Railway engineering drawing digitization and engineering quantity extraction method based on deep learning

Abstract

This invention belongs to the field of engineering geological mapping technology. To address the current problems of low efficiency, large errors, and insufficient automation in processing engineering information, it provides a deep learning-based method for digitizing railway engineering drawings and extracting quantities. This method optimizes input by establishing a railway professional icon library and employing various data augmentation techniques such as adaptive histogram equalization. It utilizes an improved YOLOv8 model (incorporating SCConv, EMA, and RFA mechanisms) to achieve high-precision detection of graphic targets, and combines this with a PaddleOCR model to extract text information. Finally, the recognition results are digitized and automatically integrated with the engineering bill of quantities for quantity calculation, outputting standardized results conforming to an enterprise-level structured coding system. This method significantly improves the accuracy and efficiency of drawing recognition, automates and standardizes quantity calculation, and provides reliable technical support for the digital management of railway engineering projects.

Description

Technical Field

[0001] This invention belongs to the field of engineering geological mapping technology, specifically involving a method for digitizing railway engineering drawings and extracting engineering quantities based on deep learning. Background Technology

[0002] With the rapid expansion of high-speed rail networks, the demand for railway construction projects is constantly increasing. In traditional railway engineering construction, information transfer from the design phase to the construction phase mainly relies on paper documents, which is not only inefficient but also prone to errors. Especially in railway communication, signaling, and power systems, where numerous charts and technical parameters are involved, manually processing this information becomes exceptionally complex and time-consuming. Furthermore, there is significant uncertainty in quantity calculations, affecting project cost control and schedule management. Summary of the Invention

[0003] In order to solve at least one of the above-mentioned technical problems in the prior art, the present invention provides a method for digitizing railway engineering drawings and extracting engineering quantities based on deep learning.

[0004] This invention is achieved using the following technical solution: a method for digitizing railway engineering drawings and extracting engineering quantities based on deep learning, comprising the following steps: S1: Establish a railway professional identification database, classify and label the icons and atlases in the railway professional identification database, and perform data augmentation to obtain standard engineering design drawings; S2: Use the improved YOLOv8 model to extract image features and detect targets in the standard engineering design drawing. The improved YOLOv8 model uses the spatial and channel reconstruction convolution SCConv module to replace the traditional convolution in the C2f module. It introduces the EMA efficient multi-scale attention mechanism in the backbone network and the neck network, and introduces the RFA receptive field attention mechanism to output the category, probability and coordinate information of the features contained in the standard engineering design drawing. S3: Use the PaddleOCR text detection model to extract the text information from the standard engineering design drawings; S4: Digitize the identification results in steps S2 and S3 into a structured data format, compare them with the pre-stored on-site engineering list, automatically calculate the engineering quantity based on the comparison results, and summarize the engineering quantity calculation results to form an enterprise-level structured coding system.

[0005] Preferably, the data augmentation includes at least one or a combination of the following methods: deep learning-based adaptive histogram equalization, random brightness adjustment, motion blur addition, and image horizontal or vertical flipping.

[0006] Preferably, the SCConv module further includes an SRU module for suppressing spatial redundancy and a CRU module for reducing channel redundancy.

[0007] Preferably, in step S2, the RFA receptive field attention mechanism is introduced in combination with convolutional blocks to improve the improved YOLOv8 model's ability to perceive large-scale convolutional kernels and the overall network performance.

[0008] Preferably, step S3 further includes: Filter drawings of the same type by drawing name through the design platform API interface, and convert the selected drawings into IMG images; An image table preprocessing operation based on OpenCV is used to extract grid lines from the IMG image to form a table frame without filled information. The PaddleOCR text recognition model is used to extract text information from the IMG image, and the relative positional relationships of the extracted text information are recorded. The extracted text information is filled into the table frame to achieve structured acquisition of chart data.

[0009] Preferably, in step S4, the digitized structured data is in JSON format, which includes the element's ID number, name, type, location coordinates, region, and other attribute information.

[0010] Preferably, in step S4, comparing with the pre-stored on-site engineering list includes: matching the name, specifications, and usage information of the digital elements with the items in the on-site list to determine the correspondence; the automatic calculation of the engineering quantity includes: calculating the engineering quantity of each element based on the unit price and quantity information in the on-site list, combined with the actual distribution in the drawings, and adjusting it according to engineering standards and experience, taking into account a reasonable loss coefficient.

[0011] Preferably, in step S4, the process of summarizing to form an enterprise-level structured coding system includes: allocating the calculated engineering quantities according to the determined enterprise-level hierarchical structure, from the project level down to the system level, subsystem level and equipment level, and allocating cross-level engineering quantities according to their actual functions and connection relationships.

[0012] Compared with the prior art, the beneficial effects of the present invention are: The deep learning-based method for digitizing railway engineering drawings and extracting quantities provided by this invention has the following significant advantages compared to traditional manual identification and calculation methods: 1. High recognition efficiency and automation: By working together with the improved YOLOv8 target detection model and PaddleOCR text recognition model, end-to-end automatic recognition and extraction of graphic symbols and text information in railway drawings is achieved, which greatly reduces the time for manual interpretation, input and verification, and significantly improves the overall efficiency of drawing digitization. 2. High recognition accuracy and robustness: The YOLOv8 model, which integrates SCConv, EMA and RFA attention mechanisms, effectively enhances the model's ability to detect small, dense and multi-scale targets in complex drawings; combined with various data augmentation methods, it improves the model's generalization performance and stability under different lighting, noise and blur conditions. 3. Structured and standardized output: The recognition results are output in structured data formats such as JSON, with complete element information, which facilitates computer parsing and subsequent processing, and provides a foundation for realizing the interconnection and system integration of engineering data; 4. Accurate and reliable quantity calculation: By automatically comparing and matching the identification results with the on-site project list, and combining preset rules and loss coefficient adjustments, the quantity of work can be accurately calculated, reducing human error and improving the accuracy of budget and material management. 5. Supports Enterprise-level Structured Code (EBS) integration: The final engineering quantity data can be automatically collected and allocated according to the EBS hierarchical system. The output results can be directly used in engineering management, cost control, schedule tracking and other systems to improve the overall level of railway engineering project management and decision support capabilities. 6. Wide applicability and strong scalability: This method is not only applicable to multiple railway professional fields such as signaling, communication, and power, but the constructed model and data augmentation strategy also have good transferability and can be applied to engineering drawing recognition and digitization tasks in other industries with slight adjustments. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the method for digitizing railway engineering drawings and extracting engineering quantities based on deep learning, provided in an embodiment of the present invention. Figure 2 This is an overall system block diagram provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] It should be noted that the structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should fall within the scope of the technical content disclosed in the present invention. It should be noted that in this specification, relational terms such as "first" and "second" are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0017] This invention provides an embodiment: like Figure 1 , Figure 2 As shown in the figure, this embodiment of the invention provides a method for digitizing railway engineering drawings and extracting engineering quantities based on deep learning, including the following steps: S1: Establish a railway professional identification database, classify and label the icons and atlases in the database, and perform data augmentation to obtain standard engineering design drawings.

[0018] In this embodiment, a railway professional identification database is established by collecting various icons and atlases related to railway engineering, including drawings related to signaling, communication, and power. Elements in the icons and atlases are categorized and labeled; for example, signal icons are divided into categories such as switch machines, signal lights, and track circuits. Each element is labeled in detail, including its name, specifications, and purpose.

[0019] Optionally, the data augmentation includes at least one or a combination of the following: deep learning-based adaptive histogram equalization, random brightness adjustment, motion blur addition, and image horizontal or vertical flipping.

[0020] In this embodiment, the original engineering design drawings are augmented and enhanced using methods such as an adaptive histogram equalization algorithm based on deep learning, random brightness, motion blur, and horizontal or vertical flipping of the drawings.

[0021] In this embodiment, the adaptive histogram equalization algorithm uses a monotonically nonlinear mapping. The histogram is distributed as follows Image A becomes distributed as Image B, monotonic nonlinear mapping The function is shown in the following formula: (1) In the formula, This represents the starting value of the pixel grayscale value in image A; This represents the interval length of pixel grayscale values ​​in image A; This represents the starting value of the pixel grayscale value in image B; This represents the range length of pixel grayscale values ​​in image B.

[0022] Formula (1) can be understood as the total number of pixels within the corresponding interval remaining constant. Therefore, there is a special case with formula (2), and under ideal circumstances... ,in It is the number of pixels. This is the grayscale depth, usually taken as 256, resulting in formula (3): (2) (3) Then monotone nonlinear mapping The solution to formula (4) is as follows: (4) In this embodiment, when using horizontal or vertical image flipping and random brightness adjustment, the classified and labeled icons and atlases are rotated at random angles and flipped horizontally or vertically, and color enhancement is used to increase the model's robustness to rotation, mirror transformation and brightness.

[0023] In this embodiment, when data augmentation is performed using noise enhancement and motion blur addition, Gaussian noise is added to improve its adaptability to noisy environments. Adding motion blur to the image increases the training difficulty of the model, enhances its generalization ability, and improves the final recognition performance.

[0024] S2: Use the improved YOLOv8 model to extract image features and detect targets in the standard engineering design drawing. The improved YOLOv8 model uses the spatial and channel reconstruction convolution SCConv module to replace the traditional convolution in the C2f module. It introduces the EMA efficient multi-scale attention mechanism in the backbone network and the neck network, and introduces the RFA receptive field attention mechanism to output the category, probability and coordinate information of the features contained in the standard engineering design drawing.

[0025] In this embodiment, the improved YOLOv8 model's network structure mainly consists of three parts: Backbone, Neck, and Head. The backbone network is used to extract target features and consists of convolutional modules (Conv), C2f structures, and the SPPF module used in YOLOv5. The Neck is used to enhance and fuse features from different dimensions, and its structure follows the Feature Pyramid Network (FPN) and Path Aggregation Network (PAN) structures. The Head part calculates the enhanced features to obtain the confidence and location of different targets. The C2f module uses spatial and channel reconstruction convolutions (SCConv) instead of traditional convolutions.

[0026] Optionally, the SCConv module further includes an SRU module for suppressing spatial redundancy and a CRU module for reducing channel redundancy.

[0027] In this embodiment, the spatial and channel reconstruction convolution SCConv is used instead of the traditional convolution in the C2f module of YOLOv8. The SRU module, which suppresses spatial redundancy through a separation-reconstruction method, and the CRU module, which reduces channel redundancy through a "segmentation-transformation-fusion" strategy, reduce computational costs and model storage, and improve model performance by reducing the spatial and channel redundancy present in standard convolution.

[0028] Optionally, the RFA receptive field attention mechanism can be combined with convolutional blocks to improve the improved YOLOv8 model's ability to perceive large-scale convolutional kernels and the overall network performance.

[0029] In this embodiment, the EMA efficient multi-scale attention mechanism is added to the backbone network and the neck network to enhance the feature extraction results. It retains the information on each channel and reduces the computational overhead, and groups the channel dimensions into multiple sub-features, so that the spatial semantic features are evenly distributed in each feature group.

[0030] In this embodiment, the idea of ​​combining the RFA receptive field attention mechanism with convolutional blocks is introduced to solve the problem that the performance of spatial attention mechanism is limited by large-scale convolutional kernels, thereby further improving network performance.

[0031] In this embodiment, YOLOv8 is used for image feature extraction and target detection of standard engineering design drawings. The output results are marked with the judgment probability, category, and coordinate information of each feature to facilitate the secondary use of the identified data. To verify the effectiveness of the YOLOv8 model constructed by this method, the following formula is used as the model evaluation index: Accuracy: Recall rate: Average precision (map): F1Score: Intersection over Union (IOU): Among them, P (Precision), R (Recall), mAP 0.5 (Mean Average Precision: the average AP of each class at IOU=0.5), mAP 0.5:0.95 (Mean Average Precision: the average AP of each class at different IOUs), and F1 Score (the harmonic mean of PR, which can distinguish the quality of the algorithm) are used as model evaluation metrics.

[0032] S3: Use the PaddleOCR text detection model to extract text information from the standard engineering design drawings.

[0033] In this embodiment, the PaddleOCR text detection model is used to extract text data from standard engineering design drawings, excluding image features. Text data from standard engineering design drawings is extracted for information such as indoor cabinets and icon structures to ensure the data integrity of the standard engineering design drawings.

[0034] Optionally, the method further includes: filtering drawings of the same type by drawing name through the design platform API interface, and uniformly converting the selected drawings into IMG images; using OpenCV-based image table preprocessing operations to extract grid lines from the IMG images to form a table frame without filled information; using the PaddleOCR text recognition model to extract text information from the IMG images and recording the relative positional relationship of the extracted text information; and filling the extracted text information into the table frame to achieve structured acquisition of chart data.

[0035] In this embodiment, the text detection PaddleOCR model is used to extract text data from standard engineering design drawings. This is for a large amount of indoor cabinet drawing information. Since the chart structures of cabinets of the same type are relatively consistent, the same type of drawings can be filtered by reading the drawing name through the design platform API interface. As a common recognition task, the drawings are first converted into IMG format. Then, an image table preprocessing operation based on OpenCV is used to extract the grid lines in the image to form the charts of each cabinet without information. Finally, the text information of the IMG format icons is extracted using the PaddleOCR text recognition model. The relative positional relationship of the extracted text information is recorded and filled into the preprocessed grid line table to realize the sequential acquisition of different types of chart data.

[0036] S4: Digitize the identification results in steps S2 and S3 into a structured data format, compare them with the pre-stored on-site engineering list, automatically calculate the engineering quantity based on the comparison results, and summarize the engineering quantity calculation results to form an enterprise-level structured coding system.

[0037] Optionally, the data is digitized into a structured data format called JSON, which includes the element's ID number, name, type, location coordinates, region, and other attribute information.

[0038] Optionally, the comparison with the pre-stored on-site engineering list includes: matching the name, specifications, and usage information of the digital elements with the items in the on-site list to determine the correspondence; the automatic calculation of the engineering quantity includes: calculating the engineering quantity of each element based on the unit price and quantity information in the on-site list, combined with the actual distribution in the standard engineering design drawings, and adjusting it according to engineering standards and experience, taking into account a reasonable loss coefficient.

[0039] Optionally, the aggregation to form an enterprise-level structured coding system includes: allocating the calculated engineering quantities according to the determined enterprise-level hierarchical structure, from the project level down to the system level, subsystem level and equipment level, and allocating cross-level engineering quantities according to their actual functions and connection relationships.

[0040] In this embodiment, after identifying the image features and text information in the engineering drawings, they are digitized and converted into a computer-processable JSON format, as shown below: { "ID": "0x1004", "name": "S4", "type": "0x044A", "point": { "x": 960, "y": 388 }, "fromPD":0, "keyArea": ​​1 } Among them, ID represents the unique identification number; name represents the name identified from the drawing; type represents the identified signal type, such as the "0x044A" identified above, which is a type of exit signal; point represents the identified signal location information; fromPD represents the attribute set for signals with slopes; and keyArea represents the throat area to which it belongs.

[0041] In this embodiment, after digitizing the standard engineering design drawings, the digitized standard engineering design drawings are then interfaced with a pre-stored on-site engineering list. By comparing the elements in the standard engineering design drawings with the items in the pre-stored on-site engineering list, the correspondence between the elements in the standard engineering design drawings and the actual project is determined. The quantities of all elements in the standard engineering design drawings are then calculated, and the calculation results are summarized. For example, a signal icon in the standard engineering design drawing corresponds to a signal item in the on-site list, including information such as the signal's specifications, quantity, location type, and electrification type. The implementation steps are as follows: (1) Preparation Ensure that the digitized railway drawings are accurate, with icons, elements, and other information clearly identifiable and correctly labeled and categorized; compile a site inventory list to ensure that project names, specifications, quantities, and other information are complete and standardized, so as to accurately compare them with the elements in the drawings.

[0042] (2) Preliminary comparison and determination of correspondence The digitized drawings and pre-stored on-site engineering lists are displayed simultaneously for easy comparison. Starting with an element in the drawing, such as the icon of a specific type of signal device, the corresponding item is found in the pre-stored on-site engineering list. Key information such as the element's name, specifications, and purpose are compared to determine if they match. For more complex elements, their detailed parameters and feature descriptions need to be examined to ensure accurate correspondence. For example, for electrified equipment, not only the model number but also its technical parameters and installation requirements must be considered. Each element in the drawing is compared sequentially to gradually determine the correspondence between all elements in the on-site engineering list.

[0043] (3) Calculation of project quantity For elements with established correspondences, calculate the quantity of that element based on the quantity information in the pre-stored on-site engineering list and the actual distribution in the drawings. For example, if the quantity of a certain type of cable in the pre-stored on-site engineering list is 100 meters, but the drawings show that the total length of the cable at different locations is 80 meters, then 80 meters is taken as the actual quantity of that element. For elements requiring special calculation methods, such as calculating the length of railway track at curves, accurate calculations must be performed based on the specific geometry and calculation formulas. Potential losses and allowances should be considered. Based on engineering experience and relevant standards, a reasonable loss coefficient should be determined, and the quantity of work should be adjusted accordingly. For example, for cable laying projects, a certain proportion of losses is usually considered and included in the quantity calculation.

[0044] (4) Summary of project quantity results The results of the quantity calculations for all elements are summarized and organized according to different parts of the project, professional fields, or other classification methods to form a detailed bill of quantities.

[0045] In this embodiment, the quantities of all elements in the drawings are calculated, the results are summarized, and finally presented using EBS (Enterprise Business Structured Coding System) for project management and cost control. The implementation steps are as follows: (1) Determine the EBS hierarchical structure Based on the characteristics and management needs of railway engineering, the hierarchical structure of the Engineering Scale (EBS) is determined. It can generally be divided into multiple levels, such as project level, system level, subsystem level, and equipment level. For example, at the project level, it can be divided into the overall railway engineering project; at the system level, it can be further divided into signaling systems, communication systems, power systems, etc.; at the subsystem level, it can be further subdivided according to specific functional modules, such as the interlocking subsystem and block subsystem within the signaling system; and at the equipment level, it refers to specific equipment and components, such as signals, switch machines, and transformers.

[0046] (2) Allocate the work volume to the EBS level Following the established EBS hierarchical structure, the compiled engineering quantity data is progressively allocated to each level. Starting from the project level, allocation proceeds downwards to the system level, subsystem level, and equipment level. For engineering quantities spanning multiple levels, allocation needs to be made rationally based on their actual affiliation and function. For example, some cables may simultaneously provide power to multiple subsystems, requiring allocation based on the equipment and functions they connect to.

[0047] (3) Refine and improve EBS During the allocation of work quantities, the EBS structure should be continuously refined and improved. For some complex subsystems or equipment, the levels can be further subdivided to better manage and control the work quantities. Necessary descriptive information and annotations should be added to make the EBS clearer and easier to understand. For example, detailed descriptions of the name, function, and included work quantities of each level should be provided to facilitate review and understanding by engineering personnel.

[0048] (4) EBS output The finalized EBS (Engineering Standards) will be output in the form of electronic documents, charts, etc., for use in project management, cost control, and schedule planning.

[0049] In this embodiment, the software is designed in a B / S mode, which improves the efficiency and accuracy of digitizing railway engineering drawings and extracting engineering quantities, and provides strong support for the design, construction and management of railway engineering.

[0050] This invention provides a deep learning-based method for digitizing railway engineering drawings and extracting quantities. It constructs a railway professional identifier database based on railway icons and atlas libraries, utilizes deep learning-based adaptive histogram equalization and other methods for data augmentation, and improves the YOLOv8 model for feature extraction and object detection in engineering design drawings. Simultaneously, it uses the PaddleOCR text detection model to recognize railway drawings used in other engineering design drawings, thus digitizing the drawings. Then, it integrates with the actual needs list from the construction site to automatically calculate detailed quantities. The final result is presented in the form of an Enterprise Structured Code (EBS). This invention improves the efficiency and accuracy of railway engineering drawing digitization and quantity extraction, providing strong support for the design, construction, and management of railway projects.

[0051] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for digitizing railway engineering drawings and extracting quantities based on deep learning, characterized in that, Includes the following steps: S1: Establish a railway professional identification database, classify and label the icons and atlases in the railway professional identification database, and perform data augmentation to obtain standard engineering design drawings; S2: Use the improved YOLOv8 model to extract image features and detect targets in the standard engineering design drawing. The improved YOLOv8 model uses the spatial and channel reconstruction convolution SCConv module to replace the traditional convolution in the C2f module. It introduces the EMA efficient multi-scale attention mechanism in the backbone network and the neck network, and introduces the RFA receptive field attention mechanism to output the category, probability and coordinate information of the features contained in the standard engineering design drawing. S3: Use the PaddleOCR text detection model to extract the text information from the standard engineering design drawings; S4: Digitize the identification results in steps S2 and S3 into a structured data format, compare them with the pre-stored on-site engineering list, automatically calculate the engineering quantity based on the comparison results, and summarize the engineering quantity calculation results to form an enterprise-level structured coding system.

2. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, The data augmentation includes at least one or a combination of the following methods: deep learning-based adaptive histogram equalization, random brightness adjustment, motion blur addition, and horizontal or vertical image flipping.

3. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, The SCConv module further includes an SRU module for suppressing spatial redundancy and a CRU module for reducing channel redundancy.

4. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, In step S2, the RFA receptive field attention mechanism is introduced and combined with convolutional blocks to improve the improved YOLOv8 model's ability to perceive large-scale convolutional kernels and the overall network performance.

5. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, Step S3 further includes: Filter drawings of the same type by drawing name through the design platform API interface, and convert the selected drawings into IMG images; An image table preprocessing operation based on OpenCV is used to extract grid lines from the IMG image to form a table frame without filled information. The PaddleOCR text recognition model is used to extract text information from the IMG image, and the relative positional relationships of the extracted text information are recorded. The extracted text information is filled into the table frame to achieve structured acquisition of chart data.

6. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, In step S4, the data is digitized into a structured data format called JSON, which includes the element's ID number, name, type, location coordinates, region, and other attribute information.

7. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, In step S4, the comparison with the pre-stored on-site engineering list includes: matching the name, specifications, and usage information of the digital elements with the items in the on-site list to determine the correspondence; the automatic calculation of the engineering quantity includes: calculating the engineering quantity of each element based on the unit price and quantity information in the on-site list, combined with the actual distribution in the drawings, and adjusting it according to engineering standards and experience, taking into account a reasonable loss coefficient.

8. The method for digitizing railway engineering drawings and extracting quantities based on deep learning according to claim 1, characterized in that, In step S4, the process of summarizing and forming an enterprise-level structured coding system includes: allocating the calculated engineering quantities according to the determined enterprise-level hierarchical structure, from the project level down to the system level, subsystem level and equipment level, and allocating cross-level engineering quantities according to their actual functions and connection relationships.

Images