Transformer substation engineering construction drawing intelligent analysis and evaluation system based on image recognition

Abstract

The invention discloses a transformer substation engineering construction drawing intelligent analysis and evaluation system based on image recognition, and relates to the technical field of transformer substation engineering construction informatization. The data processing center is in communication connection with a data acquisition layer, an image preprocessing layer, a feature extraction layer, an intelligent analysis layer, an evaluation decision-making layer and an application display layer, and all layers of architectures are in electric signal connection. According to the method, the improved convolutional neural network model is adopted, and an image preprocessing algorithm specially aiming at engineering drawing feature optimization is combined, so that the drawing recognition accuracy is remarkably improved compared with a traditional OCR technology, the manual proofreading workload is greatly reduced, a drawing element correlation analysis system based on a knowledge graph is constructed, and the engineering drawing recognition accuracy is improved. Key information such as equipment arrangement, line connection and size marking in the drawing can be automatically extracted and associated, and structural storage and intelligent retrieval of drawing content are achieved.

Description

Technical Field

[0001] The present invention relates to the field of information technology for substation engineering construction, and in particular to an intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition. Background Art

[0002] With the continuous economic growth and social progress, the power industry has ushered in unprecedented development opportunities. As an important part of the power system, the number and scale of substations are also expanding. However, traditional substation management methods, such as manual review and on-site investigation, can no longer meet the needs of the rapid development of the current power industry. Therefore, with the development of information technology, the demand for digital management of construction drawings in the field of power engineering construction is becoming increasingly urgent. By converting paper drawings into digital formats, efficient storage, rapid retrieval and convenient sharing of drawings can be achieved, thereby improving the efficiency and quality of drawing management.

[0003] The prior art has the following problems:

[0004] 1. The substation construction drawings contain a large number of design elements related to green construction. Traditional manual methods are difficult to quickly and accurately extract and identify green element information from complex drawings;

[0005] 2. Existing technical standards have put forward higher requirements for green construction, and the evaluation index system is more complete. The method of manually scoring item by item against the specifications is inefficient and prone to omissions. In addition, different evaluators have different understandings of the green factor judgment standards, resulting in a lack of consistency and comparability in the scoring results;

[0006] 3. The existing evaluation method makes it difficult to establish the correlation between the green elements of the drawings and the scoring basis, and cannot form a complete chain of evaluation evidence, which affects the traceability and persuasiveness of the evaluation results. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition to solve the problems raised in the above-mentioned background technology.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition includes a data processing center, which is communicatively connected with a data acquisition layer, an image preprocessing layer, a feature extraction layer, an intelligent analysis layer, an evaluation decision layer, and an application display layer, wherein electrical signals are connected between the architecture layers;

[0010] The data acquisition layer, as the basic data entry of the system, adopts a distributed acquisition architecture to perform parallel acquisition and processing of construction drawings of various types of substation projects, so as to obtain the original data of the substation project construction drawings and establish a distributed drawing index database for storing and managing the metadata information of the drawings;

[0011] The image preprocessing layer is used to perform standardization processing and quality optimization on the collected original drawing data. Through format conversion, image enhancement, and layout analysis, drawings in different formats are uniformly converted into a standardized image format, while maintaining the accuracy and clarity of the drawings, improving the image quality, and providing an important basis for subsequent recognition and analysis;

[0012] The feature extraction layer adopts a multi-modal feature extraction strategy, integrating text recognition, graphic feature extraction, and symbol recognition, to extract various types of feature information from the preprocessed drawings, achieving high-precision feature extraction and significantly improving the comprehensiveness and accuracy of feature extraction;

[0013] The intelligent analysis layer, based on knowledge graph and deep learning technologies, performs semantic analysis, element recognition, and correlation analysis on the extracted features, providing reliable data support for subsequent evaluation and decision-making;

[0014] The evaluation and decision-making layer constructs a comprehensive drawing quality evaluation index system, determines the weights using the fuzzy comprehensive evaluation method and the analytic hierarchy process, and conducts a comprehensive evaluation of the green construction level based on the analysis results. The evaluation results are verified and checked from multiple dimensions through an expert knowledge base and a historical case base, ensuring the reliability and persuasiveness of the evaluation results, and realizing the full-process traceability and evidence chain management of the evaluation process;

[0015] The application display layer adopts a Web technology architecture, supports multi-terminal access, provides an intuitive operation interface and automatically generates a standardized evaluation report, and uses visualization technology to intuitively display the evaluation results and analysis data, realizing real-time data update and interactive operation functions, and greatly improving the practicality and usability of the system.

[0016] A further improvement of the technical solution of the present invention lies in that in the data acquisition layer, the process of collecting the original data and establishing the distributed drawing index database is as follows:

[0017] The system receives substation project construction drawing data from different sources (such as design units, construction units, etc.), and the construction drawing data exists in the form of CAD formats (.dwg / .dxf), PDF documents, or various image formats (.jpg / .png / .tiff);

[0018] Preprocess the received drawings, including denoising and format conversion, to ensure the consistency and accuracy of subsequent processing. Use the distributed acquisition architecture to process the preprocessed drawings in parallel, extract key information from the drawings, and generate drawing summaries.

[0019] For each piece of construction drawing data collected, encode it according to its attributes, establish a three-level encoding system based on "project number-drawing type-version number", and formulate unified encoding rules to ensure the consistency and standardization of encoding. Among them, the project number is used to uniquely identify a substation project to ensure that the drawing data between different projects can be accurately distinguished. According to the content and purpose of the drawings, the drawings are classified and the types of drawings are determined, including electrical drawings, structural drawings and civil engineering drawings, etc., to facilitate subsequent classification management and retrieval. The version information of the drawings is recorded, including the initial version and subsequent revised versions, to ensure that the update and modification process of the drawings can be tracked;

[0020] The encoded drawing data is stored in the distributed drawing index database, and the metadata information of the drawings is stored and managed in the database, including the drawing name, creation time, revision record, design unit, construction unit, drawing size and scale information. The metadata information provides an important reference for the management and retrieval of drawings, and the drawing data in the database is backed up regularly.

[0021] A further improvement of the technical solution of the present invention is that: the image preprocessing layer includes a format conversion module, an image enhancement module and a layout analysis module, wherein the modules are connected by electrical signals;

[0022] The format conversion module uses parallel processing technology to convert drawings of different formats into a standardized image format and maintain the accuracy and clarity of the drawings;

[0023] The image enhancement module uses a variety of image processing algorithms to optimize the drawings, including denoising, enhancement and correction processing, to improve image quality;

[0024] The layout analysis module realizes intelligent partitioning and structured analysis of drawings based on deep learning layout analysis technology.

[0025] A further improvement of the technical solution of the present invention is that in the image preprocessing layer, the image format conversion process is:

[0026] Receive the original drawing data from the data acquisition layer through the format conversion module, identify the format of the input drawing data, determine its original format, and use parallel processing technology to simultaneously convert the drawing data in different formats into a standardized image format. For CAD format drawings, use multi-threading or multi-processing methods to simultaneously call multiple CAD conversion tools or library functions to convert them into image format. For PDF documents, use a PDF parsing library to extract graphic content in parallel and convert it into image format;

[0027] During the format conversion process, set the resolution and image quality parameters, and uniformly convert the drawings into a standard resolution of 300 DPI. This resolution is the best balance point verified through a large number of experiments, which can not only ensure the recognition accuracy but also control the consumption of computing resources, ensuring that the converted images can maintain the accuracy and clarity of the drawings. Then, transmit the converted standardized image format data to the image enhancement module through electrical signals to provide input for subsequent image optimization processing;

[0028] Use the image enhancement module to receive the standardized image format data from the format conversion module, perform denoising processing on the image using the Gaussian filtering denoising algorithm to eliminate noise points and interference information in the image, and adopt the adaptive histogram equalization method. According to the histogram distribution of the image, select the enhancement strategy and parameters to make the gray values of the image evenly distributed, thereby improving the contrast of the image;

[0029] Perform sharpening processing on the image, combine the Laplacian operator and nonlinear enhancement to highlight the edge and detail information of the image. According to the edge characteristics and sharpening requirements of the image, adjust the algorithm parameters to clearly present the lines and contours of the image, and perform correction processing on the image to correct the geometric distortion and rotation deviation problems in the image. According to the feature points or reference objects of the image, calculate the correction parameters and perform corresponding geometric transformations on the image to restore the image to the correct geometric form. Furthermore, transmit the optimized image data to the layout analysis module through electrical signals to provide high-quality image input for subsequent intelligent zoning and structured analysis;

[0030] Receive the high-quality image data from the image enhancement module through the layout analysis module, pre-train a deep learning model to obtain an improved Faster R-CNN model, and use the improved Faster R-CNN model for drawing layout analysis. Through the feature extraction and classification capabilities of the model, design regional classification criteria to establish a complete drawing regional classification system, and identify different regions in the drawing, including the title bar, design description area, main engineering drawing, node detail drawing, and equipment layout drawing;

[0031] According to the results of layout analysis recognition, different regions of the drawing are located and segmented. For each region, its position and range in the image are determined, and it is segmented from the overall image. Layout information is extracted from each of the segmented regions, including item information in the title block, text content in the design description area, and graphic symbols in the main engineering drawing, etc.;

[0032] OCR technology is used to recognize text information, image recognition and pattern matching technologies are used to recognize graphic symbols and annotation information, information on equipment layout and connection relationships is extracted through geometric analysis and topological relationships, and the extracted layout information is structured to establish a structured representation of the drawing content. The text information is organized according to paragraph and sentence structures, and the graphic symbols and annotation information are associated with the corresponding equipment to form a structured data format.

[0033] A further improvement in the technical solution of the present invention lies in: for the conversion of CAD drawings, an improved LibreCAD open-source framework is adopted. Through an optimized vector-raster conversion algorithm, the complete details of the drawing are retained. During the conversion process, an adaptive resolution control technology is adopted to perform intelligent scaling on drawings of different scales;

[0034] For the parsing and processing of PDFs, an improved PDF2Image algorithm is adopted. Through multi-threaded parallel processing technology, the conversion efficiency is improved. During the processing process, an adaptive page segmentation algorithm is introduced to recognize and extract graphic content in the PDF and control the parsing accuracy of the drawing;

[0035] For the process of using the adaptive histogram equalization method to make the gray values of the image evenly distributed, an improved CLAHE algorithm is adopted. Through dynamic block size adjustment and contrast limitation, the local contrast of the image is improved. The core of the algorithm includes image block division, histogram calculation, and bilinear interpolation. Among them, image block division adopts an adaptive block division strategy, and the block size is dynamically adjusted according to the complexity of the image content. Histogram statistics and equalization processing are performed on each block through histogram calculation, and bilinear interpolation is used for smooth processing of the transition between blocks;

[0036] The system uses an improved Faster R-CNN model for drawing layout analysis. Its core architecture includes backbone network optimization, Feature Pyramid Network (FPN), and RPN network improvement;

[0037] For backbone network optimization, ResNet-101 is used as the basic network, and the SE attention mechanism is introduced;

[0038] A Feature Pyramid Network (FPN) is used to construct multi-scale feature maps, which are divided into five levels from P2 to P6;

[0039] Through the improvement of the RPN network, dynamic anchor box generation and multi-scale feature matching are achieved.

[0040] A further improvement of the technical solution of the present invention lies in that: in the feature extraction layer, the process of feature extraction is as follows:

[0041] In the feature extraction layer, character recognition, graphic feature extraction, and symbol recognition are performed. Among them, for character recognition, it includes multi-scenario character recognition strategies and optimization of professional term recognition. For graphic feature extraction, it includes geometric feature analysis and topological relationship analysis. For symbol recognition and classification, it includes the construction of a standard symbol library and an intelligent recognition mechanism;

[0042] For the multi-scenario character recognition strategy, aiming at the character characteristics in different scenarios of engineering drawings, a hierarchical recognition strategy is adopted to classify the character areas in the drawings, including different types such as title characters, explanatory characters, dimension annotation characters, and legend characters. Each type of character has its specific font size, orientation, and density characteristics. Then, a feature model library is established for character area positioning. An improved EAST detector is used to improve the detection accuracy through multi-scale feature fusion, and character content recognition is performed. According to different scenarios, corresponding recognition models are selected. Among them, for horizontal characters, a standard OCR model is used, and for inclined characters, a direction-aware recognition model is used;

[0043] For the optimization of professional term recognition, aiming at the characteristics of professional terms in the field of construction engineering, a domain dictionary containing more than 50,000 professional entries is constructed. During the recognition process, the recognition accuracy of professional terms is adjusted through a combination of context semantic analysis and dictionary matching, and a self-learning mechanism is supported to continuously expand and optimize the professional dictionary.

[0044] A further improvement of the technical solution of the present invention lies in that: for the geometric feature analysis, a multi-level geometric feature extraction scheme is adopted to gradually construct a feature system from three levels of points, lines, and surfaces. For the topological relationship analysis, a complete topological relationship analysis system of graphic elements is established, including: spatial relationship analysis, connection relationship analysis, and inclusion relationship analysis;

[0045] For point feature extraction, key nodes in the drawing are extracted through an improved Harris corner detection algorithm. On the basis of the traditional algorithm, an adaptive threshold mechanism is added to improve the robustness of the detection. A local feature descriptor operator is introduced to describe the features of the surrounding area of each detected point. For line feature extraction, a multi-scale Hough transform combined with the LSD algorithm is used to achieve accurate extraction of straight line segments. By setting dynamic parameter thresholds, it adapts to line features of different thicknesses and lengths. For surface feature extraction, based on contour tracking and region growing algorithms, the features of closed regions are extracted, regular geometric shapes (rectangles, circles) and irregular regions are recognized, and their area and perimeter feature parameters are calculated;

[0046] For spatial relationship analysis, a spatial relationship model is established by calculating the relative positions, distances, and angle parameters between graphic elements, and a quadtree indexing structure is adopted to improve the spatial query efficiency. For connection relationship analysis, the connection methods between graphic elements are identified, including various forms such as welding, bolt connection, and hinge connection, and the connection type is accurately judged by means of feature template matching. For inclusion relationship analysis, a hierarchical structure of graphic elements is established to identify the subordinate relationships between components and support the combined analysis of complex components;

[0047] The standard symbol library includes a basic symbol set and a professional symbol set, and the intelligent recognition mechanism includes template matching, deep learning recognition, and context understanding;

[0048] A standard symbol library covering multiple specialties of architecture, structure, and hydropower is established, various engineering symbols stipulated by national standards are included, a basic symbol set including symbol graphics, meanings, and usage scenario information is established, and for special symbols in different professional fields, including water supply and drainage system symbols, HVAC symbols, and electrical symbols, classification and feature extraction are carried out to establish a professional symbol set;

[0049] For the intelligent recognition mechanism, a recognition strategy of multi-model fusion is adopted. For symbols with high standardization, an improved template matching algorithm is used for recognition by combining feature point matching and shape matching. For non-standard symbols, a deep learning model is used for recognition, and the generalization ability of the model is improved through transfer learning. Combining the text description and graphic features around the symbol, a semantic understanding model of the symbol is established to improve the recognition accuracy.

[0050] A further improvement of the technical solution of the present invention lies in that: in the intelligent parsing layer, the process of semantic analysis, element recognition, and association analysis of the extracted features is as follows:

[0051] Operations of engineering semantic understanding and intelligent rule parsing are carried out in the intelligent parsing layer, where engineering semantic understanding includes the construction of a drawing semantic network and an engineering knowledge graph, and intelligent rule parsing includes the parsing of design specifications and the intelligent extraction of parameters;

[0052] For the construction of the drawing semantic network, the in-depth understanding of engineering drawings is realized by establishing a multi-level semantic network, where the multi-level semantic network includes: a basic semantic layer, a relational semantic layer, and a knowledge semantic layer;

[0053] Capture the basic elements in the drawings at the basic semantic level, including components, dimensions, and annotation information, and establish preliminary associations between elements. Adopt improved semantic segmentation technology to classify the drawing content according to functional attributes, realize the accurate positioning and attribute extraction of elements. Analyze and establish the logical relationships between drawing elements at the relational semantic level, including spatial relationships, functional relationships, and structural relationships. Describe the complex engineering information of the connection methods between components and the force transmission paths by establishing a relational matrix. Integrate engineering domain knowledge at the knowledge semantic level, associate the drawing information with professional specifications and design standards, realize semantic understanding, and build a knowledge base containing national standards and industry specifications to support intelligent specification inspection;

[0054] Based on the semantic network, construct an engineering knowledge graph, including ontology framework design and instance relationship mapping;

[0055] Establish a professional ontology system covering the engineering domain for ontology framework design, including multiple dimensions such as component types, material properties, and construction techniques. Define standard attribute sets and relationship sets for each ontology category. Map the specific instances identified in the drawings to the ontology framework for instance relationship mapping, establish the association relationships between instances, and discover implicit engineering logical relationships through the relationship reasoning mechanism;

[0056] For the analysis of design specifications, structurally process various specification standards in the field of engineering construction, establish a specification knowledge base including clause requirements, parameter limits, and calculation formulas, and through natural language processing technology, convert the specification clauses into a computable rule set, understand the logical relationships in the clauses, and convert them into specific judgment criteria;

[0057] For intelligent parameter extraction, extract various engineering parameters from the drawings, including geometric parameter extraction and performance parameter extraction. Identify the dimension information of components through geometric parameter extraction, including basic parameters such as length, width, and height. Identify professional information such as material strength grades, component reinforcement information, and equipment performance parameters through performance parameter extraction, and automatically compare them with relevant specification requirements.

[0058] A further improvement in the technical solution of the present invention lies in: in the evaluation and decision-making layer, the process of comprehensive evaluation of the green building level is as follows:

[0059] Perform operations of constructing an evaluation index system and analyzing evaluation results in the evaluation and decision-making layer. Among them, the evaluation index system includes a basic index layer and a professional index layer, and the evaluation result analysis includes compliance analysis, integrity analysis, and consistency analysis;

[0060] Establish a comprehensive drawing quality evaluation index system, analyze the standardization of drawings, accuracy of graphic expression and design integrity at the basic index level, evaluate the integrity of basic elements of drawings, rationality of layer settings and standardization of annotations, evaluate the accuracy of component expression, clarity of detailed drawing expression and rationality of section expression, and evaluate whether the design depth meets the requirements and whether the design elements are complete;

[0061] Analyze the rationality of structural design, adaptability of building functions and coordination of equipment systems at the professional indicator level, evaluate the rationality of structural layout, component size, node practices, etc., evaluate the compliance of spatial layout, fire protection requirements, barrier-free design, etc., and evaluate the appropriateness of various pipeline layouts, equipment selection, etc.;

[0062] In the analysis of evaluation results, compliance analysis automatically checks against design specifications, identifies design content that does not meet specification requirements, generates specification compliance scores and detailed analysis reports. For integrity analysis, the integrity of design documents is checked, whether the design depth meets requirements, and missing items of key design information are identified. For consistency analysis, the coordination between professional drawings is checked, design contradictions and incoherent information are marked, and a cross-professional coordination evaluation report is generated.

[0063] A further improvement of the technical solution of the present invention is that in the application display layer, the process of generating the standardized evaluation report is as follows:

[0064] The evaluation report generation includes a comprehensive evaluation report, a special inspection report and a dynamic tracking report;

[0065] For the comprehensive evaluation report, analyze the overall score and sub-item scores, the list of major issues and rectification suggestions, and the graphical visualization of the evaluation results;

[0066] For special inspection reports, analyze the special inspection results, problem location and specific descriptions, as well as improvement suggestions and reference plans for specific design requirements;

[0067] For dynamic tracking reports, record the problem solving status during the design modification process, the dynamic change trend of evaluation indicators, and the analysis of continuous improvement effects.

[0068] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art:

[0069] 1. The present invention provides an intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition. By adopting an improved convolutional neural network model and combining an image preprocessing algorithm optimized specifically for the characteristics of engineering drawings, the drawing recognition accuracy is significantly improved compared with traditional OCR technology, greatly reducing the manual proofreading workload. A drawing element association analysis system based on a knowledge graph is constructed, which can automatically extract and associate key information such as equipment layout, line connection, and dimension marking in the drawings, realizing the structured storage and intelligent retrieval of drawing content.

[0070] 2. The present invention provides an intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition. By introducing a drawing quality evaluation mechanism based on deep reinforcement learning and comparing with industry standard specifications in real time, design defects and specification deviations existing in the drawings can be automatically identified, significantly improving the efficiency of design quality control. The system adopts a distributed architecture design, supports multi-terminal collaborative operations, and can realize parallel processing and real-time evaluation of drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0072] Figure 1 It is a block diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0074] Example 1, as Figure 1 shown, the present invention provides an intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition, including a data processing center. The data processing center is communicatively connected to a data acquisition layer, an image preprocessing layer, a feature extraction layer, an intelligent analysis layer, an evaluation and decision-making layer, and an application display layer. Among them, electrical signal connections are provided between the architectures of each layer;

[0075] The data acquisition layer, as the basic data entry of the system, adopts a distributed acquisition architecture to perform parallel acquisition and processing of construction drawings of various types of substation projects, so as to obtain the original data of substation project construction drawings, including inputs in CAD formats (.dwg / .dxf), PDF documents, and various image formats (.jpg / .png / .tiff). It has implemented a three-level coding system based on "project number - drawing type - version number" to ensure data traceability and version management, and established a distributed drawing index database for storing and managing the metadata information of the drawings. The system receives substation project construction drawing data from different sources (design units, construction units, etc.). The construction drawing data exists in the forms of CAD formats (.dwg / .dxf), PDF documents, or various image formats (.jpg / .png / .tiff). The received drawings are preprocessed, including denoising and format conversion, to ensure the consistency and accuracy of subsequent processing. And using the distributed acquisition architecture, the preprocessed drawings are processed in parallel to extract the key information in the drawings and generate drawing summaries. For each piece of construction drawing data collected, it is encoded according to its attributes, and a three-level coding system based on "project number - drawing type - version number" is established, and unified coding rules are formulated to ensure the consistency and standardization of the coding;

[0076] Among them, a substation project is uniquely identified by a project number to ensure that the drawing data between different projects can be accurately distinguished. The drawing types are determined according to the content and use of the drawings, including electrical drawings, structural drawings, civil engineering drawings, etc., which is convenient for subsequent classification management and retrieval. The version information of the drawings is recorded, including the initial version and subsequent revised versions, to ensure that the update and modification process of the drawings can be traced. The encoded drawing data is stored in the distributed drawing index database. The distributed drawing index database adopts a distributed architecture to store a large amount of drawing data and provide efficient retrieval and query functions. The metadata information of the drawings, including drawing name, creation time, revision record, design unit, construction unit, drawing size, and scale information, is stored and managed in the database. The metadata information provides an important reference basis for the management and retrieval of the drawings. The drawing data in the database is backed up regularly to ensure data security and recoverability. In case of data loss or damage, the data is restored in time to ensure the normal operation of the system;

[0077] The image preprocessing layer is used to perform standardized processing and quality optimization on the original drawing data collected. Through format conversion, image enhancement, and layout analysis, drawings in different formats are uniformly converted into a standardized image format, and the accuracy and clarity of the drawings are maintained to improve the image quality, providing an important basis for subsequent recognition and analysis;

[0078] Furthermore, the image preprocessing layer includes a format conversion module, an image enhancement module, and a layout analysis module. Among them, the modules are electrically connected. The format conversion module uses parallel processing technology to uniformly convert drawings in different formats into a standardized image format, while maintaining the accuracy and clarity of the drawings. The image enhancement module uses a variety of image processing algorithms to optimize the drawings, including denoising, enhancement, and correction processing, to improve the image quality. The layout analysis module, based on deep learning layout analysis technology, realizes intelligent partitioning and structured parsing of the drawings;

[0079] Furthermore, in the image preprocessing layer, the process of converting the image format is as follows:

[0080] The format conversion module receives the original drawing data from the data acquisition layer, identifies the format of the input drawing data, determines its original format, and uses parallel processing technology to simultaneously convert drawing data in different formats into a standardized image format. For CAD format drawings, use multi-threading or multi-processing methods to simultaneously call multiple CAD conversion tools or library functions to convert them into image format. For PDF documents, use a PDF parsing library to parallel extract graphic content and convert it into image format. During the format conversion process, set the resolution and image quality parameters, and uniformly convert the drawings into a standard resolution of 300 DPI. This resolution is the best balance point verified by a large number of experiments, which can not only ensure the recognition accuracy but also control the consumption of computing resources, ensure that the converted image can maintain the accuracy and clarity of the drawings, and transmit the converted standardized image format data to the image enhancement module through electrical signals to provide input for subsequent image optimization processing. The image enhancement module receives the standardized image format data from the format conversion module, uses the denoising algorithm of Gaussian filtering to denoise the image, eliminates the noise points and interference information in the image, adopts the method of adaptive histogram equalization, selects the enhancement strategy and parameters according to the histogram distribution of the image, makes the gray values of the image evenly distributed, thereby improving the contrast of the image, sharpens the image, combines the Laplacian operator and nonlinear enhancement to highlight the edge and detail information of the image, adjusts the algorithm parameters according to the edge characteristics and sharpening requirements of the image, makes the lines and contours of the image clearly presented, and corrects the image to correct the geometric distortion and rotation deviation problems in the image;

[0081] Calculate the correction parameters based on the feature points or reference objects of the image, perform corresponding geometric transformations on the image to restore it to the correct geometric form, and then transmit the optimized image data to the layout analysis module through electrical signals to provide high-quality image input for subsequent intelligent zoning and structured analysis. The layout analysis module receives the high-quality image data from the image enhancement module, pre-trains a deep learning model to obtain an improved Faster R-CNN model, and uses the improved Faster R-CNN model for drawing layout analysis. Through the feature extraction and classification capabilities of the model, design area classification criteria to establish a complete drawing area classification system, and identify different areas in the drawing, including the title block, design description area, main engineering drawing, detail drawing of nodes, and equipment layout drawing. Among them, for the identification of the drawing title block, the position feature is to give priority to searching in the lower right corner area, the structural feature is a regular table structure, and the content feature is keyword matching. For the identification of the design description area, it includes text density analysis, paragraph structure recognition, and font size distribution characteristics. For the identification of the main engineering drawing, it includes graphic complexity evaluation, connected component analysis, and geometric feature extraction. For the identification of the detail drawing of nodes, it includes local feature enhancement, scale detection, and annotation density analysis. For the identification of the equipment layout drawing, it includes symbol distribution density, regularity analysis, and connection relationship extraction. The identification of each area adopts a multi-feature fusion method, and the final classification result is determined through a weighted voting mechanism:

[0082]

[0083] Among them, w i is the weight of each feature, and f i (x) is the corresponding feature function;

[0084] Based on the results of layout analysis and recognition, locate and segment different areas of the drawing. For each area, determine its position and range in the image, and segment it from the overall image. Extract layout information from the segmented areas, including item information in the title block, text content in the design description area, and graphic symbols in the main engineering drawing, etc. Use OCR technology to recognize text information, and use image recognition and pattern matching technology to recognize graphic symbols and annotation information. Extract equipment layout and connection relationship information through geometric analysis and topological relationships, and perform structured processing on the extracted layout information to establish a structured representation of the drawing content. Organize the text information according to paragraph and sentence structures, associate the graphic symbols and annotation information with the corresponding equipment, and form a structured data format;

[0085] For the conversion of CAD drawings, an improved open-source LibreCAD framework is adopted. Through an optimized vector-raster conversion algorithm, the complete details of the drawings are retained. During the conversion process, an adaptive resolution control technology is used to perform intelligent scaling on drawings of different scales to ensure the clarity and recognizability of the output images. The mathematical model of the conversion process is expressed as:

[0086]

[0087] Among them, P(x,y) is the output pixel value, V(i,j) is the original vector data, and K is the conversion kernel function;

[0088] For the parsing and processing of PDFs, an improved PDF2Image algorithm is adopted. Through multi-threaded parallel processing technology, the conversion efficiency is improved. During the processing, an adaptive page segmentation algorithm is introduced to identify and extract the graphic content in the PDF and control the parsing accuracy of the drawings. The parsing accuracy control formula:

[0089]

[0090] For the process of using the adaptive histogram equalization method to make the image gray values evenly distributed, an improved CLAHE algorithm is adopted. Through dynamic block size adjustment and contrast limitation, the local contrast of the image is enhanced. The core of the algorithm includes image block division, histogram calculation, and bilinear interpolation. Among them, the image block division adopts an adaptive block division strategy, and the block size is dynamically adjusted according to the complexity of the image content. Through histogram calculation, histogram statistics and equalization processing are performed within each block, and bilinear interpolation is used for smooth processing of the transition between blocks. The processing formula:

[0091]

[0092] Among them, α(i,j) is the adaptive weight factor, which is dynamically adjusted according to local image features;

[0093] An adaptive kernel size Gaussian filter is adopted, and the kernel function is designed as follows:

[0094]

[0095] Among them, the σ value is automatically adjusted through image noise level evaluation:

[0096]

[0097] k is the adjustment coefficient, which is obtained through experimental optimization;

[0098] Image sharpening and enhancement:

[0099] The system adopts a multi-scale sharpening strategy, combined with the Laplacian operator and non-linear enhancement:

[0100]

[0101] Among them, λ is an adaptive enhancement coefficient, which is determined according to the evaluation of image clarity:

[0102] λ = f(S clarity )

[0103]

[0104] The system uses an improved Faster R-CNN model for drawing layout analysis. Its core architecture includes backbone network optimization, Feature Pyramid Network (FPN), and RPN network improvement;

[0105] For backbone network optimization, ResNet-101 is used as the basic network, and the SE attention mechanism is introduced. The expression for optimizing the gradient flow through residual connection is:

[0106] x l+1 = x l + F(x l , W l )

[0107] The expression for improving the feature extraction efficiency is:

[0108] SE(F) = F · σ(W2δ(W1F avg ))

[0109] Using the Feature Pyramid Network (FPN) to construct multi-scale feature maps, which are divided into five levels from P2 to P6. Among them, the expression for the feature fusion strategy is:

[0110] P l = Conv(U(P +1 ) + C l )

[0111] Among them, U is the upsampling operation, and C l is the feature map of the corresponding layer;

[0112] Through the improvement of the RPN network, dynamic anchor box generation and multi-scale feature matching are realized. Among them, the expression for optimizing the loss function is:

[0113] L = L cls + λL reg + γL center ;

[0114] For the feature extraction layer, a multi-modal feature extraction strategy is adopted, integrating text recognition, graphic feature extraction, and symbol recognition. Various types of feature information are extracted from the preprocessed drawings, achieving high-precision feature extraction and significantly improving the comprehensiveness and accuracy of feature extraction;

[0115] The intelligent analysis layer, based on knowledge graph and deep learning technologies, performs semantic analysis, element recognition, and correlation analysis on the extracted features, providing reliable data support for subsequent evaluation and decision-making;

[0116] The evaluation and decision-making layer constructs a comprehensive drawing quality evaluation index system, determines weights using the fuzzy comprehensive evaluation method and the analytic hierarchy process, and conducts a comprehensive evaluation of the green building level based on the analysis results. It verifies and checks the evaluation results from multiple dimensions through an expert knowledge base and a historical case base, ensuring the reliability and persuasiveness of the evaluation results and realizing the full traceability of the evaluation process and evidence chain management;

[0117] The application display layer adopts a Web technology architecture, supports multi-terminal access, provides an intuitive operation interface and automatically generates a standardized evaluation report, visually displays the evaluation results and analysis data using visualization technology, and realizes real-time data update and interactive operation functions, greatly improving the practicality and usability of the system.

[0118] Embodiment 2, as Figure 1 shown, based on Embodiment 1, the present invention provides a technical solution: Preferably, in the feature extraction layer, the process of feature extraction is as follows:

[0119] In the feature extraction layer, character recognition, graphic feature extraction, and symbol recognition are performed. Among them, for character recognition, it includes multi-scene character recognition strategies and optimization of professional term recognition. For graphic feature extraction, it includes geometric feature analysis and topological relationship analysis. For symbol recognition and classification, it includes the construction of a standard symbol library and an intelligent recognition mechanism. For the multi-scene character recognition strategy, aiming at the character characteristics in different scenes of engineering drawings, a hierarchical recognition strategy is adopted to classify the character areas in the drawing into different types, including title characters, explanatory characters, dimension annotation characters, and legend characters. Each type of character has its specific font size, orientation, and density characteristics. Then, a feature model library is established for character area positioning. An improved EAST (Efficient and Accurate Scene Text) detector is used to improve the detection accuracy through multi-scale feature fusion, and character content recognition is performed. According to different scenes, the corresponding recognition model is selected. Among them, for horizontal characters, a standard OCR model is used, and for inclined characters, a direction-aware recognition model is used. For the optimization of professional term recognition, aiming at the characteristics of professional terms in the field of construction engineering, a domain dictionary containing more than 50,000 professional entries is constructed. During the recognition process, the recognition accuracy of professional terms is adjusted through a combination of context semantic analysis and dictionary matching, and a self-learning mechanism is supported to continuously expand and optimize the professional dictionary;

[0120] Furthermore, for geometric feature analysis, a multi-level geometric feature extraction scheme is adopted to gradually construct a feature system from three levels: points, lines, and surfaces. For topological relationship analysis, a complete topological relationship analysis system for graphic elements is established, including: spatial relationship analysis, connection relationship analysis, and inclusion relationship analysis. For point feature extraction, through an improved Harris corner detection algorithm, key nodes in the drawing are extracted. On the basis of the traditional algorithm, an adaptive threshold mechanism is added to improve the robustness of detection. A local feature description operator is introduced to describe the features of the surrounding area of each detected point. For line feature extraction, a multi-scale Hough transform combined with the LSD (Line Segment Detector) algorithm is used to achieve accurate extraction of straight line segments. By setting dynamic parameter thresholds, it adapts to line features of different thicknesses and lengths. For surface feature extraction, based on contour tracing and region growing algorithms, closed region features are extracted, regular geometric shapes (rectangles, circles) and irregular regions are identified, and their characteristic parameters of area and perimeter are calculated. For spatial relationship analysis, by calculating the relative position, distance, and angle parameters between graphic elements, a spatial relationship model is established, and a quadtree index structure is used to improve the spatial query efficiency. For connection relationship analysis, the connection methods between graphic elements are identified, including various forms such as welding, bolt connection, and hinge connection. Through feature template matching, the connection type is accurately judged;

[0121] For inclusion relationship analysis, a hierarchical structure of graphic elements is established to identify the subordinate relationships between components and support the combined analysis of complex components. The standard symbol library includes a basic symbol set and a professional symbol set. The intelligent recognition mechanism includes template matching, deep learning recognition, and context understanding. A standard symbol library covering multiple professions such as architecture, structure, and hydropower is established, and various engineering symbols specified by national standards are included. A basic symbol set containing symbol graphics, meanings, and usage scenario information is established, and for special symbols in different professional fields, including water supply and drainage system symbols, HVAC symbols, and electrical symbols, classification and feature extraction are carried out to establish a professional symbol set. For the intelligent recognition mechanism, a multi-model fusion recognition strategy is adopted. For highly standardized symbols, an improved template matching algorithm is used for recognition by combining feature point matching and shape matching. For non-standard symbols, a deep learning model is used for recognition, and the generalization ability of the model is improved through transfer learning. Combining the text description and graphic features around the symbol, a semantic understanding model of the symbol is established to improve the accuracy of recognition;

[0122] In the intelligent parsing layer, the process of semantic analysis, element recognition, and association analysis of the extracted features is as follows:

[0123] Operations of engineering semantic understanding and intelligent rule parsing are carried out in the intelligent parsing layer. Among them, engineering semantic understanding includes the construction of drawing semantic networks and engineering knowledge graphs, and intelligent rule parsing includes design specification parsing and parameter intelligent extraction. For the construction of drawing semantic networks, through the establishment of multi-level semantic networks, in-depth understanding of engineering drawings is achieved. Among them, the multi-level semantic networks include: the basic semantic layer, the relational semantic layer, and the knowledge semantic layer. In the basic semantic layer, basic elements in the drawing are captured, including components, dimensions, and annotation information, and preliminary associations between elements are established. An improved semantic segmentation technology is adopted to classify the drawing content according to functional attributes, realizing the precise positioning and attribute extraction of elements. In the relational semantic layer, the logical relationships between drawing elements are analyzed and established, including spatial relationships, functional relationships, and structural relationships. By establishing a relational matrix, complex engineering information such as the connection method between components and the force transmission path is described. In the knowledge semantic layer, engineering domain knowledge is integrated, and drawing information is associated with professional specifications and design standards to achieve semantic understanding. A knowledge base containing national standards and industry specifications is built-in to support intelligent specification inspection. Based on the semantic network, an engineering knowledge graph is constructed, including ontology framework design and instance relationship mapping. For ontology framework design, a professional ontology system covering the engineering domain is established, including multiple dimensions such as component types, material properties, and construction techniques;

[0124] For each ontology category, a standard set of attributes and relationships is defined. For instance relationship mapping, the specific instances identified in the drawing are mapped to the ontology framework to establish the association relationships between instances. Through the relationship reasoning mechanism, implicit engineering logical relationships are discovered. For design specification parsing, various specification standards in the field of engineering construction are structured, and a specification knowledge base containing clause requirements, parameter limits, and calculation formulas is established. Through natural language processing technology, the specification clauses are transformed into a set of computable rules, understanding the logical relationships in the clauses and transforming them into specific judgment criteria. For parameter intelligent extraction, various engineering parameters are extracted from the drawing, including geometric parameter extraction and performance parameter extraction. Through geometric parameter extraction, the dimension information of components is identified, including basic parameters such as length, width, and height. Through performance parameter extraction, professional information such as material strength grades, component reinforcement information, and equipment performance parameters is identified, and automatic comparison with relevant specification requirements is carried out;

[0125] In the evaluation and decision-making layer, the process of comprehensive evaluation of the green building level is as follows:

[0126] Operations for constructing an evaluation index system and analyzing evaluation results are carried out in the evaluation and decision-making layer. Among them, the evaluation index system includes a basic index layer and a professional index layer, and the evaluation result analysis includes compliance analysis, integrity analysis, and consistency analysis. A comprehensive drawing quality evaluation index system is established. In the basic index layer, the drawing standardization, graphic expression accuracy, and design integrity are analyzed. The integrity of the basic elements of the drawing, the rationality of layer settings, and the annotation standardization are evaluated. The accuracy of component expression, the clarity of detail drawing expression, and the rationality of section expression are evaluated. Whether the design depth meets the requirements and whether the design elements are complete are evaluated. In the professional index layer, the rationality of structural design, the adaptability of building functions, and the coordination of equipment systems are analyzed. The rationality of structural layout, component dimensions, joint practices, etc. is evaluated. The compliance of space layout, fire protection requirements, barrier-free design, etc. is evaluated. The appropriateness of the layout of various pipelines, equipment selection, etc. is evaluated. In the evaluation result analysis, for the compliance analysis, it is automatically checked against the design specification requirements to identify the design content that does not meet the specification requirements, and a specification compliance score and a detailed analysis report are generated. For the integrity analysis, the integrity of the design documents is checked, whether the design depth meets the requirements is evaluated, and the missing items of key design information are identified. For the consistency analysis, the coordination between the drawings of different specialties is checked, the design contradictions and inconsistent information are marked, and a cross-professional coordination evaluation report is generated;

[0127] In the application and display layer, the process of generating a standardized evaluation report is as follows:

[0128] The generation of the evaluation report includes a comprehensive evaluation report, a special inspection report, and a dynamic tracking report. For the comprehensive evaluation report, the overall score and the scores of each sub-item, the list of main problems and rectification suggestions, and the chart visualization display of the evaluation results are analyzed. For the special inspection report, the special inspection results for specific design requirements, problem location and specific description, as well as improvement suggestions and reference solutions are analyzed. For the dynamic tracking report, the problem-solving situation during the design modification process, the dynamic change trend of evaluation indicators, and the analysis of continuous improvement effects are recorded.

[0129] As mentioned above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claimed rights.

Claims

1. An intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition, including a data processing center, is characterized by: The data processing center is communicatively connected with a data acquisition layer, an image preprocessing layer, a feature extraction layer, an intelligent analysis layer, an evaluation and decision layer, and an application display layer, wherein electrical signals are connected between the architectures of each layer; The data collection layer adopts a distributed collection architecture to perform parallel collection and processing of multiple types of substation engineering construction drawings to obtain the original data of the substation engineering construction drawings and establish a distributed drawing index database; The image preprocessing layer is used to perform standardization processing and quality optimization on the collected original drawing data, and convert drawings of different formats into a standardized image format; The feature extraction layer adopts a multimodal feature extraction strategy, integrating text recognition, graphic feature extraction and symbol recognition to extract various feature information from the preprocessed drawings; The intelligent parsing layer performs semantic analysis, element recognition and association analysis on the extracted features based on knowledge graph and deep learning technology; The evaluation decision-making layer constructs a comprehensive drawing quality evaluation index system, uses fuzzy comprehensive evaluation method and hierarchical analysis method to determine weights, and conducts a comprehensive evaluation of the green construction level based on the analysis results; The application display layer adopts the Web technology architecture, supports multi-terminal access, provides an operation interface and automatically generates standardized evaluation reports.

2. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 1 is characterized by: In the data collection layer, the process of collecting original data and establishing a distributed drawing index database is as follows: The system receives substation engineering construction drawing data from different sources in the form of CAD format, PDF document or multiple image formats; Preprocess the received drawings, including denoising and format conversion, and use the distributed acquisition architecture to parallel process the preprocessed drawings, extract key information from the drawings, and generate a drawing summary; For each piece of construction drawing data collected, encode it according to its attributes, establish a three-level encoding system based on "project number-drawing type-version number", and formulate unified encoding rules, in which the project number is used to uniquely identify a substation project, classify the drawings according to their content and purpose, determine the drawing type, including electrical drawings, structural drawings and civil engineering drawings, and record the version information of the drawings, including the initial version and subsequent revised versions; The encoded drawing data is stored in a distributed drawing index database, and the metadata information of the drawings is stored and managed in the database, including the drawing name, creation time, revision record, design unit, construction unit, drawing size and scale information, and the drawing data in the database is backed up regularly.

3. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 2 is characterized by: The image preprocessing layer includes a format conversion module, an image enhancement module and a layout analysis module, wherein the modules are connected by electrical signals; The format conversion module uses parallel processing technology to uniformly convert drawings of different formats into a standardized image format; The image enhancement module uses a variety of image processing algorithms to optimize the drawings, including denoising, enhancement and correction processing; The layout analysis module realizes intelligent partitioning and structured analysis of drawings based on deep learning layout analysis technology.

4. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 3 is characterized by: In the image preprocessing layer, the image format conversion process is as follows: The format conversion module receives the original drawing data from the data acquisition layer, identifies the format of the input drawing data, determines its original format, and uses parallel processing technology to simultaneously convert drawing data of different formats into a standardized image format; During the format conversion process, the resolution and image quality parameters are set to uniformly convert the drawings to a standard resolution of 300 DPI, and the converted standardized image format data is transmitted to the image enhancement module via electrical signals; The image enhancement module receives the standardized image format data from the format conversion module, and uses the Gaussian filter denoising algorithm to denoise the image. The adaptive histogram equalization method is used to select the enhancement strategy and parameters according to the histogram distribution of the image to make the gray value of the image evenly distributed. The image is sharpened by combining the Laplacian operator and nonlinear enhancement to highlight the edge and detail information of the image. According to the edge features and sharpening requirements of the image, the algorithm parameters are adjusted to make the lines and contours of the image clear. The image is corrected by calculating the correction parameters according to the feature points or reference objects of the image, and the image is geometrically transformed accordingly to restore the image to the correct geometric form. The optimized image data is then transmitted to the layout analysis module through electrical signals. The layout analysis module receives high-quality image data from the image enhancement module, pre-trains the deep learning model to obtain the improved Faster R-CNN model, and uses the improved Faster R-CNN model to perform drawing layout analysis. Through the model's feature extraction and classification capabilities, the design area classification criteria establish a complete drawing area classification system to identify different areas in the drawing, including the title bar, design description area, main engineering drawing, node detail drawing, and equipment layout drawing; According to the results of layout analysis and recognition, different areas of the drawing are located and segmented. For each area, its position and range in the image are determined, and it is segmented from the overall image. Layout information is extracted from each segmented area, including project information in the title bar, text content in the design description area, and graphic symbols in the main engineering drawing. Use OCR technology to recognize text information, use image recognition and pattern matching technology to recognize graphic symbols and annotation information, extract equipment layout and connection relationship information through geometric analysis and topological relationships, and structure the extracted layout information to establish a structured representation of the drawing content, organize text information according to paragraph and sentence structure, associate graphic symbols and annotation information with corresponding equipment, and form a structured data format.

5. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 4 is characterized in that: For the conversion of CAD drawings, the improved LibreCAD open source framework is used, and the complete details of the drawings are retained through the optimized vector-to-raster conversion algorithm. During the conversion process, the adaptive resolution control technology is used to intelligently scale drawings of different scales. For PDF parsing, the improved PDF2 Image algorithm is used to improve the conversion efficiency through multi-threaded parallel processing technology. During the processing, an adaptive page segmentation algorithm is introduced to identify and extract the graphic content in PDF and control the parsing accuracy of the drawings. For the process of using adaptive histogram equalization method to make the grayscale value of the image evenly distributed, the improved CLAHE algorithm is used to improve the local contrast of the image through dynamic block size adjustment and contrast limitation. The core of the algorithm includes image segmentation, histogram calculation and bilinear interpolation. Among them, the image segmentation adopts an adaptive segmentation strategy, the block size is dynamically adjusted according to the complexity of the image content, the histogram is calculated, the histogram statistics and equalization processing are performed in each block, and the bilinear interpolation is used to perform smooth transition between blocks.

6. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 5 is characterized by: In the feature extraction layer, the feature extraction process is as follows: In the feature extraction layer, text recognition, graphic feature extraction and symbol recognition are performed. For text recognition, it includes multi-scenario text recognition strategy and professional terminology recognition optimization. For graphic feature extraction, it includes geometric feature analysis and topological relationship analysis. For symbol recognition and classification, it includes standard symbol library construction and intelligent recognition mechanism. For the multi-scenario text recognition strategy, a hierarchical recognition strategy is adopted according to the text characteristics in different scenarios in engineering drawings, and the text areas in the drawings are classified into different types, including title text, description text, dimension text and legend text. Each type of text has its own specific font size, orientation and density characteristics, and then a feature model library is established to locate the text area and recognize the text content. The corresponding recognition model is selected according to different scenarios, wherein a standard OCR model is used for horizontal text, and a recognition model with direction perception is used for inclined text; For the optimization of professional terminology recognition, a domain dictionary containing professional terms is constructed based on the characteristics of professional terminology in the field of construction engineering. During the recognition process, the recognition accuracy of professional terminology is adjusted by combining contextual semantic analysis and dictionary matching.

7. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 6 is characterized by: For the geometric feature analysis, a multi-level geometric feature extraction scheme is adopted to gradually build a feature system from the three levels of point, line and surface. For the topological relationship analysis, a complete graphic element topological relationship analysis system is established, including: spatial relationship analysis, connection relationship analysis and inclusion relationship analysis; For point feature extraction, the improved Harris corner detection algorithm is used to extract key nodes in the drawings. The local feature description operator is introduced to describe the features of the surrounding area of ​​each detection point. For line feature extraction, multi-scale Hough transform combined with LSD algorithm is used to extract straight line segments. By setting dynamic parameter thresholds, it adapts to line features of different thicknesses and lengths. For surface feature extraction, based on contour tracking and region growing algorithms, closed area features are extracted, regular geometric shapes and irregular areas are identified, and the characteristic parameters of their area and perimeter are calculated; For spatial relationship analysis, the spatial relationship model is established by calculating the relative position, distance and angle parameters between graphic elements. For connection relationship analysis, the connection methods between graphic elements are identified, including welding, bolt connection and hinged connection. The connection type is determined by feature template matching. For inclusion relationship analysis, the hierarchical structure of graphic elements is established, the subordinate relationship between components is identified, and the combination analysis of complex components is supported. The standard symbol library includes a basic symbol set and a professional symbol set, and the intelligent recognition mechanism includes template matching, deep learning recognition and context understanding; Establish a standard symbol library covering multiple professions such as architecture, structure, and water and electricity, establish a basic symbol set containing symbol graphics, meanings, and usage scenario information, and classify and extract features of special symbols in different professional fields, including water supply and drainage system symbols, HVAC symbols, and electrical symbols, to establish a professional symbol set; A multi-model fusion recognition strategy is adopted for the intelligent recognition mechanism. For symbols with a high degree of standardization, an improved template matching algorithm is used to identify them by combining feature point matching and shape matching. For non-standard symbols, a deep learning model is used for identification, and a semantic understanding model of the symbol is established in combination with the text description and graphic features around the symbol.

8. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 7 is characterized by: In the intelligent parsing layer, the process of performing semantic analysis, element recognition and association analysis on the extracted features is as follows: In the intelligent analysis layer, engineering semantic understanding and intelligent rule analysis are performed. Engineering semantic understanding includes the construction of drawing semantic network and engineering knowledge graph, and intelligent rule analysis includes design specification analysis and parameter intelligent extraction. For the construction of the drawing semantic network, a multi-level semantic network is established to achieve a deep understanding of engineering drawings. The multi-level semantic network includes: basic semantic layer, relationship semantic layer and knowledge semantic layer; At the basic semantic layer, the basic elements in the drawings are captured, including components, dimensions and annotation information, and preliminary associations between elements are established. The improved semantic segmentation technology is used to classify the drawing content according to functional attributes to achieve element positioning and attribute extraction. At the relational semantic layer, the logical relationships between drawing elements are analyzed and established, including spatial relationships, functional relationships and structural relationships. By establishing a relationship matrix, the complex engineering information of the connection mode and force transmission path between components is described. At the knowledge semantic layer, engineering domain knowledge is integrated, and the drawing information is associated with professional specifications and design standards to achieve semantic understanding. A knowledge base including national standards and industry specifications is built in. Based on the semantic network, we build an engineering knowledge graph, including ontology framework design and instance relationship mapping; A professional ontology system covering the engineering field is established for the ontology framework design, including multiple dimensions such as component type, material properties and construction technology. Each ontology category defines a standard attribute set and relationship set. For instance relationship mapping, the specific instances identified in the drawings are mapped to the ontology framework, and the association relationship between instances is established. Through the relational reasoning mechanism, the implicit engineering logic relationship is discovered. For the analysis of design specifications, various specifications and standards in the field of engineering construction are structured, and a specification knowledge base including clause requirements, parameter limits and calculation formulas is established. Through natural language processing technology, the specification clauses are converted into a computable rule set, the logical relationship in the clauses is understood, and it is converted into specific judgment criteria; For intelligent parameter extraction, various engineering parameters are extracted from the drawings, including geometric parameter extraction and performance parameter extraction. Geometric parameter extraction is used to identify the size information of the components, including the basic parameters of length, width and height. Performance parameter extraction is used to identify professional information such as material strength grade, component reinforcement information and equipment performance parameters, and automatically compare them with relevant specifications.

9. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 8 is characterized by: In the evaluation decision-making layer, the process of comprehensive evaluation of green construction level is as follows: In the evaluation decision-making layer, the evaluation index system is constructed and the evaluation results are analyzed. The evaluation index system includes the basic index layer and the professional index layer, and the evaluation results analysis includes compliance analysis, integrity analysis and consistency analysis. Establish a comprehensive drawing quality evaluation index system, analyze the standardization of drawings, accuracy of graphic expression and design integrity at the basic index level, evaluate the integrity of basic elements of drawings, rationality of layer settings and standardization of annotations, evaluate the accuracy of component expression, clarity of detailed drawing expression and rationality of section expression, and evaluate whether the design depth meets the requirements and whether the design elements are complete; Analyze the rationality of structural design, adaptability of building functions and coordination of equipment systems at the professional indicator level, evaluate the rationality of structural layout, component size and node practices, evaluate the compliance of spatial layout, fire protection requirements and barrier-free design, and evaluate the appropriateness of various pipeline layouts and equipment selection; In the analysis of evaluation results, compliance analysis automatically checks against design specifications, identifies design content that does not meet specification requirements, generates specification compliance scores and detailed analysis reports. For integrity analysis, the integrity of design documents is checked, whether the design depth meets requirements, and missing items of key design information are identified. For consistency analysis, the coordination between professional drawings is checked, design contradictions and incoherent information are marked, and a cross-professional coordination evaluation report is generated.

10. The intelligent analysis and evaluation system for substation engineering construction drawings based on image recognition according to claim 9 is characterized in that: In the application presentation layer, the process of generating the standardized evaluation report is as follows: The evaluation report generation includes a comprehensive evaluation report, a special inspection report and a dynamic tracking report; For the comprehensive evaluation report, analyze the overall score and sub-item scores, the list of major issues and rectification suggestions, and the graphical visualization of the evaluation results; For special inspection reports, analyze the special inspection results, problem location and specific descriptions, as well as improvement suggestions and reference plans for specific design requirements; For dynamic tracking reports, record the problem solving status during the design modification process, the dynamic change trend of evaluation indicators, and the analysis of continuous improvement effects.

Images