Electromechanical installation quality detection method based on image recognition

Abstract

The application relates to the technical field of building electromechanical installation quality detection, and discloses an electromechanical installation quality detection method based on image recognition, which comprises the following steps: acquiring a field image and design data of a region to be detected; performing quality checking and scale calibration on the field image to obtain a calibrated image; performing component recognition on the calibrated image to obtain a component result; matching the component result with the design data to obtain a design matching result; constructing installation relations and shielding relations of the design matching result to obtain a relation graph; presetting a rule table, performing rule detection on the relation graph and the rule table to obtain a rule violation result; performing shielding credibility determination on the relation graph to obtain a shielding result; and performing evidence integrity determination on the relation graph. The relation graph and the rule detection technical scheme achieve the technical effect of regularizing the component relations, and realize the conversion of installation quality from appearance recognition to rule compliance judgment.

Description

Technical Field

[0001] This invention relates to the field of building electromechanical installation quality inspection technology, specifically a method for electromechanical installation quality inspection based on image recognition. Background Technology

[0002] In scenarios involving the quality inspection of electromechanical pipelines, cable trays, air ducts, and supports in high-density computer rooms, underground pipe corridors, and integrated ceilings, air ducts, water pipes, cable trays, sprinkler pipes, supports, seismic braces, sleeves, and fireproof sealing components are installed crosswise in confined spaces. On-site images are affected by overlapping pipelines, bracket obstruction, changes in shooting angles, and design changes, making it difficult to reliably determine whether the image recognition results are consistent with the design position, installation sequence, support spacing, pipeline clearance, slope requirements, and concealed acceptance requirements.

[0003] Currently, existing technologies typically collect construction site images or point cloud data using mobile terminals, fixed cameras, inspection robots, or laser scanning equipment, and then compare the collected results with... The model, construction drawings, or preset quality rules are compared, and the air ducts, cable trays, pipes and supports are identified through target detection, semantic segmentation and instance segmentation. At the same time, the identification box, pixel position or spatial projection relationship is used to determine whether the components are missing, whether the installation position is deviated, and whether the appearance is abnormal.

[0004] In existing technologies, image recognition results mainly focus on judging the presence of components, the approximate correctness of component positions, and the obvious abnormalities in component appearance. They fail to transform component recognition results into spatial relationships and rule-based evidence that can be used in installation specifications. This makes it difficult to reliably detect issues such as whether the spacing between supports and hangers meets requirements, whether the clearance between cable trays and water pipes meets requirements, whether the duct elevation affects maintenance space, whether the pipe slope is reversed, whether seismic supports and hangers are arranged in designated sections, and whether fireproof sealing corresponds to wall sleeves. Therefore, this invention proposes an image recognition-based electromechanical installation quality inspection method to solve the aforementioned problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an image recognition-based method for inspecting the quality of electromechanical installations, thereby at least partially solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for inspecting the quality of electromechanical installation based on image recognition, comprising:

[0007] Acquire on-site images and design data of the area to be inspected;

[0008] The on-site images are subjected to quality verification and scale calibration to obtain calibrated images;

[0009] Component identification is performed on the calibration image to obtain component results;

[0010] The component results and design data are matched to obtain the design matching results;

[0011] The design matching results are used to construct installation and occlusion relationships to obtain a relationship graph;

[0012] A preset rule table is used to perform rule detection on the relationship graph and the rule table to obtain violation results;

[0013] The occlusion confidence level of the relationship graph is determined to obtain the occlusion result;

[0014] The relationship graph is subjected to an evidence integrity assessment to obtain the evidence results;

[0015] The violation results, occlusion results, and evidence results are fused together to obtain the quality inspection results.

[0016] Preferably, the on-site images are subjected to quality verification and scale calibration to obtain calibrated images, including:

[0017] The image quality parameters are obtained by extracting the sharpness, exposure status, and tilt status of the scene image;

[0018] The image quality parameters are compared with preset quality conditions to obtain a qualified image;

[0019] The design data is processed to extract component specifications, thus obtaining the design specifications.

[0020] The component pixel lengths are extracted from the qualified images to obtain the image specifications;

[0021] The scale factor is calculated based on the design specifications and image specifications to obtain the scale factor.

[0022] Based on the scale coefficient, the qualified image is scaled to obtain a calibrated image.

[0023] Preferably, the component results and design data are matched to obtain design matching results, including:

[0024] The component results are processed by extracting category, location, orientation, and specifications to obtain on-site component information;

[0025] The design data is processed by extracting its category, location, orientation, and specifications to obtain the design component information;

[0026] Candidate matching is performed on the on-site component information and the design component information to obtain candidate matching results;

[0027] The candidate matching results are compared for deviation to obtain the matching determination result;

[0028] Based on the matching determination result, the component result is marked accordingly to obtain the design matching result.

[0029] Preferably, the design matching results are used to construct installation relationships and occlusion relationships to obtain a relationship graph, including:

[0030] The design matching results are processed to extract the corresponding state of the components, and the corresponding state results are obtained.

[0031] Based on the corresponding state results, the components with installation and mating relationships are associated and determined to obtain the installation relationship;

[0032] Based on the corresponding state results, occlusion determination is performed on components with occlusion influence relationships to obtain the occlusion relationships;

[0033] The installation and occlusion relationships are correlated and integrated to obtain a relationship graph.

[0034] Preferably, a preset rule table is used to perform rule detection on the relationship graph and the rule table to obtain violation results, including:

[0035] The installation and testing requirements are extracted from the design data to obtain rule parameters;

[0036] The rule parameters are categorized and organized to obtain a rule table containing detection objects, detection relationships, and judgment conditions;

[0037] The installation relationships in the relationship graph are extracted to obtain the relationships to be detected;

[0038] The relationship to be detected is matched with the rule table to obtain the rule matching result;

[0039] Based on the rule matching results, the relationship to be detected is judged for compliance, and a violation result is obtained.

[0040] Preferably, the occlusion confidence level of the relationship graph is determined to obtain the occlusion result, including:

[0041] The occlusion relationships in the relationship graph are extracted to obtain the occlusion relationship results;

[0042] Based on the occlusion relationship results, the components in the relationship graph are distinguished by occlusion source and occlusion state to obtain occlusion state results;

[0043] By mapping the occlusion state results to the component results, the occlusion range results are obtained;

[0044] Based on the occlusion range result, the corresponding state of the component in the design matching result is verified to obtain the occlusion confidence result;

[0045] The occlusion confidence results are integrated to obtain the occlusion result.

[0046] Preferably, the relationship graph is subjected to an evidence integrity determination to obtain evidence results, including:

[0047] The installation and occlusion relationships in the relationship graph are extracted to obtain the evidence relationship results;

[0048] By mapping the evidence relationship results and component results, the component evidence correspondence results are obtained;

[0049] The component evidence correspondence results and design matching results are compared to obtain the evidence comparison results;

[0050] Based on the evidence comparison results, the integrity of the corresponding states of the components in the relationship graph is determined, and the integrity determination results are obtained.

[0051] The integrity determination results are integrated to obtain the evidence results.

[0052] Preferably, the scale factor is calculated based on the design specifications and image specifications to obtain the scale factor, including:

[0053] The design specifications and image specifications are matched to obtain the specification correspondence results;

[0054] The consistency verification of the specification correspondence results is performed to obtain the valid specification correspondence results;

[0055] The ratio of the design specifications and image specifications in the effective specification correspondence results is calculated to obtain the individual scale value;

[0056] The individual scale values ​​are subjected to deviation screening to obtain effective scale values;

[0057] The effective scale values ​​are fused and calculated to obtain the scale coefficients.

[0058] Preferably, the component results are processed by extracting category, location, orientation, and specifications to obtain on-site component information, including:

[0059] The component results are divided into single components to obtain single component results;

[0060] The category of the single component is obtained by reading the category of the single component result;

[0061] Based on the calibration image, the position of the single component is calculated to obtain the component position;

[0062] The extension direction of the single component is determined to obtain the component direction;

[0063] Based on the component location and calibration image, the specifications of the single component are converted to obtain the component specifications;

[0064] The component category, component location, component orientation, and component specifications are integrated to obtain on-site component information.

[0065] Preferably, based on the rule matching result, a compliance determination is made on the relationship to be detected to obtain a violation result, including:

[0066] The detection object, detection relationship, and judgment condition are extracted from the rule matching results to obtain rule judgment information;

[0067] The detection object, detection relationship, and relationship parameters are extracted from the relationship to be detected to obtain the information to be determined;

[0068] The consistency of the information to be determined and the rule-based determination information are verified to obtain valid determination information.

[0069] Based on the judgment conditions in the rule judgment information, the relationship parameters in the valid judgment information are compared for compliance to obtain the compliance comparison result.

[0070] The relationships to be detected that do not meet the judgment criteria in the compliance comparison results are marked to obtain individual violation results;

[0071] The individual violation results are categorized and integrated to obtain the violation results.

[0072] This invention provides a method for inspecting the quality of electromechanical installation based on image recognition. It has the following beneficial effects:

[0073] 1. This invention adopts a relationship graph and rule detection technology solution to achieve the effect of rule-based judgment of component relationships, and realizes the transformation of installation quality from appearance recognition to rule compliance judgment.

[0074] 2. This invention adopts a technical solution for determining the credibility of occlusion and the integrity of evidence, achieving the technical effect of joint judgment of occlusion status and acceptance evidence, and realizing the credibility quality assessment of the occluded area. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0076] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0077] The present invention will now be described in detail with reference to the accompanying drawings:

[0078] Please see the appendix Figure 1 This embodiment provides an image recognition-based method for inspecting the quality of electromechanical installations. This method is used to inspect the quality of electromechanical installations within an area to be inspected, which can be a high-density computer room, underground pipe gallery, integrated ceiling, equipment mezzanine, or an area with concentrated electromechanical pipelines. On-site images reflect the actual installation status within the area to be inspected, while design data reflects the designed installation status and installation inspection requirements. The quality inspection result is obtained by performing quality verification, scale calibration, component identification, design matching, relationship graph construction, rule detection, occlusion credibility determination, evidence integrity determination, and quality result fusion on the on-site images.

[0079] In this embodiment, the on-site images are acquired by mobile terminals, handheld shooting devices, fixed cameras, inspection robots, or construction site image acquisition equipment. Design data originates from... The data includes models, 2D construction drawings, detailed design drawings, component specification sheets, installation and testing requirement sheets, construction change records, and project quality acceptance documents. After data parsing, the design data is transformed into design component data and testing requirement data that can be calculated and processed. Design component data must at least include component category, component design location, component orientation, and component specifications; testing requirement data must at least include the testing object, testing relationship, and judgment criteria.

[0080] In this embodiment, components include air ducts, water pipes, cable trays, sprinkler pipes, supports, seismic braces, sleeves, fireproof sealing components, valves, flanges, and insulation layers. Component results include component category, component outline, component location, component orientation, component specifications, identification confidence level, and visible area of ​​the component. Relationship diagrams are used to express the installation relationships and occlusion relationships between components, and quality inspection results are used to express the installation quality status within the inspected area.

[0081] When acquiring on-site images and design data for the area to be inspected, first determine the area identifier. The area identifier consists of floor, axis, room, pipe gallery section, equipment room number, or construction sequence section. Bind the area identifier to the on-site images so that design data can be retrieved according to the same area.

[0082] For on-site images, the image acquisition location, shooting orientation, acquisition time, and the identification of the area to be inspected are recorded during acquisition. If multiple on-site images exist for the same area to be inspected, they are grouped into the same image group. The on-site images in the image group were acquired from different directions, heights, or distances to reduce the impact of occlusion on the detection results.

[0083] For design data, if it comes from Model, Read The model includes component numbers, component categories, component coordinates, component orientation, component specifications, and component association attributes. If the data originates from two-dimensional construction drawings, the axis positions, elevations, component numbers, pipe diameters, cable tray widths, duct dimensions, and support and hanger layout requirements are extracted through drawing analysis and converted into tabular design data. If the data originates from construction change records, the original design data is updated with the changed component positions, component specifications, and installation requirements.

[0084] After parsing, the design data forms the following structure: ,

[0085] in, Represents the design data set. Indicates the first Component design Indicates the category of design components. Indicates the location of the designed components. Indicates the direction of the designed components. Indicates the specifications of the designed components. This indicates the installation and testing requirements related to the design component.

[0086] This ensures that on-site images and design data correspond within the same inspection area, avoiding inconsistencies between the images and the design source.

[0087] The purpose of quality verification and scale calibration of on-site images is to ensure a reliable basis for component identification and size conversion. This processing includes image quality parameter extraction, quality condition comparison, design specification extraction, image specification extraction, scale coefficient calculation, and scale calibration.

[0088] Image quality parameters are extracted by analyzing the sharpness, exposure, and tilt of the on-site image.

[0089] Sharpness is used to characterize whether the boundaries of components in an image are clear. It is obtained jointly through image gradient, edge response, and contour stability. (For on-site images) Gradient calculations are performed on the pixels to obtain the gradient response; the second-order response of the image is then calculated to determine whether the image is blurry. The sharpness parameter can be expressed as:

[0090] ,

[0091] in, Indicates the sharpness parameter, Indicates the image region involved in the calculation. Represents pixels in an image region. Represents pixels gradient at, Represents the second-order response of the image. This represents the second-order response variance. This represents the evaluation value of the continuity of the component boundary. , , This represents the weighting coefficient.

[0092] Exposure status is used to characterize whether there is darkness or brightness in a scene image. It is calculated based on the gray-level mean, the degree of gray-level distribution concentration, and the proportion of saturated pixels.

[0093] ,

[0094] in, Indicates the exposure status parameters, This represents the average grayscale value of the image. Indicates the target grayscale. Indicates the allowable grayscale range. Indicates the percentage of pixels that are exposed or underexposed. Indicates the uniformity of grayscale distribution. Indicates a correction term. , , This represents the weighting coefficient.

[0095] The tilt state is used to characterize the deviation between the component orientation and the design orientation or the main orientation of the image. Orientation is extracted from linear structures in the field image to obtain the main orientation, and the difference between this and the corresponding orientation in the design data is calculated. ,

[0096] in, Indicates the tilt state parameter. This indicates the main direction extracted from the scene image. Indicates the corresponding direction in the design data. This indicates the directional deviation control parameter. Indicates a correction term.

[0097] By fusing sharpness, exposure status, and tilt status, image quality parameters are obtained:

[0098] ,

[0099] in, Indicates image quality parameters, , , This represents the weighting coefficient.

[0100] The quality condition comparison compares the image quality parameters with preset quality conditions to obtain qualified images. These preset quality conditions include sharpness, exposure, and tilt conditions. If the on-site image simultaneously meets all of these quality conditions, it is considered a qualified image. If the on-site image does not meet the quality conditions required for scale calibration but some components can still be identified, it can be used as an auxiliary image for occlusion or evidence assessment. If the on-site image does not meet the basic recognition conditions, it will not participate in subsequent detection.

[0101] Design specifications are extracted from the design data. These specifications are derived from the pipe diameter, cable tray width, duct side length, support crossarm length, sleeve size, or other component dimensions that can be used for dimensional conversion within the design data. The design specifications are stored indexed by component category and component number.

[0102] Image specification extraction involves extracting the pixel lengths of components from qualified images to obtain image specifications. Image specifications are obtained through the pixel distances between component edge points, contour points, centerline endpoints, or feature corner points. For linear components, the pixel distance between centerline endpoints is extracted; for planar components, the width or height of the contour boundary is extracted; for support / hanger components, the pixel lengths of horizontal or vertical components are extracted.

[0103] Scale factor calculation is performed based on design specifications and image specifications. First, the design specifications and image specifications are matched to obtain the specification correspondence results. During matching, constraints such as consistent component category, close component location, and consistent component orientation are used to ensure that the design specifications and image specifications of the same component correspond.

[0104] A consistency check is performed on the specification mapping results to obtain valid specification mapping results. The consistency check is used to remove specification mapping results with severe occlusion, incomplete boundaries, or incorrect matches. For the first... Based on the results of the group specifications, calculate the consistency evaluation value: ,

[0105] in, This represents the consistency evaluation value. Indicates category consistency. Indicates positional consistency. Indicates consistency in direction. Indicates visibility completeness. , , , This represents the weighting coefficient. Specification correspondence results that meet the preset consistency conditions are determined as valid specification correspondence results.

[0106] The ratio of the design specifications to the image specifications in the valid specification correspondence results is calculated to obtain the individual scale value: ,

[0107] in, Indicates the first Individual scale value, Indicates the first Component design specifications Indicates the first Image specifications of components, Indicates a correction term.

[0108] The individual scale values ​​are filtered for bias to obtain the effective scale values. The bias filtering uses a weighted robust filtering method: , ,

[0109] in, This represents the initial scale mean. Indicates the first Individual scale value, Indicates the first Reliable weights of individual scale values, Indicates the first The degree of deviation of individual scale values, This indicates a correction item. Individual scale values ​​that do not meet the preset screening criteria are removed to obtain valid scale values.

[0110] The effective scale values ​​are fused and calculated to obtain the scale coefficients: ,

[0111] in, Represents the scaling factor. Indicates the number of valid scale values. Represents the robust loss function. Indicates the first The permissible range of fluctuations in the effective scale value. This indicates a correction term. This method reduces the impact of edge extraction errors on the scale factor.

[0112] Scale calibration involves scaling the qualified image based on a scale factor to obtain a calibrated image. The calibrated image retains the original image content and includes the conversion relationship between pixel coordinates and actual scale coordinates. If the pixel distance between two points in the qualified image is... The corresponding actual distance is: ,

[0113] in, Indicates the actual distance. Represents the scaling factor. This represents pixel distance. Through this processing, component location, component specifications, and relational parameters can all be converted from pixel results to engineering-scale results.

[0114] When identifying components from calibration images, the images are first preprocessed to enhance component edges, reduce noise interference, and highlight component outlines. Then, an image recognition model is used to identify the components in the calibration images, yielding the component results.

[0115] Component recognition includes candidate region generation, category determination, contour extraction, centerline extraction, and confidence score calculation. Candidate region generation is used to determine the location of possible components in the image; category determination is used to determine the component category to which the candidate region belongs; contour extraction is used to obtain the outer boundary of the component; centerline extraction is used to obtain the direction and orientation of long components such as pipes, cable trays, and air ducts; and confidence score calculation is used to evaluate the reliability of the recognition results.

[0116] The component result is represented as follows: ,

[0117] in, Represents the set of component results. Indicates the first Component results Indicates the component category, Represents the outline of the component. Represents the geometric information of the component. Indicates the position and orientation of the component. This indicates the confidence level of identification.

[0118] For elongated components, the component geometry information includes the centerline, endpoints, and width; for planar components, the component geometry information includes the outer contour, boundary direction, and area; for support and hanger components, the component geometry information includes the support points, connection direction, and positional relationship of adjacent supported components.

[0119] When extracting on-site component information from the component results, the component results are first divided into individual components to obtain individual component results. Individual component division is used to organize the multiple recognition regions output by the image recognition model into a group of independent components.

[0120] The category of a single component is determined by reading its category. The component category is derived from the category label output by the image recognition model and verified in conjunction with the component's outline shape. For example, pipes typically have linear or circular cross-sectional features, cable trays typically have a long rectangular structure, and supports and hangers typically have a vertical or oblique connection to the supported component.

[0121] Based on the calibration image, the position of a single component is calculated. If the pixel coordinates of the component's center point in the image are... The component position is obtained through the scaling factor and image coordinate transformation relationship: ,

[0122] in, Indicates the first Component location, Represents the scaling factor. This represents the conversion function from pixel coordinates to calibration coordinates.

[0123] The extension direction of a single component is determined to obtain its orientation. For elongated components, the orientation is obtained by fitting the centerline; for planar components, the orientation is obtained through the principal axis direction. The component orientation is represented as follows: ,

[0124] in, Indicates the direction of the component. and Indicates the endpoint of the component's centerline or main axis.

[0125] Based on the component location and calibration image, the specifications of a single component are converted to obtain the component specifications. The component specifications are obtained by multiplying the pixel specifications by a scale factor. ,

[0126] in, Indicate component specifications, Indicates the pixel specifications in the image. This represents the scaling factor.

[0127] The component category, location, orientation, and specifications are integrated to obtain on-site component information. This on-site component information is then used for subsequent matching with design data.

[0128] When matching component results with design data, the component results are first extracted by category, location, orientation, and specifications to obtain on-site component information; then the design data is extracted by category, location, orientation, and specifications to obtain design component information.

[0129] The information on the on-site components is represented as follows: ,

[0130] in, Indicates the first On-site component information, Indicates the component category, Indicates the location of the component. Indicates the direction of the component. Indicates component specifications.

[0131] The design component information is represented as follows: ,

[0132] in, Indicates the first Design component information, Indicates the component category, Indicates the location of the component. Indicates the direction of the component. Indicates component specifications.

[0133] Candidate matching is performed between on-site component information and design component information to obtain candidate matching results. Candidate matching is constrained by category, location, orientation, and specifications, and a matching score is calculated.

[0134] ,

[0135] in, Indicates on-site components With design components Match rating, Indicates category consistency. Indicates positional correspondence, Indicates consistency in direction. Indicates consistency of specifications. , , , This represents the weighting coefficient.

[0136] Positional correspondence is represented as: ,

[0137] Directional consistency can be expressed as: ,

[0138] Specification consistency can be expressed as: ,

[0139] in, This indicates the allowable positional deviation control amount. This indicates the allowable deviation control amount for direction.

[0140] The candidate matching results are compared for deviation to obtain the matching judgment result. The deviation comparison includes position deviation, direction deviation, and specification deviation. If the matching score meets the preset matching conditions, the on-site component and the design component are marked as matched; if the categories are the same but the position, direction, or specifications do not meet the preset deviation conditions, they are marked as suspected misalignment; if no corresponding design component is found for the on-site component, it is marked as suspected addition; if no corresponding on-site component is found for the design component, it is marked as suspected missing.

[0141] Based on the matching determination results, the component results are marked accordingly to obtain the design matching results. The design matching results record whether the components match, as well as the matching object, deviation type, and corresponding status, providing a data source for the construction of the relationship graph.

[0142] The design matching results are used to extract the corresponding states of the components, resulting in corresponding state results. These state results include matched, suspected misalignment, suspected addition, and suspected missing. The corresponding state results are used to determine whether a component can participate in the construction of installation relationships and occlusion relationships.

[0143] Based on the corresponding state results, components with installation coordination relationships are associated and determined to obtain the installation relationships. Installation relationships are used to describe the functional and spatial relationships between components in electromechanical installation. Installation relationships include support relationships, connection relationships, through relationships, parallel relationships, intersection relationships, clearance relationships, and upstream and downstream relationships.

[0144] The establishment of installation relationships is based on component category, component location, component orientation, component specifications, and design matching status. For example, when support and hanger components are spatially close to pipe components and the support and hanger are located in the direction of pipe support, a support relationship is established between them; when pipe components and sleeve components have overlapping crossing locations, a crossing relationship is established; when cable trays and pipes are arranged adjacently and the spacing needs to be determined, a clearance relationship is established; when the orientations of the two components are close and they are arranged adjacently, a parallel relationship is established; when the orientations of the two components intersect and their spatial projections intersect, an intersection relationship is established.

[0145] The installation relationship determination value is expressed as follows: ,

[0146] in, Representation of components and components Interval The value used to determine the installation relationship. Indicates the first Function for determining class installation relationship. , Indicates the component category, , Indicates the location of the component. , Indicates the direction of the component. , Indicates component specifications.

[0147] Based on the corresponding state results, occlusion is determined for components with occlusion influence relationships, thus obtaining the occlusion relationship. The occlusion relationship is jointly determined by component contour overlap, edge interruption, positional relationship, recognition confidence change, and design matching state.

[0148] The occlusion effect value is expressed as: ,

[0149] in, Representation of components and components The occlusion impact value, Representation of components The outline area, Representation of components The outline area, This represents the overlapping area of ​​the outline region. Representation of components The outline area, Indicates the degree of interruption at the component boundary. Representation of components The confidence level of identification, Indicates the support of the front and rear occlusion positions. , , , This represents the weighting coefficient.

[0150] By associating and integrating installation relationships and occlusion relationships, a relationship graph is obtained. The relationship graph can be represented as follows: ,

[0151] in, Representing a relational graph, Represents a set of component nodes. Represents a set of installation relationships. Represents the set of occlusion relationships. This represents a set of node and edge attributes. Node attributes include component category, location, orientation, specifications, and corresponding status; edge attributes include relationship type, relationship parameters, occlusion impact value, and relationship confidence.

[0152] When setting up the rule table, installation and testing requirements are extracted from the design data to obtain rule parameters. These rule parameters originate from the installation and testing requirements, quality acceptance requirements, component layout requirements, and project testing standards within the design data. The rule parameters are not directly used as testing conclusions but are instead organized into an executable rule table.

[0153] The rule parameters are categorized and organized to obtain a rule table containing detection objects, detection relationships, and judgment conditions. Each rule in the rule table includes a rule number, detection object, detection relationship, source of relationship parameters, and judgment condition. The detection object is used to limit the component categories to which the rule applies; the detection relationship is used to limit the relationship types to which the rule applies; and the judgment condition is used to determine whether the relationship to be detected is compliant.

[0154] The rule table is represented as follows: ,

[0155] in, Represents the rule table, Indicates the first Rule 1 Indicates the object to be detected. Indicates the detection relationship. This indicates the judgment condition.

[0156] The installation relationships in the relationship graph are extracted to obtain the relationships to be detected. These relationships originate from the installation relationship edges in the relationship graph and include the object to be detected, the relationship to be detected, and relationship parameters. For example, support spacing parameters can be extracted from support relationships; component spacing parameters can be extracted from clearance relationships; the corresponding crossing state can be extracted from crossing relationships; and the complete connection state can be extracted from connection relationships.

[0157] The relationships to be detected are matched against the rule table to obtain the rule matching results. During matching, first, it compares whether the objects to be detected in the relationships to be detected are consistent with the objects to be detected in the rule table; then it compares whether the relationships to be detected in the relationships to be detected are consistent with the relationships to be detected in the rule table; finally, it extracts the corresponding decision conditions. Upon successful matching, the rule matching results are generated.

[0158] Based on the rule matching results, a compliance determination is made regarding the relationship to be detected, resulting in a violation result. Specifically, the following processing is performed:

[0159] The detection object, detection relationship, and decision condition are extracted from the rule matching results to obtain rule decision information. This rule decision information clarifies which decision condition should be applied to the current relationship to be detected.

[0160] The detection object, detection relationship, and relationship parameters are extracted to obtain the information to be determined. The relationship parameters are distance, direction deviation, position deviation, specification difference, corresponding state, or connection state.

[0161] The consistency of the information to be judged and the rule judgment information is checked to obtain valid judgment information. If the detection object and the detection relationship are consistent, the relationship to be judged will enter the compliance comparison; if they are inconsistent, the corresponding rule will not be executed on the relationship to be judged.

[0162] Based on the judgment conditions in the rule judgment information, a compliance comparison is performed on the relational parameters in the valid judgment information to obtain the compliance comparison result. The compliance comparison is represented as: ,

[0163] in, Indicates the first The compliance comparison results of the rules This represents the rule-based decision function. This represents the relation parameters in the relation to be detected. This indicates the decision conditions in the rule decision information.

[0164] When the judgment criteria are within the allowable range, the compliance comparison is expressed as follows: ,

[0165] in, Indicates the first The permitted scope of the rule, This indicates compliance. This indicates non-compliance.

[0166] The non-compliant relationships in the compliance comparison results are marked to obtain individual violation results; these individual violation results are then categorized and integrated to obtain the violation results. The violation results include the violation object, violation relationship, violation type, violation source, and violation severity.

[0167] The degree of violation is expressed as follows: ,

[0168] in, Indicates the first The degree of violation of the results of the violation Indicates the actual relational parameters. Indicates the reference value for the rule. Indicates the normalization scale. This represents the importance coefficient of the rule. Indicates a correction term.

[0169] When determining the credibility of occlusion in a relationship graph, the occlusion relationships are first extracted to obtain the occlusion relationship results. The occlusion relationship results include the occlusion source component, the occluded component, the occlusion influence value, and the occlusion edge attributes.

[0170] Based on the occlusion relationship results, the components in the relationship graph are distinguished by occlusion source and occlusion state to obtain occlusion state results. The occlusion source is the component that causes the occlusion effect; the occlusion state is used to indicate whether the target component is partially visible, severely occluded, or in a state of insufficient evidence. This state is jointly determined by the occlusion effect value, the visible area of ​​the component, the identification confidence level, and the design matching state.

[0171] By mapping the occlusion state results to the component results, the occlusion range results are obtained. The occlusion range results indicate which part of the component experiences occlusion. The occlusion range can be determined by the overlapping area between the component outline and the outline of the component from which the occlusion originates, and can also be determined by comparing the missing area of ​​the component boundary with the component outline in the design matching results. The occlusion range scale is expressed as follows: ,

[0172] in, Indicates the first The visible proportion of the components, Indicates the first The area obstructed by the component, Indicates the first The overall outline area of ​​the component. Indicates a correction term.

[0173] Based on the occlusion range results, the corresponding states of components in the design matching results are verified to obtain the occlusion confidence score. The occlusion confidence score is used to determine whether the current state of a component is subject to uncertainty due to occlusion. The calculation is expressed as follows: ,

[0174] in, Indicates the first The credibility of component occlusion. Indicates the visible proportion. Indicates the confidence level of identification. Indicates the reliability of the design matching. Indicates the degree of impact from occlusion. , , , Indicates the weighting coefficient. Indicates a correction term.

[0175] If the occlusion confidence level is high, it means the current image can support the judgment of the component's state; if the occlusion confidence level is low, it means the recognition or matching result of the component may be affected by occlusion, and it is not appropriate to directly draw conclusions of missing or illegality. The occlusion confidence level results are integrated to obtain the occlusion results. The occlusion results include the occlusion object, occlusion source, occlusion range, occlusion confidence level, and occlusion state.

[0176] When determining the completeness of evidence in a relationship graph, the installation and occlusion relationships are first extracted to obtain the evidence relationship results. These results indicate which components and relationships in the current image can be used as a basis for quality judgment.

[0177] By mapping the evidence relationship results to the component results, the component evidence correspondence results are obtained. The component evidence correspondence results include whether the component is identified, whether the component outline is complete, whether the component participates in the installation relationship, whether the component has any occlusion effect, and whether the component forms a valid correspondence with the design data.

[0178] The component evidence matching results are compared with the design matching results to obtain the evidence comparison results. This comparison is used to determine whether the components and installation relationships that should be detected in the design data form evidence that can be used for judgment in the field images. If a component exists in the design data, but the field component results do not identify it and the occlusion results show insufficient occlusion confidence, the component is not directly judged as missing, but rather a state of insufficient evidence is formed. If the component is identified, the relationship is established, and the occlusion confidence meets the judgment requirements, a state of valid evidence is formed.

[0179] Based on the evidence comparison results, the integrity of the corresponding states of the components in the relationship graph is determined, and the integrity determination result is obtained. Evidence integrity is expressed by the following formula: ,

[0180] in, Indicates the first Completeness of evidence regarding components or relationships Indicates the number of evidentiary items involved in the component or relationship. Indicates the first The weighting of the importance of each piece of evidence. Indicates the first Whether the evidence items form a valid correspondence, Indicates the first The credibility of the obscuration corresponding to the evidence item.

[0181] Evidence includes component identification evidence, design matching evidence, installation relationship evidence, occlusion relationship evidence, and rule detection evidence. Higher evidence completeness indicates that the current image better supports the quality detection conclusion.

[0182] The results of the integrity assessment are integrated to obtain the evidence results. The evidence results include the state of complete evidence, partially complete evidence, and insufficient evidence, and record the objects and sources of the insufficient evidence.

[0183] When integrating violation results, occlusion results, and evidence results, the severity of the installation rule violation, the credibility of the occlusion, and the completeness of the evidence are comprehensively considered. The purpose of this integration process is to avoid directly outputting a definitive violation conclusion when occlusion is severe or evidence is insufficient, while also avoiding underreporting quality issues when rule violations are clear and evidence is sufficient.

[0184] The quality fusion value is expressed as: ,

[0185] in, Indicates the first The quality fusion value of the detected object, Represents the normalized mapping function. Indicates the degree of violation. Indicates the credibility of the occlusion. Indicates the completeness of evidence. Indicates the degree of design matching deviation. Indicates the risk weight of the detection relationship. , , , , This indicates the fusion weight.

[0186] When the violation is clear and the evidence is complete, the quality inspection result outputs the violation object, violation relationship, and violation type; when the violation exists but the evidence is insufficient, the quality inspection result outputs a review status; when the occlusion credibility is low and the evidence is insufficient, the quality inspection result outputs an insufficient evidence status; when the violation does not exist and the occlusion result and evidence meet the judgment requirements, the quality inspection result outputs a qualified status.

[0187] The quality inspection results should include at least component identification results, design matching results, relationship diagram results, violation results, occlusion results, evidence results, and overall quality status. Through these results, construction personnel can determine the components corresponding to quality problems, their installation relationships, whether occlusion exists, and whether the current image evidence is sufficient to support acceptance judgment.

[0188] In actual execution, firstly, on-site images and design data of the area to be inspected are acquired, and the two are associated according to area identifiers. Then, the on-site images are quality-verified, and images that meet the quality conditions are scaled to obtain calibration images that can be used for actual size conversion.

[0189] Next, component identification is performed on the calibration image to obtain component results. On-site component information is extracted from the component results, and design component information is extracted from the design data. Design matching results are obtained through joint matching of category, location, orientation, and specifications. The design matching results are used to characterize the correspondence between on-site components and design components.

[0190] After obtaining the design matching results, a relationship graph is constructed. Nodes in the relationship graph represent components, and edges represent installation relationships or occlusion relationships. Installation relationships are used for rule detection, while occlusion relationships are used to determine occlusion credibility and evidence integrity.

[0191] During the rule detection phase, installation and testing requirements are extracted from the design data to form a rule table; the relationships to be tested are extracted from the relationship graph, and the relationships to be tested are matched with the rule table. The successfully matched relationships are then judged for compliance to obtain the violation results.

[0192] In the occlusion confidence determination stage, occlusion relationships are extracted from the relationship graph to determine the occlusion source and the occlusion state. The occlusion confidence is calculated by combining the component results and design matching results to obtain the occlusion result.

[0193] In the evidence integrity assessment stage, the installation relationship, occlusion relationship, component results and design matching results are compared to determine whether the current image evidence is sufficient to support the quality inspection conclusion and obtain the evidence result.

[0194] The results of violations, obstructions, and evidence are combined to obtain the quality inspection results. These results indicate whether installation quality issues exist, whether these issues are affected by obstructions, and whether the current evidence is sufficient to support the acceptance judgment.

[0195] In a further embodiment, the quality detection results are manually reviewed, and the image quality weight, scale coefficient fusion weight, matching score weight, rule importance weight, occlusion credibility weight, and quality fusion weight are updated based on the manual review results.

[0196] When a violation result is confirmed to be correct by manual review, the on-site image, design data, component results, relationship diagram, and rule judgment result are used as positive samples; when a detection result is confirmed to be a misjudgment by manual review, it is used as an error sample. The weights are adjusted based on the positive and error samples to adapt to different projects, different shooting conditions, and different electromechanical installation requirements.

[0197] Through the above implementation methods, image recognition results can be correlated with design data and installation inspection requirements to form a relationship graph that can be used to judge installation quality. At the same time, by judging the credibility of occlusion and the integrity of evidence, missing components, occlusion of components and insufficient evidence can be distinguished to improve the accuracy, credibility and acceptance applicability of electromechanical installation quality inspection results.

[0198] Example 1:

[0199] This was used to verify that the relationship graph and rule detection technology solution can further transform the component identification results into installation relationship and rule compliance judgment results, so as to solve the shortcomings of existing image recognition methods that can only identify the appearance of components and are not easy to judge whether the installation rules are met.

[0200] An area for electromechanical installation inspection was selected within the ceiling of a high-density computer room. This area contains air ducts, water pipes, cable trays, and supports. The pipelines in this area are densely intersecting, with cable trays and water pipes installed simultaneously, and some air ducts installed adjacent to supports. During on-site manual inspection, it is necessary to simultaneously determine whether the components exist, whether the components correspond to their designed positions, whether the supports are supporting the corresponding components, and whether the clearance between components meets the design requirements.

[0201] In this embodiment, on-site images and design data of the area to be inspected are first acquired. After quality verification, the on-site images are determined to meet the requirements for subsequent identification and scale calibration. Then, scale calibration is performed based on the component specifications in the design data and the image specifications in the on-site images to obtain calibration images. After component identification is performed on the calibration images, component results are obtained, and the component results are matched with the design data to obtain design matching results.

[0202] After completing the design matching, instead of directly outputting a simple result of "component exists" or "component is missing," the design matching results are further processed to construct installation and occlusion relationships, resulting in a relationship graph. In this graph, support relationships are established between supports and water pipes, clearance relationships between cable trays and water pipes, installation fit relationships between air ducts and adjacent components, and intersection relationships between intersecting pipelines. A rule table is constructed based on the installation inspection requirements in the design data, and rule checks are performed on the relationship graph and rule table to obtain violation results.

[0203] Table 1: Processing results of component identification and relationship construction in the inspection area according to the present invention

[0204] air duct 3 3 3 6 water pipe 8 8 8 14 Cable trays 4 4 4 9 Supports and hangers 16 15 15 15

[0205] As shown in Table 1, this invention first completes component identification, but component identification is not the final detection objective. After identifying air ducts, water pipes, cable trays, and supports, this invention further matches the identified components with design data, and establishes the installation relationships between components based on this. In the support project, the number of identified components is less than the number of design data points, but this invention does not directly use this result as the sole criterion for judgment. Instead, it continues to determine whether the supports form a supporting relationship with the corresponding pipelines through the design matching results and relationship diagrams.

[0206] During the rule-based detection phase, the relationships to be detected are extracted from the relationship graph, and these relationships are matched against the detection objects, relationships, and judgment conditions in the rule table. If the relationship parameters in the relationship to be detected do not meet the judgment conditions, a violation result is generated. Through this process, the detection object is expanded from a single component to the installation relationships between components, allowing support and hanger spacing, component clearance, installation direction, and maintenance space to all be included in quality inspection.

[0207] Table 2: Comparison of detection results between the method of the present invention and existing component identification methods

[0208] Do air ducts, water pipes, cable trays, and supports exist? Does the pipe direction meet the design requirements? It can identify components and generate design matching results. Do the supports and hangers support the corresponding water pipes? Unable to make a stable judgment 15 sets of support relationships were identified. Do the supports and hangers support the corresponding water pipes? Unable to determine Two instances were found where the spacing between supports and hangers did not meet the requirements. Do the spacing between supports and hangers meet the requirements? Unable to determine One location was found where the clearance did not meet the requirements. Does the clearance between the cable tray and the water pipe meet the requirements? Unable to determine One area with insufficient maintenance space was found. Do the ductwork and maintenance space meet the requirements? Relying on manual review One directional anomaly was detected.

[0209] In this embodiment, existing component recognition methods can identify ducts, water pipes, cable trays, and supports in images, but the output mainly focuses on "whether the component is identified". For example, existing methods can identify a water pipe and a support, but cannot determine whether the support is used to support the water pipe, nor can they determine whether the installation relationship between adjacent supports meets the design requirements.

[0210] In this embodiment, the present invention identified five violations of installation rules, including two instances where the spacing between supports and hangers did not meet requirements, one instance where the clearance between the cable tray and the water pipe did not meet requirements, one instance where the maintenance space for the air duct was insufficient, and one instance where the pipe direction was abnormal. These violations cannot be consistently identified solely through component appearance recognition; rather, they require the combined efforts of component identification, design matching, relationship mapping, and rule detection.

[0211] Therefore, this embodiment can intuitively demonstrate that the present invention adopts a relationship graph and rule detection technology solution to achieve the effect of component relationship rule-based judgment technology, realize the transformation of installation quality from appearance recognition to rule compliance judgment, and solve the shortcomings of existing technologies in that it is difficult to identify whether installation rules are met.

[0212] Example 2:

[0213] This is used to verify that after adopting the occlusion credibility and evidence integrity judgment technology, a credible judgment can be made on the occluded area, and whether the current on-site image is sufficient to support the concealed acceptance conclusion, so as to solve the shortcomings of existing image recognition methods that cannot form a credible judgment on the occluded area and have incomplete concealed acceptance evidence.

[0214] Before the ceiling of a certain integrated structure is sealed, an area for inspection of electromechanical installations is selected. This area contains issues such as pipe intersections, cable tray obstructions, partially invisible supports and hangers, and fireproof sealing components being obscured by adjacent structures. This area is a key inspection area before the acceptance of concealed works. Once the ceiling is sealed, the anchorage of supports and hangers, the perimeter of sleeves, fireproof sealing, and cable tray connections will be difficult to directly observe again. Therefore, the inspection results need to determine the presence of the components and whether the current site images are sufficient as evidence for acceptance.

[0215] In this embodiment, firstly, on-site images and design data of the area to be inspected are acquired. The on-site images are then subjected to quality verification and scale calibration to obtain calibration images. Next, component identification is performed on the calibration images to obtain component results, which are then matched with the design data to obtain design matching results. Following this, installation relationships and occlusion relationships are constructed based on the design matching results to obtain a relationship graph.

[0216] In the relationship diagram, the installation relationships between components are recorded, as well as the occlusion relationships. For example, a water pipe in front occupies the support behind it, the side wall of the cable tray partially occupies the support, the edge of the air duct occupies the fireproof sealing area, and the insulation layer occupies the sleeve area. Based on the mentioned occlusion relationships, the credibility of the occlusion is further determined to obtain the occlusion result; at the same time, the integrity of the evidence in the relationship diagram is determined to obtain the evidence result.

[0217] Table 1: Results of Determining the Credibility and Evidence Completeness of the Obscured Object

[0218] Support A Water pipe in front Partially visible 0.74 0.78 Supports quality assessment Support and hanger B cable tray sidewall Locally invisible 0.41 0.46 It is not advisable to directly determine that the item is missing. Fireproof sealing C Duct edge Partially visible 0.57 0.62 Further confirmation is needed. Sleeve D Insulation layer Severe obstruction 0.33 0.38 Insufficient evidence Cable tray connecting piece E Cross water pipes Basically visible 0.82 0.82 Supports quality assessment

[0219] As shown in Table 1, this invention does not simply determine whether a component is qualified based on the recognition confidence level. Instead, it combines the occlusion relationships in the relationship graph to comprehensively judge the source of occlusion, the state of occlusion, and the completeness of evidence. For example, if the support bracket B only shows a partial outline in the on-site image, it might be judged as missing or failed to be recognized according to existing image recognition methods. However, this invention determines through occlusion relationships that it is mainly obstructed by the side wall of the cable tray, and the completeness of the evidence is insufficient. Therefore, it outputs the result "not suitable for direct judgment of missing". The sleeve D is severely obstructed by the insulation layer, and the current image cannot fully display the sleeve part. Therefore, this invention outputs "insufficient evidence" instead of directly outputting missing or unqualified.

[0220] Fireproof sealing C can be partially identified, but due to the obstruction of key parts by the edge of the duct, the current image can only support partial judgment. Therefore, this invention outputs "Needs further confirmation". In contrast, support bracket A and cable tray connecting piece E are obstructed to varying degrees, but the credibility of the obstruction and the completeness of the evidence are both high, which can support quality judgment. Therefore, the output is a result that supports quality judgment.

[0221] Table 2: Comparison of results between existing image recognition methods and the method of this invention in occluded scenarios.

[0222] Support A It was identified as existing. Supports quality assessment This invention further confirms the usability of current image evidence. Support and hanger B Judged as suspected missing It is not advisable to directly determine that the item is missing. This invention identifies the impact of occlusion on the determination of missing items. Fireproof sealing C Judged as suspected missing Further confirmation is needed. This invention distinguishes between defects and incomplete evidence. Sleeve D Determined as missing Insufficient evidence This invention avoids misjudgment caused by occlusion. Cable tray connecting piece E It was identified as existing. Supports quality assessment The reliability of synchronous output evidence in this invention

[0223] Table 2 shows that existing image recognition methods in occluded scenarios mainly rely on whether the component is identified and the level of confidence in the identification. When the component is occluded, resulting in incomplete identification, it is easy to judge the support B and sleeve D as missing, and the fireproof seal C as a suspected defect. However, in the concealed acceptance scenario, missing, defect, and insufficient evidence are three different states. If insufficient evidence is directly judged as a defect, it may lead to unnecessary rework; if the occluded area is directly ignored, the concealed quality risk may be overlooked.

[0224] This invention identifies the source and occluded object of occlusion through occlusion relationships in a relationship graph, and further determines the credibility of the occlusion by combining component results and design matching results. Based on this, the relationship graph is used to determine the completeness of evidence, assessing whether the current on-site images are sufficient to support the acceptance conclusion. Finally, this invention incorporates the occlusion results and evidence results into the quality inspection results, making the inspection results include "whether a problem exists" and "whether the current evidence is sufficient to support the judgment."

[0225] In this embodiment, for the support bracket A and cable tray connecting piece E, which have high credibility and completeness of obstruction evidence, the present invention outputs a result that supports quality judgment; for the sleeve D, which has low credibility and insufficient completeness of evidence, the present invention outputs insufficient evidence; for the fireproof sealing C, which is in an intermediate state, the present invention outputs that supplementary confirmation is needed. This result meets the requirements for evidence reliability in the concealed works acceptance scenario.

[0226] Therefore, this embodiment can intuitively demonstrate that the present invention adopts the technical solution of determining the credibility of occlusion and the integrity of evidence, so as to achieve the technical effect of joint judgment of occlusion status and acceptance evidence, realize the credible quality assessment of the occluded area, and solve the shortcomings of the existing technology of incomplete hidden acceptance evidence.

[0227] Embodiments of the present invention have been presented and described. It will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for inspecting the quality of electromechanical installation based on image recognition, characterized in that, include: Acquire on-site images and design data of the area to be inspected; The on-site images are subjected to quality verification and scale calibration to obtain calibrated images; Component identification is performed on the calibration image to obtain component results; The component results and design data are matched to obtain the design matching results; The design matching results are used to construct installation and occlusion relationships to obtain a relationship graph; A preset rule table is used to perform rule detection on the relationship graph and the rule table to obtain violation results; The occlusion confidence level of the relationship graph is determined to obtain the occlusion result; The relationship graph is subjected to an evidence integrity assessment to obtain the evidence results; The violation results, occlusion results, and evidence results are fused together to obtain the quality inspection results.

2. The electromechanical installation quality inspection method according to claim 1, characterized in that, The on-site images are subjected to quality verification and scale calibration to obtain calibrated images, including: The image quality parameters are obtained by extracting the sharpness, exposure status, and tilt status of the scene image; The image quality parameters are compared with preset quality conditions to obtain a qualified image; The design data is processed to extract component specifications, thus obtaining the design specifications. The component pixel lengths are extracted from the qualified images to obtain the image specifications; The scale factor is calculated based on the design specifications and image specifications to obtain the scale factor. Based on the scale coefficient, the qualified image is scaled to obtain a calibrated image.

3. The electromechanical installation quality inspection method according to claim 1, characterized in that, The component results and design data are matched to obtain design matching results, including: The component results are processed by extracting category, location, orientation, and specifications to obtain on-site component information; The design data is processed by extracting its category, location, orientation, and specifications to obtain the design component information; Candidate matching is performed on the on-site component information and the design component information to obtain candidate matching results; The candidate matching results are compared for deviation to obtain the matching determination result; Based on the matching determination result, the component result is marked accordingly to obtain the design matching result.

4. The electromechanical installation quality inspection method according to claim 1, characterized in that, The design matching results are used to construct installation and occlusion relationships to obtain a relationship graph, including: The design matching results are processed to extract the corresponding state of the components, and the corresponding state results are obtained. Based on the corresponding state results, the components with installation and mating relationships are associated and determined to obtain the installation relationship; Based on the corresponding state results, occlusion determination is performed on components with occlusion influence relationships to obtain the occlusion relationships; The installation and occlusion relationships are correlated and integrated to obtain a relationship graph.

5. The electromechanical installation quality inspection method according to claim 1, characterized in that, A preset rule table is used to perform rule detection on the relationship graph and the rule table to obtain violation results, including: The installation and testing requirements are extracted from the design data to obtain rule parameters; The rule parameters are categorized and organized to obtain a rule table containing detection objects, detection relationships, and judgment conditions; The installation relationships in the relationship graph are extracted to obtain the relationships to be detected; The relationship to be detected is matched with the rule table to obtain the rule matching result; Based on the rule matching results, the relationship to be detected is judged for compliance, and a violation result is obtained.

6. The electromechanical installation quality inspection method according to claim 1, characterized in that, The occlusion confidence level of the relationship graph is determined to obtain the occlusion result, including: The occlusion relationships in the relationship graph are extracted to obtain the occlusion relationship results; Based on the occlusion relationship results, the components in the relationship graph are distinguished by occlusion source and occlusion state to obtain occlusion state results; By mapping the occlusion state results to the component results, the occlusion range results are obtained; Based on the occlusion range result, the corresponding state of the component in the design matching result is verified to obtain the occlusion confidence result; The occlusion confidence results are integrated to obtain the occlusion result.

7. The electromechanical installation quality inspection method according to claim 1, characterized in that, The relationship graph is subjected to an evidence integrity assessment to obtain evidence results, including: The installation and occlusion relationships in the relationship graph are extracted to obtain the evidence relationship results; By mapping the evidence relationship results and component results, the component evidence correspondence results are obtained; The component evidence correspondence results and design matching results are compared to obtain the evidence comparison results; Based on the evidence comparison results, the integrity of the corresponding states of the components in the relationship graph is determined, and the integrity determination results are obtained. The integrity determination results are integrated to obtain the evidence results.

8. The electromechanical installation quality inspection method according to claim 2, characterized in that, Scale factors are calculated based on the design specifications and image specifications to obtain scale factors, including: The design specifications and image specifications are matched to obtain the specification correspondence results; The consistency verification of the specification correspondence results is performed to obtain the valid specification correspondence results; The ratio of the design specifications and image specifications in the effective specification correspondence results is calculated to obtain the individual scale value; The individual scale values ​​are subjected to deviation screening to obtain effective scale values; The effective scale values ​​are fused and calculated to obtain the scale coefficients.

9. The electromechanical installation quality inspection method according to claim 3, characterized in that, The component results are processed by extracting category, location, orientation, and specifications to obtain on-site component information, including: The component results are divided into single components to obtain single component results; The category of the single component is obtained by reading the category of the single component result; Based on the calibration image, the position of the single component is calculated to obtain the component position; The extension direction of the single component is determined to obtain the component direction; Based on the component location and calibration image, the specifications of the single component are converted to obtain the component specifications; The component category, component location, component orientation, and component specifications are integrated to obtain on-site component information.

10. The electromechanical installation quality inspection method according to claim 5, characterized in that, Based on the rule matching results, the relationship to be detected is judged for compliance, and a violation result is obtained, including: The detection object, detection relationship, and judgment condition are extracted from the rule matching results to obtain rule judgment information; The detection object, detection relationship, and relationship parameters are extracted from the relationship to be detected to obtain the information to be determined; The consistency of the information to be determined and the rule-based determination information are verified to obtain valid determination information. Based on the judgment conditions in the rule judgment information, the relationship parameters in the valid judgment information are compared for compliance to obtain the compliance comparison result. The relationships to be detected that do not meet the judgment criteria in the compliance comparison results are marked to obtain individual violation results; The individual violation results are categorized and integrated to obtain the violation results.

Images