresponding size within the induced reference frame. This includes moving the sliding window to select several regions, and then individually identifying the images within the selected regions using the model. In practice, the regions corresponding to the induced reference frame often have densely distributed components, complex and intersecting outlines, and significant occlusion or adhesion. Relying solely on overall region identification can easily lead to multiple internal components being misidentified as a single entity or missing key components. For example, in circuit breaker clusters or relay groups, the graphic symbols of different components may partially overlap, intersect, or even almost merge due to limitations in graphic size accuracy. If the induced reference frame is used as a unit for overall identification, the model may not be able to fully perceive and separate closely adjacent independent components, potentially misidentifying multiple components as a single composite graphic. Therefore, a sliding window is constructed for the induced reference frame, and individual identification is performed. The introduction of sliding windows can better decompose complex areas within the induced reference box into multiple smaller sub-regions. By individually identifying the images of each selected region, subtle features can be captured, providing a more reliable basis for the subsequent depth annotation module's judgment. This better avoids misjudgment and omissions caused by overlapping, sticking, or occlusion of component outlines, thereby improving the accuracy and reliability of component identification and annotation.

[0041] In particular, this invention determines whether to individually or collectively label the induced reference frame based on the sliding verification results. Because component layout densities vary in power engineering drawings, legend drawing standards differ, and there are numerous instances of graphic overlap and occlusion, uniformly labeling the entire frame might incorrectly merge multiple spatially adjacent independent components, potentially leading to the loss of key component identification information. Conversely, uniformly labeling each component individually might over-segment what should be a unified composite functional unit, generating numerous graphic fragments without engineering significance and disrupting its complete electrical semantics. Therefore, the image selected by the sliding verification is identified, and labeling is performed based on the identification results. If the recording model identifies independent and complete components during the sliding verification process, it indicates that the induced reference frame actually contains multiple independent components densely adjacent in space, rather than a single graphic entity. In this case, each identified independent component is individually labeled. Conversely, if no independent and complete components are identified, it indicates that the graphic elements in that area are highly integrated, with outlines and internal features coupled together; forced segmentation would lead to semantic errors or the generation of meaningless graphic fragments. At this point, all graphics within the induced reference frame are treated as a complete whole with specific composite functions, and thus annotated as a whole. This adaptive annotation mechanism based on recognition results can better cope with the complexity of component identification in power engineering drawings, thereby improving the accuracy and reliability of component identification and annotation. Attached Figure Description

[0042] Figure 1This is a schematic diagram of the processing system for processing multimodal power engineering drawing data according to an embodiment of the invention;

[0043] Figure 2 This is a logic decision diagram for determining dense reference frames in an embodiment of the invention.

[0044] Figure 3 A logic decision diagram with several induced reference boxes marked for embodiments of the invention;

[0045] Figure 4 This is a logic diagram for determining whether to individually or collectively annotate the induced reference frame in an embodiment of the invention. Detailed Implementation

[0046] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0047] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0048] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0049] Please see Figure 1 The diagram shown is a structural schematic of a processing system for processing multimodal power engineering drawing data according to an embodiment of the present invention. The processing system for processing multimodal power engineering drawing data of the present invention includes:

[0050] The data acquisition module is used to acquire power engineering drawings uploaded by the user.

[0051] The induced analysis module, which is connected to the acquisition module, is used to extract contour features from the power engineering drawing data and perform induced feature analysis on the power engineering drawing data based on the contour features. This includes dividing the power engineering drawing data into several regions, clustering based on the contour features within the regions, determining the region clusters, and determining dense reference boxes based on the distribution density of the contour features in the region clusters.

[0052] The induced annotation module, which is connected to the induced parsing module, is used to match each of the dense reference boxes with the sample legend and to annotate several induced reference boxes based on the matching results.

[0053] The preliminary annotation module, which is connected to the induced parsing module and the induced annotation module respectively, is used to identify the components of the non-induced reference frame in the power engineering drawing based on the preset model and to annotate them;

[0054] The sliding verification module, which is connected to the induced annotation module, is used to determine the sliding window size based on the size of the identified components, and to construct a sliding window of the corresponding size in the induced reference frame for sliding verification. This includes moving the sliding window, selecting several regions, and individually identifying the images in the selected regions using the model.

[0055] The depth annotation module, which is connected to the sliding verification module, is used to determine whether to annotate the induced reference box individually or as a whole based on the sliding verification result.

[0056] Specifically, there are no restrictions on the structure of the induced parsing module, induced annotation module, preliminary annotation module, sliding verification module, and depth annotation module. They can be composed of logical components or combinations of logical components, including field-programmable processors, computers, or microprocessors in computers.

[0057] Specifically, there are no restrictions on the structure of the acquisition module. It can obtain power engineering drawings uploaded by the user client by calling the user client upload interface, or obtain multimodal drawing data of different formats and sources by accessing specified network storage paths, database interfaces and file transfer services, etc. It is only necessary to meet the integrity and accuracy of drawing data transmission, which will not be elaborated further.

[0058] Specifically, there are no restrictions on the method for extracting contour features from power engineering drawing data. Contour extraction can be performed by using a convolutional neural network model or by using traditional image segmentation algorithms. As long as the geometric contours and structural boundaries of the components in the drawing can be accurately identified and separated, and the accuracy and completeness of the contour extraction are met, this will not be elaborated further.

[0059] Specifically, there is no limitation on the way the power engineering drawing data is divided into several regions. It can be divided by a fixed grid-based method, such as dividing the drawing into rectangles or adaptive grids, or by other methods. It is only necessary to divide it into several regions of uniform size and spatial continuity, ensuring that the divided regions can cover the entire drawing. This will not be elaborated further.

[0060] Specifically, after the model identifies the component type by labeling the components, a corresponding virtual type label is set for the component.

[0061] Furthermore, based on virtual type tags, it is possible to quickly count the number of components, assist drawing readers in determining the corresponding component types, and archive relevant component data, which will not be elaborated further.

[0062] Specifically, the induced parsing module is used to perform clustering based on contour features within the region to determine the region's clusters, wherein...

[0063] The clustering condition is that each region in the said region cluster is adjacent and all have contour features.

[0064] In practice, cluster analysis is performed on spatially adjacent regions that all contain valid contours, and regions that are adjacent to each other and all have contour features are identified as region clusters.

[0065] Please see Figure 2 As shown, this is a logical decision diagram for determining dense reference boxes according to an embodiment of the invention. Specifically, the induced parsing module is used to determine dense reference boxes based on the distribution density of contour features in region clusters.

[0066] If the contour density of the contour features in a region cluster is greater than or equal to a predetermined contour density threshold, then the edge of the region cluster is determined as a dense reference box.

[0067] In implementation, the purpose of the predetermined contour density threshold is to characterize the density of contour features in regional clusters to distinguish regions with different densities. The contour density threshold is predetermined. Those skilled in the art can collect a large amount of power engineering drawing data to determine the ratio of the pixel area occupied by contour features within each regional cluster to the total area of ​​the region, i.e., the pixel percentage. The average pixel percentage is then determined to represent the general density of component distribution under normal conditions. To indicate a relatively dense contour feature distribution in regional clusters, the contour density threshold is set as the product of the average value and a density accuracy coefficient. Typically, the density accuracy coefficient is selected within the range of [1.35, 1.55], and in implementation, 1.45 is preferred.

[0068] This invention determines regional clusters by clustering based on contour features within a region, and then determines dense reference frames based on the distribution density of contour features within these clusters. In practice, due to the complexity of drawing structures, large differences in component layout density, and factors such as inconsistent historical drawing standards, varying image quality, scanning distortion, or manual drawing errors, component contours often overlap, stick together, or obscure. If a global fixed scale is directly used for identification, misjudgment is very likely to occur in high-density areas, misidentifying multiple independent components as a single composite graphic. For example, in substation layout diagrams or power distribution network diagrams, small components such as circuit breakers and relays are often densely arranged, with their legend contours intersecting or even partially missing. Traditional methods often incorrectly identify multiple independent components as a single composite component, or fail to identify the effective component due to the loss of local features, which may lead to the failure of subsequent annotation and data extraction. Therefore, by clustering regions using contour features, it is possible to better distinguish regions with different density distributions and identify high-density regions as dense reference boxes. This provides a reliable basis for subsequently labeling several induced reference boxes and then identifying components through sliding verification, thereby improving the accuracy and reliability of component identification and labeling.

[0069] Specifically, the induced annotation module is used to match each of the dense reference boxes with the sample legend, including:

[0070] Used to extract image features of each of the dense reference boxes and sample image features of the sample legend;

[0071] Used to determine the similarity between the image features of the dense reference box and the corresponding sample image features.

[0072] In practice, the similarity between the image features of the dense reference box and the corresponding sample image features can be calculated using the cosine similarity method. The image is vectorized and the cosine similarity is calculated as the similarity. Of course, other methods can also be used, as long as they can represent the similarity between images. This will not be elaborated further.

[0073] Please see Figure 3 As shown, it is a logic decision diagram for annotating several induced reference boxes according to an embodiment of the invention. Specifically, the induced annotation module is used to annotate several induced reference boxes based on the matching results, including...

[0074] If the similarity is greater than or equal to the similarity threshold, then the dense reference box is labeled as an induced reference box;

[0075] If the similarity is less than the similarity threshold, then the dense reference boxes are labeled as non-induced reference boxes.

[0076] Specifically, the similarity threshold is predetermined. The purpose of setting the similarity threshold is to characterize the situation where image similarity is prone to inducement. Those skilled in the art can mark several similar components in the power engineering drawings in advance, compare the average similarity between the components, and determine the similarity threshold by the product of the average similarity and the error coefficient. The error coefficient is selected in the range [0.85, 0.95].

[0077] Specifically, the sliding verification module is used to determine the sliding window size based on the size of the identified components, including:

[0078] Used to determine the average width of the minimum circumscribed rectangle for each component;

[0079] A square sliding window is constructed based on a predetermined multiple of the average width of the minimum bounding rectangle as the side length of the sliding window.

[0080] In implementation, there are no restrictions on the method for determining the minimum bounding rectangle width of each component. Image processing techniques, such as edge detection algorithms, can be used to extract the component outlines, and then geometric algorithms or computer vision libraries can be applied to calculate the minimum bounding rectangle of the outlines to obtain the width. Alternatively, semantic segmentation techniques from deep learning can be used to first separate the components from the background before calculating the bounding rectangle. As long as the outlines of the components can be accurately obtained and the minimum bounding rectangle width can be calculated, it is acceptable, which will not be elaborated further.

[0081] In practice, preferably, the predetermined multiple is 1.2 times the average width of the minimum circumscribed rectangle.

[0082] Specifically, the sliding verification module is used to construct a sliding window of a corresponding size within the induced reference frame for sliding verification, including:

[0083] Used to determine the width of the induced reference frame;

[0084] Used to set the step size of the sliding window to a predetermined proportion of the width;

[0085] Used to move the sliding window based on the step size, and select several regions of the image.

[0086] In practice, preferably, the predetermined ratio is one-half the width of the sliding window.

[0087] In practice, the sliding window starts from the top left corner of the induced reference frame and moves to the right in steps, selecting a new region for verification after each move. When the sliding window reaches the right boundary of the induced reference frame, it moves down one step and starts moving right again from the left, continuing to select and verify the region. This process continues until the sliding window covers the entire induced reference frame area.

[0088] Specifically, the sliding verification module is used to individually identify the image in the selected area using the model, including:

[0089] Used to determine the image within the area selected by the sliding window;

[0090] This is used to input the image into a preset recording model for individual recognition.

[0091] Specifically, there are no restrictions on the form of the preset model. Existing open-source image processing models that can identify electrical components in images can be used, or the model can be trained independently. There are no restrictions on the architecture of the model. For example, a neural network architecture can be used to obtain several image samples labeled with component types to train the model, so that the model can identify the component types in the image. This will not be elaborated further.

[0092] This invention performs sliding verification by constructing a sliding window of corresponding size within an induced reference frame. This includes moving the sliding window to select several regions, and then individually identifying the images within the selected regions using the model. In practice, the regions corresponding to the induced reference frame often have densely distributed components, complex and intersecting outlines, and significant occlusion or adhesion. Relying solely on overall region identification can easily lead to multiple internal components being misidentified as a single entity or missing key components. For example, in circuit breaker clusters or relay groups, the graphic symbols of different components may partially overlap, intersect, or even almost merge due to limitations in image size accuracy. If the induced reference frame is used as a unit for overall identification, the model may not be able to fully perceive and separate closely adjacent independent components, potentially misidentifying multiple components as a single composite graphic. Therefore, a sliding window is constructed for the induced reference frame, and individual identification is performed. The introduction of sliding windows can better decompose complex areas within the induced reference box into multiple smaller sub-regions. By individually identifying the images of each selected region, subtle features can be captured, providing a more reliable basis for the subsequent depth annotation module's judgment. This better avoids misjudgment and omissions caused by overlapping, sticking, or occlusion of component outlines, thereby improving the accuracy and reliability of component identification and annotation.

[0093] Please see Figure 4 As shown, this is a logic diagram for determining whether to individually or entirely annotate the induced reference box according to an embodiment of the invention. Specifically, the depth annotation module is used to determine whether to individually or entirely annotate the induced reference box based on the sliding verification result, including...

[0094] If the recording model identifies an independent and complete component, the induced reference frame is labeled separately;

[0095] If the recording model fails to identify an independent and complete component, the induced reference frame is labeled as a whole.

[0096] This invention uses a sliding verification method to determine whether to individually or collectively annotate the induced reference frame. Due to variations in component density, inconsistent legend drawing standards, and numerous instances of graphic overlap and occlusion in power engineering drawings, uniform collective annotation might incorrectly merge multiple spatially adjacent independent components, potentially leading to the loss of key component identification information. Conversely, uniform individual annotation might over-segment what should be a unified composite functional unit, generating numerous graphic fragments without engineering significance and disrupting its complete electrical semantics. Therefore, the invention identifies components based on the image selected by the sliding verification and annotates them based on the identification results. If the recording model identifies independent and complete components during the sliding verification process, it indicates that the induced reference frame actually contains multiple independent components densely adjacent in space, rather than a single graphic entity. In this case, each identified independent component is individually annotated. Conversely, if no independent and complete components are identified, it indicates that the graphic elements in that area are highly integrated, with contours and internal features coupled together; forced segmentation would lead to semantic errors or the generation of meaningless graphic fragments. At this point, all graphics within the induced reference frame are treated as a complete whole with specific composite functions, and thus annotated as a whole. This adaptive annotation mechanism based on recognition results can better cope with the complexity of component identification in power engineering drawings, thereby improving the accuracy and reliability of component identification and annotation.

[0097] Specifically, the individual annotation includes annotating only each of the independent complete components, while the overall annotation includes individually identifying and annotating the image within the induced reference box using a preset model.

[0098] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A processing system for handling multimodal power engineering drawing data, characterized in that, include, The data acquisition module is used to acquire power engineering drawings uploaded by the user. The induced feature parsing module, connected to the acquisition module, is used to extract contour features from the power engineering drawing data. Based on these contour features, it performs induced feature parsing on the power engineering drawing data, including dividing the data into several regions, clustering based on contour features within each region, and determining the region clusters. Cluster analysis is performed on spatially adjacent regions that all contain valid contours, and the regions that are adjacent to each other and all have contour features are identified as region clusters. Dense reference boxes are determined based on the distribution density of contour features in regional clusters; The induced annotation module, which is connected to the induced parsing module, is used to match each of the dense reference boxes with the sample legend and to annotate several induced reference boxes based on the matching results. The preliminary annotation module, which is connected to the induced parsing module and the induced annotation module respectively, is used to identify the components of the non-induced reference frame in the power engineering drawing based on the preset model and to annotate them; The sliding verification module, which is connected to the induced annotation module, is used to determine the sliding window size based on the size of the identified components, and to construct a sliding window of the corresponding size in the induced reference frame for sliding verification. This includes moving the sliding window, selecting several areas, and individually recognizing the images in the selected areas using a preset recording model. A depth annotation module, connected to the sliding verification module, is used to determine whether to annotate the guided reference box individually or as a whole based on the sliding verification result, including: If the recording model identifies an independent and complete component, the induced reference frame is labeled separately; If the recording model fails to identify an independent and complete component, the entire induced reference frame is labeled. The individual annotation includes annotating only each of the independent complete components, while the overall annotation includes individually identifying and annotating the image within the induced reference box using a preset model.

2. The processing system for processing multimodal power engineering drawing data according to claim 1, characterized in that, The induced parsing module is used to determine dense reference boxes based on the distribution density of contour features in region clusters, wherein, If the contour density of the contour features in a region cluster is greater than or equal to a predetermined contour density threshold, then the edge of the region cluster is determined as a dense reference box.

3. The processing system for processing multimodal power engineering drawing data according to claim 1, characterized in that, The induced annotation module is used to match each of the dense reference boxes with the sample legend, including... Used to extract image features of each of the dense reference boxes and sample image features of the sample legend; Used to determine the similarity between the image features of the dense reference box and the corresponding sample image features.

4. The processing system for processing multimodal power engineering drawing data according to claim 3, characterized in that, The induced annotation module is used to annotate several induced reference boxes based on the matching results, including: If the similarity is greater than or equal to the similarity threshold, then the dense reference box is labeled as an induced reference box; If the similarity is less than the similarity threshold, then the dense reference boxes are labeled as non-induced reference boxes.

5. The processing system for processing multimodal power engineering drawing data according to claim 1, characterized in that, The sliding verification module is used to determine the sliding window size based on the dimensions of the identified components, including: Used to determine the average width of the minimum circumscribed rectangle for each component; A square sliding window is constructed based on a predetermined multiple of the average width of the minimum bounding rectangle as the side length of the sliding window.

6. The processing system for processing multimodal power engineering drawing data according to claim 1, characterized in that, The sliding verification module is used to construct a sliding window of corresponding size in the induced reference frame for sliding verification, including: Used to determine the width of the induced reference frame; Used to set the step size of the sliding window to a predetermined proportion of the width; Used to move the sliding window based on the step size, and select several regions of the image.

7. The processing system for processing multimodal power engineering drawing data according to claim 1, characterized in that, The sliding verification module is used to individually identify the image in the selected area using a preset recording model, including... Used to determine the image within the area selected by the sliding window; This is used to input the image into a preset recording model for individual recognition.

Images