A method for detecting defects in an aluminum magnesium manganese metal roof

Abstract

This invention belongs to the field of building engineering quality inspection technology, and discloses a defect detection method for aluminum-magnesium-manganese metal roofs. This method is designed for phased construction and acceptance scenarios of metal roofs. Based on the design documents, it establishes inspection benchmarks such as panel type parameters, vertical joint parameters, purlin axis parameters, fixing seat arrangement parameters, moisture-proof layer overlap parameters, and insulation layer laying parameters. It integrates exposed base images during the construction phase, visible light images and infrared temperature images or elevation data during the sealing phase, and completes registration mapping, measured parameter extraction, abnormal area projection overlap analysis, and consistency verification under a unified roof coordinate system, realizing traceable output of defect type, defect coordinates, and defect level.

Description

Technical Field

[0001] This invention belongs to the field of building engineering quality inspection technology, specifically a defect detection method for aluminum-magnesium-manganese metal roofs. Background Technology

[0002] Because aluminum-magnesium-manganese metal roofing can be laid continuously, has multiple structural layers, and requires close connection between on-site procedures, it is usually constructed and inspected in stages in industrial plants and large-span buildings. In current engineering practice, the inspection work is mostly focused on the appearance inspection and spot checks after the roof is sealed, and the inspection basis is mainly the visible characteristics at the sealing stage.

[0003] In this construction mode, problems such as the center offset of the fixed seat, insufficient overlap of the moisture-proof layer, opening of the joint of the insulation layer and local settlement formed in the base stage are not easily traced back to the original construction position after entering the sealing stage. Even if images of the construction stage, visible light images of the sealing stage and infrared temperature images or elevation data are collected on site at the same time, the data of different stages are still often affected by the difference in collection angle, imaging scale and coordinate reference, making it difficult to form a stable one-to-one correspondence.

[0004] Looking further, although the existing process can extract the base parameters and sealing parameters separately and make threshold judgments, in the cross-stage correlation link, there is often a lack of a continuous mechanism for template constraints after unified coordinate mapping, abnormal area projection overlap analysis, and corresponding relationship verification. This results in an incomplete evidence chain for the same defect between different stages, and insufficient positioning accuracy and consistency in re-inspection and rework.

[0005] The core challenge of the current technology lies in the lack of a continuous correspondence mechanism for cross-construction stage inspection data under a unified spatial benchmark, which in turn affects the traceability, verifiability, and efficiency of defect confirmation and engineering handling. Summary of the Invention

[0006] The purpose of this invention is to provide a defect detection method for aluminum-magnesium-manganese metal roofs, so as to solve the problems mentioned in the background art.

[0007] Current inspections of aluminum-magnesium-manganese metal roofs typically focus on post-installation inspections. Issues such as misaligned fixing seats, insufficient overlap of the moisture barrier, and abnormal installation of the insulation layer, which occur during the construction phase, are difficult to directly correlate with their original positions after installation. Because the coordinate references for data from the construction phase and the installation phase are not consistent, it is difficult to continuously track the same defect across different stages, resulting in insufficient accuracy in re-inspection and rework.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a defect detection method for aluminum-magnesium-manganese metal roofs. This method first reads the design documents and establishes detection benchmark data, including roof panel type parameters, vertical edge height parameters, purlin axis parameters, fixing seat arrangement parameters, damp-proof layer overlap parameters, insulation layer laying parameters, and threshold boundary values ​​corresponding to each measured parameter. Then, images of the exposed state of the base layer during the construction phase are collected, along with visible light images during the panel sealing phase. Simultaneously, one or both of infrared temperature images and elevation data are collected according to the detection requirements. A unified roof coordinate system is established using the eaves baseline, gable baseline, and ridge direction line, unifying the above multi-stage data into the same coordinate frame.

[0009] Under a unified coordinate system, roof panel outline templates, vertical joint centerline templates, and fixed seat position templates are established based on the detection benchmark data, and the measured parameters of the base layer and the measured parameters of the sealing panel are extracted accordingly. The measured parameters for the base layer include the center offset of the fixed seat, the overlap width of the moisture-proof layer, and the opening width of the joint of the insulation layer; the measured parameters for the sealing plate include the vertical joint continuity parameters, the temperature anomaly area parameters, and the elevation gradient connectivity area parameters. The measured parameters are compared with the corresponding threshold boundary values ​​to generate the base layer abnormal area and the sealing plate abnormal area. After projecting the base layer abnormal area onto the sealing plate stage coordinates, the overlap rate between the base layer abnormal area and the sealing plate abnormal area is calculated. The overlap rate is defined as the ratio of the intersection area of ​​the two to the area of ​​the base layer abnormal area after projection. Then, perform the corresponding checks: the check for the offset of the fixed seat and the continuity of the vertical joint is a mandatory check; when collecting infrared temperature images, perform the check for the overlap anomaly and temperature anomaly area; when collecting elevation data, perform the check for the insulation anomaly and the elevation gradient connected area. Finally, the comprehensive deviation score is calculated based on the parameter deviation and overlap rate, and the defect type, defect level and defect coordinates are output. The parameter deviation is the absolute value of the difference between the measured parameter value and the corresponding threshold boundary value.

[0010] This invention places the hidden structural information of the construction stage and the observable anomalies of the sealing stage in the same coordinate system for parameter-level correspondence verification, forming a closed-loop detection process from parameter extraction, anomaly generation, spatial mapping to hierarchical output, which can improve the consistency of location and reliability of judgment of the same defect in cross-stage detection.

[0011] Furthermore, in order to establish a one-to-one spatial correspondence between the exposed base layer images during the construction phase and the visible light images, infrared temperature images, and elevation data during the sealing phase, a roof coordinate system is first constructed. The roof coordinate system uses the eaves baseline as the horizontal reference, the gable baseline as the vertical reference, and the ridge direction line to define the coordinate direction. For multi-stage detection data, control points are first registered and then perspective correction is performed. The control points are selected from the eaves corner, gable corner, ridge intersection, and gutter corner. The control points can be identified and their geometric positions are stable during the construction and slab sealing stages. After control point registration and perspective correction, the images of the exposed base layer during the construction phase and the detection data during the sealing phase can be jointly mapped to a unified roof coordinate system, reducing positional deviations caused by differences in shooting angle and imaging ratio, and providing a unified spatial benchmark for subsequent abnormal area projection, overlap rate calculation and consistency verification.

[0012] Furthermore, the threshold library is established based on calibration samples before the testing process is executed and is called during the testing process. The calibration samples include roof samples without defects and roof samples with known defects. To ensure the comparability of parameters between the two stages, both the calibration samples and the data to be tested are labeled and statistically analyzed using a unified roof coordinate system. The threshold library consists of a threshold sub-library for the construction stage and a threshold sub-library for the sealing stage. The threshold sub-library for the construction stage corresponds to the center offset of the fixed seat, the exposed length of the fasteners, the coverage of the thermal insulation pad, the overlap width of the moisture-proof layer, the proportion of the length of the continuous overlap section, the opening width of the insulation layer joint, and the local settlement. The threshold sub-library for the sealing stage corresponds to the vertical joint height, the vertical joint overlap width, and the vertical edge inclination angle. The aforementioned threshold boundary values ​​are used to determine whether the measured parameters in the construction stage and the sealing stage exceed the limit: when the measured parameters exceed the corresponding threshold boundary values, they are marked as abnormal parameters and an abnormal region is generated. The generated abnormal region is further used for overlap rate calculation, consistency verification and defect classification, so that the defect determination in the construction stage and the sealing stage is completed based on the same parameter benchmark.

[0013] Furthermore, under a unified roof coordinate system, the exposed state images of the base layer during the construction phase are parametrically analyzed to obtain the measured parameters during the construction phase. The measured parameters during the construction phase include the center offset of the fixed seat, the exposed length of the fasteners, the coverage rate of the heat insulation pad, the overlap width of the moisture-proof layer, the proportion of the length of the continuous overlap section, the integrity rate of the insulation layer coverage, the width of the joint opening, and the amount of local settlement.

[0014] Among them, the center offset of the fixed seat is defined as the shortest distance from the center point of the fixed seat to the corresponding purlin axis; the exposed length of the fastener is defined as the normal distance from the top of the fastener to the upper surface of the fixed seat; the coverage rate of the heat insulation pad is defined as the ratio of the effective coverage area of ​​the heat insulation pad to the area of ​​the upper surface of the fixed seat; the overlap width of the moisture-proof layer is defined as the shortest distance between the overlap boundaries of adjacent moisture-proof layers; the proportion of the length of the continuous overlap section is defined as the ratio of the total length of the continuous section with an overlap width not less than the overlap width threshold boundary value to the total length of the corresponding overlap boundary; the insulation layer coverage integrity rate is defined as the ratio of the measured coverage area to the target coverage area of ​​the corresponding zone in the design document; the joint opening width is defined as the maximum opening distance between the boundaries of adjacent insulation layers; the local settlement is defined as the maximum negative deviation of the measured elevation of the insulation layer relative to the zone reference plane, wherein the zone reference plane is obtained by fitting the non-settled area of ​​the same zone.

[0015] The threshold boundary values ​​corresponding to the measured parameters mentioned above are called in the threshold library for comparison. Parameters that exceed the corresponding threshold boundary values ​​are marked as base layer anomaly indicators. Base layer anomaly regions are generated based on the spatial connectivity of the base layer anomaly indicators. The base layer anomaly regions serve as inputs for subsequent cross-stage spatial projection, overlap rate calculation, and consistency verification, so that the structural state of the construction stage and the anomaly information of the sealing stage can be established in a calculable correspondence under the same roof coordinate system.

[0016] Furthermore, during the data analysis process of the sealing stage, the measured parameters of the sealing stage are extracted. The measured parameters of the sealing stage include the vertical joint height, vertical joint overlap width, vertical edge inclination angle, and vertical joint centerline continuity. The vertical joint height, vertical joint overlap width, and vertical edge inclination angle are used to characterize the vertical joint forming state. The vertical joint centerline continuity is used to characterize the degree of continuous extension of the vertical joint within the corresponding vertical joint detection area. The calculation process of the vertical joint centerline continuity is as follows: the total interruption length of the vertical joint centerline within the corresponding vertical joint detection area is counted. When the total length of the corresponding vertical joint is greater than zero, the total interruption length is divided by the total length of the corresponding vertical joint to obtain the interruption length ratio. Then, the interruption length ratio is subtracted from one to obtain the vertical joint centerline continuity. The measured parameters of the above sealing stage are compared with the corresponding threshold boundary values ​​in the threshold library to generate sealing abnormal areas. The sealing abnormal areas serve as inputs for subsequent spatial correspondence and consistency verification with the base layer abnormal areas, and are used for defect type determination and defect level calculation.

[0017] Furthermore, when infrared temperature images exist during the sealing stage, temperature gradients are calculated along the vertical joint direction and perpendicular to the vertical joint direction under a unified roof coordinate system to obtain a bidirectional temperature gradient distribution. The bidirectional temperature gradient distribution is compared with the corresponding temperature gradient threshold boundary values ​​in the threshold library, and continuous threshold-crossing strips are extracted as temperature anomaly zones. Then, connected component extraction is performed on the temperature anomaly zones to form temperature anomaly regions. The temperature anomaly regions correspond one-to-one with the corresponding vertical joint detection areas under the same roof coordinate system, serving as inputs for overlap anomaly correspondence verification and anomaly region overlap rate calculation. Through the above processing, thermal anomaly information during the sealing stage is incorporated into the cross-stage consistency verification process for defect type determination and defect level calculation.

[0018] Furthermore, after acquiring elevation data during the sealing stage, gradient amplitude calculations are performed on the elevation data under a unified roof coordinate system to form an elevation gradient amplitude distribution. This distribution is then compared with elevation gradient threshold boundary values ​​in a threshold library. Regions exceeding the threshold boundary values ​​are marked, and connected component extraction is performed on these regions to obtain elevation gradient connected regions. These regions are then spatially overlaid with the insulation layer detection area to establish a one-to-one correspondence. Based on this, insulation anomaly correspondence verification and anomaly region overlap rate calculation are performed. By incorporating the elevation gradient connected regions into the cross-stage consistency verification process, deformation anomaly information during the sealing stage can be associated with the insulation layer status during the construction stage, which can be used for defect type determination and defect level calculation.

[0019] Furthermore, to quantify the spatial correspondence between the base layer abnormal area and the sealing plate abnormal area, the overlap rate is used as a consistency verification parameter. Both the base layer abnormal area and the sealing plate abnormal area are generated by comparing the corresponding measured parameters with the threshold boundary value. The base layer abnormal area is projected onto the sealing plate stage coordinates to obtain the projected base layer abnormal area. The intersection area of ​​the projected base layer abnormal area and the sealing plate abnormal area is calculated, and the overlap rate is calculated according to the following formula: the overlap rate is equal to the intersection area divided by the area of ​​the projected base layer abnormal area. The intersection area is defined as the area of ​​the overlapping part of the projected base layer abnormal area and the sealing plate abnormal area. The overlap rate is calculated using the following formula: Overlap rate = Intersection area or area of ​​abnormal area in the base layer after projection When the area of ​​the abnormal region in the base layer after projection is greater than zero, the overlap rate is calculated according to the above formula; when the area of ​​the abnormal region in the base layer after projection is equal to zero, the overlap rate is zero. The overlap rate is used as an input parameter for consistency verification and defect level calculation to characterize the spatial matching degree of the abnormal region in cross-stage data.

[0020] Furthermore, after completing the extraction of measured parameters during the construction phase, the extraction of measured parameters during the sealing phase, and the calculation of the overlap rate, defect confirmation is carried out. The defect confirmation uses the parameter threshold boundary value and the overlap rate threshold in the threshold library as a unified criterion. First, the measured parameters during the construction phase are compared with the corresponding threshold boundary values ​​one by one. If any parameter exceeds the corresponding threshold boundary value range, it is determined that the parameter in the construction phase is out of limit. Then, the measured parameters during the sealing phase are compared with the corresponding threshold boundary values ​​one by one. If any parameter exceeds the corresponding threshold boundary value range, it is determined that the parameter in the sealing phase is out of limit. Finally, it is determined whether the overlap rate is not less than the overlap rate threshold. Only when all three conditions are met simultaneously—the parameter in the construction phase is out of limit, the parameter in the sealing phase is out of limit, and the overlap rate is not less than the overlap rate threshold—is the location confirmed as a defect location. This confirmation rule incorporates parameter limit information and spatial correspondence into the judgment process, enabling the detection results of the construction stage and the sealing stage to be correlated under the same judgment caliber, thereby improving the stability and consistency of defect confirmation.

[0021] Furthermore, after the defect is confirmed, the defect level is calculated for each defect location. The defect level calculation is based on the deviation of the measured parameters during the construction stage, the deviation of the measured parameters during the sealing stage, and the overlap rate as inputs. The parameter deviation is defined as the minimum distance from the measured parameter value to the corresponding threshold boundary value and is normalized according to the parameter dimensions. The weight of each input item is predetermined by the statistical results of the calibration sample and remains unchanged during the detection process of the same roof zone, and the total weight is one. The comprehensive deviation score is calculated using weighted averages and then compared with a three-level threshold interval to determine the defect level. The three-level threshold interval is divided by a first threshold and a second threshold, ultimately forming a defect record. The defect record includes construction stage identifier, roof zoning identifier, panel type template number, defect type, defect coordinates, measured parameter values, trigger threshold boundary values ​​and their numbers, and defect level. This classification method incorporates data from the construction stage and the panel sealing stage into the same evaluation scale, which is used for cross-stage result comparison, re-inspection and positioning, and quality traceability.

[0022] The beneficial effects of this invention are as follows: 1. This invention establishes inspection benchmark data by reading design documents and constructs a unified roof coordinate system using the eaves baseline, gable baseline, and ridge direction line. Then, it uses control point registration and perspective correction to map the construction stage data and the sealing stage data in a unified manner. Under the same coordinate frame, the roof panel outline template, vertical joint center line template, and fixed seat position template are loaded synchronously, so that the hidden construction information and the visible anomalies after sealing form a stable correspondence, thereby reducing cross-stage positioning deviation and re-inspection and rework misalignment problems, and improving the continuity and spatial consistency of defect tracking.

[0023] 2. This invention establishes a phased threshold library by calibrating samples, and makes unified judgments on measured parameters such as the center offset of the fixed seat, the exposed length of the fastener, the overlap width of the moisture-proof layer, the opening width of the insulation layer joint, the height of the vertical joint, the overlap width of the vertical joint, and the inclination angle of the vertical edge. When infrared temperature images or elevation data are available, temperature anomaly areas and elevation gradient connected areas are extracted respectively, so that the parameter limit identification is transformed from a single image judgment to a collaborative judgment of geometric information and auxiliary information, thereby reducing missed detections and false detections and improving the stability and reliability of defect identification under different working conditions.

[0024] 3. This invention projects abnormal areas of the base layer onto the coordinates of the sealing plate stage and calculates the overlap rate. It also superimposes the corresponding checks of the fixed seat offset and vertical joint continuity, the corresponding checks of the overlap abnormality and temperature abnormality zone, and the corresponding checks of the insulation abnormality and elevation gradient connection zone. Then, it calculates the comprehensive deviation score by weighting the base layer parameter deviation, sealing plate parameter deviation, and overlap rate. This makes defect confirmation, type determination, and level output form a unified closed-loop process, thereby realizing the traceable output of defect coordinates, defect level, and threshold trigger items, which is convenient for quality review, process accountability, and subsequent process correction. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the overall process for defect detection in aluminum-magnesium-manganese metal roofs according to the present invention. Figure 2 This is a flowchart illustrating the core calculation process for abnormal region generation and overlap rate in this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figures 1 to 2 As shown, this embodiment of the invention provides a defect detection method for aluminum-magnesium-manganese metal roofs, including: In this embodiment, testing benchmark data is first established based on the design documents. The testing benchmark data includes roof panel type parameters, vertical edge height parameters, purlin axis parameters, fixing seat arrangement parameters, moisture-proof layer overlap parameters, insulation layer laying parameters, and a threshold library. The threshold library is pre-established by calibration samples, which include defect-free sample roofs and known defect sample roofs. The sample distribution of each measured parameter is statistically analyzed to form parameter threshold boundary values. The parameter threshold boundary values ​​are written into the threshold library for parameter limit judgment, abnormal area generation, and defect classification.

[0028] Subsequently, images of the exposed state of the base layer during the construction phase were collected, as well as visible light images during the slab sealing phase. Infrared temperature images and elevation data, or both, were collected according to site conditions. To achieve cross-phase correspondence, a roof coordinate system was established. The roof coordinate system was determined by the eaves baseline, gable baseline, and ridge direction line. The corner points of the eaves, gable, ridge, and gutter were selected as control points. The control points were first registered, and then perspective correction was performed to map the construction phase data and the slab sealing phase data to the same roof coordinate system.

[0029] Under a unified roof coordinate system, roof panel outline templates, vertical joint centerline templates, and fixed seat position templates are established based on the detection benchmark data. The templates are then mapped to the corresponding detection data. Based on the template constraints, measured parameters during the construction phase are extracted. These measured parameters include fixed seat center offset, exposed fastener length, insulation pad coverage, moisture barrier overlap width, overlap continuous segment length ratio, insulation layer coverage integrity rate, joint opening width, and local settlement. The fixed seat center offset is the shortest distance from the fixed seat center point to the corresponding purlin axis. The overlap continuous segment length ratio is the ratio of the total length of continuous segments that meet the overlap width threshold boundary value to the total length of the overlap boundary. The measured parameters during the construction phase are compared with the corresponding parameter threshold boundary values ​​in the threshold library, and abnormal areas of the base layer are generated according to spatial connectivity.

[0030] Similarly, measured parameters for the sealing stage are extracted under a unified roof coordinate system. These parameters include the height of the vertical joint, the overlap width of the vertical joint, the inclination angle of the vertical edge, and the continuity of the vertical joint centerline. The continuity of the vertical joint centerline is obtained by subtracting the ratio of the interrupted length of the vertical joint centerline to the total length of the corresponding vertical joint. The measured parameters for the sealing stage are compared with the corresponding threshold boundary values ​​in the threshold library, and abnormal sealing areas are generated according to spatial connectivity.

[0031] Project the abnormal area of ​​the base layer onto the coordinates of the sealing stage, and calculate the overlap rate between the abnormal area of ​​the base layer and the abnormal area of ​​the sealing stage after projection. The overlap rate is the ratio of the intersection area of ​​the two to the area of ​​the abnormal area of ​​the base layer after projection. When the area of ​​the abnormal area of ​​the base layer after projection is zero, the overlap rate is zero.

[0032] During the consistency verification phase, the corresponding verification of the fixed seat offset and the vertical joint continuity is performed. When infrared temperature images are available, the temperature gradient along the vertical joint direction and perpendicular to the vertical joint direction is calculated, temperature anomaly zones are extracted, and the corresponding verification of overlap anomalies and temperature anomaly zones is performed. When elevation data is available, the elevation gradient amplitude is calculated, the elevation gradient connected region is extracted, and the corresponding verification of insulation anomalies and elevation gradient connected regions is performed. When both infrared temperature images and elevation data are available, both types of corresponding verification are performed.

[0033] Based on the parameter limit exceedance results, overlap rate, and consistency verification results, the defect type is confirmed. After confirming the defect type, the parameter deviation is calculated. The parameter deviation is the normalized deviation of the measured parameter value relative to the corresponding parameter threshold boundary value. The base parameter deviation, sealing plate parameter deviation, and overlap rate are calculated into a comprehensive deviation score according to preset weights. The preset weights are determined by the calibration sample statistics and the total weights are one. The defect level is determined based on the three-level threshold interval where the comprehensive deviation score is located, and the defect coordinates and inspection records are output. The inspection records include the construction stage identifier, roof zoning identifier, plate type template number, defect type, measured parameter value, triggered parameter threshold boundary value, and defect level.

[0034] In this embodiment, to address the differences in shooting position, shooting angle, and imaging scale between the exposed base layer images during the construction phase and the data during the sealing phase, a unified roof coordinate system is first established, and then control point registration and perspective correction are performed to ensure that data from different phases correspond under the same spatial reference.

[0035] The roof coordinate system is established according to the following rules: the intersection of the eaves baseline and the gable baseline is taken as the origin of the coordinate system; the direction of the eaves baseline is taken as the positive direction of the horizontal coordinate axis; the direction of the gable baseline is taken as the positive direction of the vertical coordinate axis; and the ridge direction line is taken as the direction constraint line. The eaves baseline, gable baseline, and ridge direction line are extracted from the design documents and their geometric consistency is checked with the on-site boundary measurement results. If any baseline is missing in the same roof zone, the zone will not enter the automatic registration process and will be transferred to the manual review process.

[0036] Control points are established at the eaves corner, gable corner, ridge intersection, and gutter corner. The eaves corner is the intersection of the eaves baseline and the roof boundary line; the gable corner is the intersection of the gable baseline and the roof boundary line; the ridge intersection is the intersection of the ridge direction line and the zone boundary line; and the gutter corner is the intersection of the turning point of the gutter boundary line. The target coordinates of the control points are first generated in the design file, and then the corresponding control points are extracted from the exposed base layer images during the construction phase and the visible light images during the slab sealing phase. Each roof zone has no fewer than four valid control points for registration, and they are not collinear. If there are fewer than four, the automatic registration of that zone is stopped and a verification mark is output.

[0037] Control point registration establishes a mapping relationship based on pairs of corresponding points. The mapping from the exposed base layer image to the roof coordinate system during the construction phase and the mapping from the visible light image to the roof coordinate system during the sealing phase both employ planar projection transformation and are solved using the least squares criterion to minimize the sum of squares of the reprojection errors of all control points. The reprojection error is defined as the Euclidean distance between the mapped coordinates of the control points and the target coordinates. If the average reprojection error is not greater than the error threshold, the registration is deemed valid. If the average reprojection error is greater than the error threshold, control points are re-extracted and the solution is repeated. If the error threshold is not met after two consecutive attempts, the process is transferred to manual review. The error threshold is determined statistically from the calibration samples.

[0038] Perspective correction is performed after control point registration is completed. It is used to eliminate scale and shape distortion caused by differences in viewing angle. After perspective correction, the corrected images for the construction stage and the corrected images for the sealing stage are output, and the mapping relationship between pixels and roof coordinates is preserved. The infrared temperature image and elevation data of the sealing stage adopt the same coordinate mapping relationship as the visible light image of the sealing stage, and are unified to the same roof coordinate system.

[0039] To ensure the stability of subsequent parameter calculations, a consistency check is performed after registration. Within the same roof zone, the directional deviation of the vertical joint centerline of the two-stage calibration images is no greater than the directional deviation threshold. The distance sequence from the fixed seat position template mapping point to the nearest vertical joint centerline maintains a monotonically changing relationship with the design layout pitch sequence. If any condition is not met, it is recorded as a coordinate unification failure zone and will not enter the automatic defect confirmation process. The directional deviation threshold is determined by the calibration sample statistics.

[0040] Through the above steps, the images of the exposed base layer during the construction phase, the visible light images during the sealing phase, the infrared temperature images during the sealing phase, and the elevation data during the sealing phase can establish a recalcible spatial correspondence under the same roof coordinate system, providing a unified coordinate basis for subsequent abnormal area projection, overlap rate calculation, and consistency verification.

[0041] In this embodiment, the threshold library is established before the detection task is executed and is called during the detection process according to the roof partition identifier and panel template number. The data used to establish the threshold library is the calibration sample, which includes defect-free sample roofs and known defect sample roofs. Defect-free sample roofs are roof partitions that have not been detected by on-site verification, and known defect sample roofs are roof partitions whose defect type and defect location have been manually verified. Both types of samples are labeled and statistically analyzed under a unified roof coordinate system.

[0042] The threshold library contains parameter threshold boundary values ​​that correspond one-to-one with the measured parameters. The parameters include the center offset of the fixed seat, the exposed length of the fastener, the coverage of the heat insulation pad, the overlap width of the moisture-proof layer, the proportion of the length of the continuous overlap section, the opening width of the insulation layer joint, the local settlement, the height of the vertical joint, the overlap width of the vertical joint, and the inclination angle of the vertical edge. The units, dimensions, and calculation methods of each parameter are fixed in the parameter definition table and do not change with the partition. The height of the vertical joint, the overlap width of the vertical joint, and the inclination angle of the vertical edge are extracted at equal intervals within the same vertical joint detection area. The interval between the sections is given by the system configuration file.

[0043] The parameter threshold boundary values ​​are divided into upper limit threshold, lower limit threshold, and two-sided threshold according to the parameter risk direction. The threshold determination process is as follows: first, candidate boundary intervals are formed in the defect-free samples, and then boundary search is performed in combination with known defective samples; for each candidate boundary value, the false negative rate and false positive rate are calculated and compared according to the comprehensive judgment cost function, which is the product of the false negative rate weight and the false negative rate plus the product of the false positive rate weight and the false positive rate; the boundary value with the minimum comprehensive judgment cost is taken as the parameter threshold boundary value. The false negative rate weight, false positive rate weight, and minimum recognition accuracy requirement are all given in the system configuration file and the version number is recorded.

[0044] Before writing the threshold, back-substitution verification is performed. If the recognition accuracy of the validation set is not lower than the minimum recognition accuracy requirement, the threshold is written into the threshold library. If it is not met, additional calibration samples are added and recalculated. When each threshold item is entered into the library, the parameter name, threshold direction, threshold boundary value, number of samples, establishment time and version number are recorded synchronously.

[0045] During inspection, the threshold items are first located according to the roof zoning identifier and panel template number. Then, the corresponding parameter threshold boundary values ​​are called according to the inspection stage. If the target zoning lacks a threshold item, the alternative threshold is called in the priority order of the same panel type, same slope, and same inspection stage. If there is still no available threshold, the zoning is marked as a threshold missing zoning and transferred to manual review, and does not enter the automatic defect confirmation process.

[0046] During the operation phase, the measured parameters during the construction phase and the measured parameters during the sealing phase are compared with the corresponding parameter threshold boundary values. Parameters that exceed the threshold boundary value range are marked as abnormal indicators. Abnormal areas are generated according to the eight-neighbor connectivity rule, and connected domains with areas smaller than the minimum connected area threshold are removed to obtain the abnormal areas of the base layer and the abnormal areas of the sealing layer. The minimum connected area threshold is determined by the calibration sample statistics and fixed in the system configuration file.

[0047] Subsequent overlap rate calculations, consistency checks, and defect classifications are all performed based on the aforementioned abnormal indicators and abnormal regions, ensuring that the same threshold caliber is used from parameter comparison to defect output. By establishing a partitioned, panel-shaped, and phased threshold library under a unified roof coordinate system, the drift of the single-stage global threshold under different partitions and imaging conditions can be avoided, thus improving the consistency of cross-stage judgments.

[0048] In this embodiment, the extraction of measured parameters of the base layer is performed after the roof coordinate unification, roof panel outline template mapping, fixed seat position template mapping, and purlin axis mapping are completed. The input data includes the base layer exposed state image during the construction stage, the partition boundary under the unified roof coordinate system, the corresponding purlin axis, fixed seat position template, moisture-proof layer area template, insulation layer area template, and parameter items in the threshold library corresponding to the current roof partition identifier and panel template number. The output data includes the fixed seat center offset, fastener exposed length, insulation pad coverage, moisture-proof layer overlap width, overlap continuous section length ratio, insulation layer coverage integrity rate, joint opening width, and local settlement, and the spatial position index corresponding to each parameter is output simultaneously.

[0049] Before parameter extraction, effective area constraints are performed. The exposed state image of the base layer during the construction stage is cropped according to the partition boundary. Pixels outside the partition, strong reflective areas, occluded areas, and areas with incomplete boundaries are removed. When the ratio of the effective area area to the standard area of ​​the partition is lower than the lower limit given by the threshold library, the partition is marked as a partition to be reviewed and will not enter the automatic judgment process.

[0050] The center offset of the fixed seat is extracted bit by bit according to the fixed seat position template. First, the outer contour of the fixed seat is extracted in the template window, and the geometric center of the outer contour is taken as the center point of the fixed seat. Then, the corresponding purlin axis obtained by mapping the design file is used as a reference to calculate the vertical distance from the center point of the fixed seat to the purlin axis, which is taken as the center offset of the fixed seat. This reference relationship is kept consistent in all partitions.

[0051] The exposed length of the fastener is extracted within the detection area of ​​the fixed seat. After identifying the axial center line of the fastener, the top point of the fastener and the reference line of the upper surface of the fixed seat are determined. The distance along the axial direction of the fastener is calculated as the exposed length of the fastener. The conversion coefficient from pixel to actual length is calibrated using the known structural scale in the current partition template and fixed within the same partition.

[0052] The coverage rate of the heat insulation pad is calculated based on the area ratio. The theoretical coverage area is determined by the boundary of the upper surface of the fixing seat. The actual coverage area of ​​the heat insulation pad is extracted, and the ratio of the actual coverage area to the theoretical coverage area is calculated. The area exceeding the theoretical boundary is not included in the actual coverage area.

[0053] The overlap width of the moisture-proof layer is calculated based on the local normal distance between the boundaries of adjacent moisture-proof layers. Samples are taken at equal intervals along the overlap direction to form an overlap width sequence. The overlap width parameter is taken as the minimum value within the stable section. The stable section is defined as the section where the number of continuous sampling points is not less than the minimum number of points given in the threshold library and the width fluctuation amplitude does not exceed the upper limit of the fluctuation given in the threshold library.

[0054] The proportion of continuous overlap length is calculated based on the overlap width sequence. Sampling points that are not less than the overlap width threshold boundary value are marked as valid overlap points. Continuous valid overlap points form continuous segments. The proportion of continuous overlap length is defined as the ratio of the sum of the lengths of all continuous segments to the total length of the overlap boundary of the zone. The total length of the overlap boundary is determined by the template of the moisture-proof layer area.

[0055] The insulation layer coverage integrity rate is calculated by zone. Based on the theoretical coverage area given by the insulation layer zone template, the actual coverage area of ​​the insulation material is extracted, and the area ratio of the two is calculated. Holes, gaps, and unlaid areas within the zone are all included in the uncovered area.

[0056] The joint opening width is measured between the boundaries of adjacent insulation layer panels. After extracting the center line of the joint, samples are taken at equal intervals along the center line. The lateral spacing between adjacent boundaries is calculated to obtain the opening width sequence. The maximum value of the joint opening width after denoising is taken, and the corresponding roof coordinates are recorded. The denoising rule adopts the minimum coherent length constraint given by the threshold library. Isolated single-point peak values ​​are not included in the maximum value calculation.

[0057] The local settlement amount is calculated from the partitioned elevation grid reconstructed from the multi-view images of the exposed base layer during the construction phase. First, the non-settled boundary points in the same partition are selected to fit the partition reference plane. Then, the height difference between the detection point and the reference plane is calculated. The local settlement amount is defined as the maximum value of the absolute value of the negative height difference. If the current partition cannot form an effective elevation grid, the local settlement amount is marked as missing and manual review is triggered. Automatic grading is not performed on this partition.

[0058] After the eight measured parameters at the grassroots level are extracted, the parameters are matched with the threshold library one by one according to their names to perform the limit judgment. Each limit-breaking parameter is bound to a spatial location to generate an abnormal parameter identification map. The abnormal parameter identification map is merged according to the eight-neighbor connectivity rule to form a grassroots abnormal region. Connected regions with an area smaller than the minimum connected area threshold given by the threshold library are removed. The grassroots abnormal region outputs a partition identifier, abnormal parameter item, abnormal coordinate boundary and trigger threshold item, providing input for subsequent cross-stage consistency verification.

[0059] In this embodiment, the extraction of measured parameters during the sealing stage is performed after the roof coordinates are unified. The input data includes the visible light image of the sealing stage, the roof panel outline template, the vertical joint centerline template, the roof partition boundary, and the threshold items in the threshold library corresponding to the current roof partition identifier and panel template number. When the sealing stage elevation data exists, the sealing stage elevation data is used as an additional input for parameter calculation. When the sealing stage elevation data does not exist, the image geometry measurement aperture is used and the data source is marked in the result. To ensure the consistency of the measurement aperture, the sealing stage data is first cut according to the roof partition boundary, and then the vertical joint detection area is determined according to the vertical joint centerline template. Only within the vertical joint detection area are four measured parameters of the sealing stage: vertical joint height, vertical joint overlap width, vertical edge inclination angle, and vertical joint centerline continuity.

[0060] The vertical joint height is obtained by cross-sectional measurement. Measurement sections are laid out along the center line of the vertical joint according to the cross-sectional spacing parameters. The cross-sectional spacing parameters are given by the corresponding partition parameter item of the threshold library. The shortest distance between the top point of the vertical joint and the reference points on both sides of the plate is extracted on each cross-section to form a cross-sectional height sequence. The vertical joint height parameter is the median value of the cross-sectional height sequence, and the maximum and minimum values ​​are recorded for verification.

[0061] The vertical joint overlap width is measured on the same cross section to find the shortest distance between the two overlap boundaries, forming an overlap width sequence. The vertical joint overlap width parameter is taken as the minimum value within the stable section. The stable section is defined as having a continuous measurement point number not less than the minimum point number threshold and an adjacent point change not greater than the fluctuation threshold. The minimum point number threshold and the fluctuation threshold are both given by the corresponding parameter items in the threshold library.

[0062] The vertical edge inclination angle is determined by the angle between the vertical edge outline and the zoning reference plane in the unified roof coordinate system. The inclination angle sequence is obtained by calculating point by point according to the cross section, and the vertical edge inclination angle parameter is taken as the median value of the inclination angle sequence.

[0063] The continuity of the vertical joint centerline is calculated based on the proportion of interruptions. First, the current vertical joint centerline trajectory is extracted based on the vertical joint centerline template. Then, interrupted segments in the trajectory are identified, and the total length of the interrupted segments is taken as the interruption length. The total length of the vertical joint trajectory within the detection area is taken as the corresponding total length of the vertical joint. The continuity of the vertical joint centerline is defined as the ratio of the interruption length to the corresponding total length of the vertical joint. The smaller the ratio, the better the continuity; the larger the ratio, the worse the continuity. When the corresponding total length of the vertical joint is zero, the vertical joint is marked as an invalid vertical joint and is transferred to manual review.

[0064] After the four parameters are extracted, an effectiveness check is performed. If the number of effective sections in the same vertical joint detection area is lower than the minimum section number threshold, or the center line interruption identification result exceeds the template boundary tolerance, then the vertical joint detection area is marked as a low confidence detection area and a verification mark is output. It does not directly enter the defect classification. The minimum section number threshold and the template boundary tolerance are given by the corresponding parameter items in the threshold library.

[0065] For the detection areas that pass the validity check, the vertical joint height, vertical joint overlap width, vertical edge inclination angle, and vertical joint centerline continuity are compared with the corresponding thresholds in the threshold library item by item to generate sealing plate abnormal parameter markers. These markers are then merged according to the eight-neighbor connectivity rule to form sealing plate abnormal areas. The sealing plate abnormal areas serve as inputs for subsequent overlap rate calculation and consistency check with the base layer abnormal areas.

[0066] By uniformly extracting geometric height, overlap width, posture angle, and centerline integrity within the same vertical joint detection area, and completing threshold determination and region merging under a unified roof coordinate system, cross-stage alignable sealing plate parameter results can be generated.

[0067] In this embodiment, when an infrared temperature image exists during the sealing stage, a temperature anomaly detection process is executed. The input data includes the infrared temperature image of the sealing stage under a unified roof coordinate system, the visible light image of the sealing stage, the vertical joint centerline template, the roof partition boundary, and temperature parameter items in the threshold library. The temperature parameter items include the gradient threshold along the vertical joint direction, the gradient threshold perpendicular to the vertical joint direction, the joint response threshold, the minimum strip length threshold, the minimum connected area threshold, the merging distance threshold, and the direction difference threshold.

[0068] First, the infrared temperature image of the sealing stage is mapped to a unified roof coordinate system and spatially aligned with the visible light image of the sealing stage. Then, within each roof section, the direction of the vertical joint and the direction perpendicular to the vertical joint are determined based on the center line template of the vertical joint. Based on the temperature field, the temperature gradient along the vertical joint direction and the temperature gradient perpendicular to the vertical joint direction are calculated respectively. To reduce the influence of noise, the temperature field is edge-preserving smoothed before the gradient calculation. The smoothing window and sampling step size are given by the corresponding partition parameter item of the threshold library.

[0069] The criteria for determining candidate points of temperature anomalies are as follows: When the temperature gradient along the vertical seam is not less than the corresponding threshold and the temperature gradient perpendicular to the vertical seam is not less than the corresponding threshold, it is determined to be a candidate point of the first type; when neither of the two conditions is met simultaneously, the joint response value is calculated. The joint response value is the weighted sum of the gradients in the two directions according to a preset weight. When the joint response value is not less than the joint response threshold, it is determined to be a candidate point of the second type. The candidate points of the first type and the candidate points of the second type are merged to form a set of candidate points of temperature anomalies.

[0070] Candidate points of temperature anomalies are continuously connected along the vertical seam direction. When connecting, the interval between adjacent candidate points in the vertical seam direction is not greater than the connection interval threshold. After connecting, candidate strips are obtained. Candidate strips with a length less than the minimum strip length threshold are removed. The remaining strips are defined as temperature anomaly strips.

[0071] Connectivity extraction is performed on the temperature anomaly zone. The connectivity rule adopts eight-neighbor connectivity. After the connectivity is extracted, the area is first filtered, and the connectivity with an area less than the minimum connectivity area threshold is removed. Then, adjacent connectivity is merged. When the distance between the center lines of two connectivity is not greater than the merging distance threshold and the directional angle is not greater than the directional difference threshold, the merging result is defined as the temperature anomaly region.

[0072] The output fields for the temperature anomaly area include roof zoning identifier, anomaly area boundary coordinates, anomaly zone centerline, area area, trigger threshold, and confidence level. The confidence level is determined based on the over-limit strength and connectivity stability. Over-limit strength is the ratio of the actual gradient to the corresponding threshold, and connectivity stability is the ratio of the length of the continuous strip segment to the total length. The temperature anomaly area serves as the input for subsequent consistency verification, corresponding to the verification of the moisture barrier overlap anomaly execution space, and participates in the calculation of anomaly area overlap rate and defect type determination.

[0073] If there is no infrared temperature image during the sealing stage, the temperature anomaly detection process is not executed, and the temperature process is marked as not enabled in the detection record. Subsequently, only the verification process corresponding to the enabled data item is executed.

[0074] In this embodiment, when elevation data exists during the sealing stage, an elevation anomaly detection process is executed to form an elevation gradient connected region. This process is initiated after the roof coordinate unification, roof partition boundary determination, and vertical joint centerline template mapping are completed. The input data includes the sealing stage elevation data under the unified roof coordinate system, roof partition boundaries, vertical joint centerline template, insulation layer detection area boundaries, and elevation parameter items in the threshold library. The elevation parameter items include elevation gradient amplitude threshold, minimum connected area threshold, minimum connected length threshold, connection spacing threshold, and direction difference threshold.

[0075] First, establish a reference plane for each roof section. The elevation points of the stable area are selected according to the following rules: they are located within the boundary of the section and are not in the gutter boundary, corner boundary, or missing data area. At the same time, the local slope is not greater than the slope threshold of the stable point. The reference plane of the section is fitted with the elevation points of the stable area, and the elevation residuals relative to the reference plane are calculated for the effective elevation points in the section to obtain the normalized elevation field of the section.

[0076] The elevation gradient magnitude is calculated on the partitioned normalized elevation field. The horizontal and vertical axes are defined as the horizontal and vertical coordinate axes of the unified roof coordinate system, respectively. For each valid elevation point, the horizontal elevation change rate and the vertical elevation change rate are calculated using central difference, and the elevation gradient magnitude is obtained by taking the square root of the sum of squares. The central difference grid spacing adopts the grid spacing parameter corresponding to the current partition. The elevation gradient magnitude is compared with the elevation gradient magnitude threshold point by point, and points exceeding the threshold are marked as candidate points for elevation anomalies.

[0077] Eight-neighbor connected component extraction is performed on candidate elevation anomalies to form initial connected components. Then, filtering and merging are performed: connected components with an area smaller than the minimum connected area threshold are removed, and connected components with a principal axis length smaller than the minimum connected length threshold are removed; when the distance between the centerlines of two connected components is not greater than the connection spacing threshold and the angle between the principal directions is not greater than the direction difference threshold, they are merged. The connected components after filtering and merging are defined as elevation gradient connected regions, and the length of the connected region is defined as the principal axis length of the connected component.

[0078] After the elevation gradient connected region is formed, structural constraint verification is performed. The elevation gradient connected region is spatially superimposed with the insulation layer detection area, and only the connected regions located within the insulation layer detection area or intersecting with the boundary of the insulation layer detection area are retained. Then, the retained connected regions are checked for directional consistency with the vertical joint centerline template. Connected regions whose directional deviation exceeds the directional difference threshold are marked as verification connected regions and are not included in the automatic judgment.

[0079] The final output is the elevation gradient connected region. The output fields include roof partition identifier, connected region boundary coordinates, connected region area, connected region length, main direction angle, trigger threshold item, and data validity identifier. If no connected region meets the filtering conditions, an empty connected region is output and the data validity identifier is set to no valid connected region. The above output is used as input for subsequent verification of insulation anomalies and calculation of the overlap rate of anomaly areas, and is not used as a separate conclusion for the defect level.

[0080] If no elevation data exists during the sealing stage, this process will not be executed. Instead, an "elevation process not enabled" flag will be written into the inspection record, and the system will continue to execute the inspection process corresponding to the enabled data item.

[0081] In this embodiment, the overlap rate is used to characterize the degree of spatial correspondence between base layer anomalies and sealing plate anomalies within the same roof partition. The overlap rate calculation is performed after the roof coordinate unification, base layer anomaly area projection, and sealing plate anomaly area extraction are completed. The input data includes the projected set of base layer anomaly areas, the set of sealing plate anomaly areas, and the roof partition boundary, all of which are located in the unified roof coordinate system.

[0082] First, the two types of abnormal areas are trimmed according to the roof partition boundary to obtain the effective area within the current partition. Then, the overlapping sub-areas within the abnormal areas of the projected base layer are merged to obtain the effective area of ​​the projected base layer. The overlapping sub-areas within the abnormal areas of the sealing plate are merged to obtain the effective area of ​​the sealing plate, thus avoiding duplicate area calculation for the same type of sub-area. The intersection of the effective area of ​​the projected base layer and the effective area of ​​the sealing plate is obtained, and the overlap rate is calculated using the following formula: Overlap rate = Intersection area / Effective area of ​​the base layer after projection; Wherein, the intersection area is the total area of ​​the intersection region; the effective area of ​​the base layer after projection is the total effective area of ​​the base layer after projection. If the effective area of ​​the base layer after projection is less than the minimum effective area threshold, the overlap rate of the partition is recorded as 0 and marked as an area without effective projection. The process is then transferred to manual review and will not proceed to the automatic defect confirmation process. When multiple abnormal sub-regions exist, the intersection area is calculated by summing the areas of all intersection sub-regions, and the effective area of ​​the base layer after projection is calculated by summing the areas of all projected sub-regions. Both use the same area caliber, and the area calculation is uniformly performed under the roof coordinate system. For raster data, the calculation is performed by accumulating the unit surface, and for vector surface regions, the calculation is performed by surface integration.

[0083] When outputting the overlap rate, the roof zoning identifier, intersection area, effective area of ​​the base layer after projection, overlap rate value, and threshold comparison results are recorded simultaneously and written into the detection record for subsequent consistency verification, defect confirmation, and defect classification.

[0084] In this embodiment, defect confirmation is performed after the extraction of base layer measured parameters, the extraction of sealing plate measured parameters, the generation of abnormal areas and the calculation of overlap rate are completed. The input data includes base layer measured parameters and their threshold items, sealing plate measured parameters and their threshold items, sub-region overlap rate and overlap rate threshold, roof zoning identifier, board type template number and data validity identifier. The threshold call adopts the same roof zoning identifier, the same board type template number and the same threshold version number.

[0085] Defect confirmation is first performed at the sub-region level, and then the data is summarized by region. Within each abnormal sub-region, the following flags are calculated: base layer exceeding limit flag, sealing plate exceeding limit flag, and overlap passing flag. The base layer exceeding limit flag is set to 1 if at least one of the measured parameters of the base layer in the sub-region exceeds the limit, otherwise it is set to 0. The sealing plate exceeding limit flag is set to 1 if at least one of the measured parameters of the sealing plate in the sub-region exceeds the limit, otherwise it is set to 0. The overlap passing flag is set to 1 if the overlap rate of the sub-region is not less than the overlap rate threshold, otherwise it is set to 0.

[0086] The rules for exceeding the limit for parameters are uniformly as follows: for upper limit threshold items, the measured value is determined to be exceeding the threshold; for lower limit threshold items, the measured value is determined to be exceeding the threshold; and for bilateral threshold items, the measured value exceeds the upper or lower boundary interval.

[0087] The condition for confirming defects in a sub-region is that all three flags are 1 at the same time: the base layer exceeding the limit flag is 1, the sealing plate exceeding the limit flag is 1, and the overlapping passage flag is 1. When these conditions are met, the sub-region is determined to be a confirmed defect sub-region. When any one of the flags is 0, the sub-region is determined to be an unconfirmed defect sub-region.

[0088] Data validity gating takes precedence over the above judgment rules. If a sub-region has no valid value for the measured parameters at the base level, no valid value for the measured parameters at the sealing plate, or no valid value for the overlap rate of the sub-region, then the sub-region is marked as a manually reviewed sub-region and will not be automatically confirmed. The low confidence level is marked by the preceding process, and the low confidence level sub-region is treated as a manually reviewed sub-region.

[0089] The rules for partitioning are as follows: when at least one sub-region within a partition is a confirmed defect sub-region, the partition is determined to be a confirmed defect partition; when all sub-regions within a partition are unconfirmed defect sub-regions, the partition is determined to be an unconfirmed defect partition; when there are sub-regions requiring manual review within a partition and no confirmed defect sub-regions, the partition is determined to be a partition awaiting review.

[0090] The output includes two levels of content. The sub-region level output includes sub-region coordinate boundaries, base layer exceedance parameters, sealing plate exceedance parameters, sub-region overlap rate, overlap rate threshold, sub-region confirmation flag, and data validity flag. The zoning level output includes construction stage flag, roof zoning flag, panel type template number, zoning confirmation status, and trigger sub-region index. These results serve as inputs for subsequent defect classification and report generation.

[0091] In this embodiment, the comprehensive deviation score is calculated after the defect is confirmed. The calculation object is the confirmed defect sub-region within the same roof partition. The input data includes the measured parameters and threshold items of the base layer, the measured parameters and threshold items of the sealing plate, the sub-region overlap rate, parameter weight, stage weight, and grade interval threshold. The threshold library is called using the same roof partition identifier, the same board type template number, and the same threshold version number.

[0092] The deviation of grassroots parameters is weighted and summarized according to the extent of exceeding the limit of the measured parameters at the grassroots level. For any parameter, the normalized deviation value is calculated first. The normalized deviation value of the upper limit threshold parameter is the measured value minus the threshold value and then divided by the denominator constant. The normalized deviation value of the lower limit threshold parameter is the threshold value minus the measured value and then divided by the denominator constant. For the bilateral threshold parameter, the upper boundary deviation value and the lower boundary deviation value are calculated separately and the larger value is taken. The denominator constant is the larger value between the absolute value of the threshold value and the minimum denominator constant. The normalized deviation value is zero if it is less than zero and one if it is greater than one. The deviation of grassroots parameters is the result of the weighted average of the normalized deviation values ​​of effective grassroots parameters according to the parameter weights.

[0093] The deviation of the sealing plate parameters adopts the same normalization and weighting rules as the deviation of the base plate parameters. The parameter weights are given by the threshold library and are fixed under the same plate type template number.

[0094] The overall deviation score is calculated using the following formula: The comprehensive deviation score equals the weight of the grassroots stage multiplied by the deviation of the grassroots parameters, plus the weight of the sealing stage multiplied by the deviation of the sealing parameters, plus the overlap rate weight multiplied by the sub-region overlap rate.

[0095] The weights for the grassroots stage, the sealing stage, and the overlap rate are all non-negative, and the sum of the three is one. The overall deviation score is limited to the range of zero to one.

[0096] The defect level is determined using a three-level interval. The thresholds for the first and second levels are set to zero to less than or equal to one. When the comprehensive deviation score is less than the first level threshold, it is determined to be a level three defect. When the comprehensive deviation score is not less than the first level threshold and less than the second level threshold, it is determined to be a level two defect. When the comprehensive deviation score is not less than the second level threshold, it is determined to be a level one defect. Boundary values ​​are classified according to the higher level.

[0097] The defect type is determined by the combination of triggering parameters and the consistency check results. When the center offset of the fixed seat or the exposed length of the fastener is triggered and the continuity of the vertical joint centerline is abnormal, it is judged as a fixed connection related defect. When the overlap width of the moisture-proof layer or the proportion of the length of the continuous overlap section is triggered and the corresponding temperature abnormal area is established, it is judged as an overlap sealing related defect. When the insulation layer coverage integrity rate or the joint opening width is triggered and the corresponding elevation gradient connected area is established, it is judged as an insulation layer related defect. When the vertical joint height, vertical joint overlap width or vertical edge inclination angle is triggered, it is judged as a vertical joint geometry related defect. If the same sub-region meets multiple conditions at the same time, the contribution score of each type is calculated separately and the one with the highest contribution score is taken as the main defect type. The rest are secondary type records. The type contribution score is calculated by weighting the normalized deviation value of the triggering parameter of that type and the corresponding parameter weight.

[0098] Data validity gating takes precedence over hierarchical execution. If any of the following parameters—baseline parameter deviation, sealing plate parameter deviation, or sub-region overlap rate—has no valid value, the sub-region will not output automatic hierarchical results, will be marked as a manually reviewed sub-region, and the name of the missing item and the reason for the missing item will be output.

[0099] The output results include construction stage identifier, roof zoning identifier, panel type template number, sub-region identifier, main defect type, secondary defect type, defect coordinates, measured parameter values, threshold items, comprehensive deviation score, defect level, and data validity identifier. The defect coordinates adopt the sub-region boundary coordinates under the unified roof coordinate system. The measured parameter values ​​and threshold items are output in one-to-one correspondence according to the parameter names. The zoning-level results are obtained by summarizing the sub-region results.

[0100] This embodiment achieves consistent defect classification and type determination across stages by using a combined weighted classification of base layer parameter deviation, sealing plate parameter deviation, and sub-region overlap rate.

[0101] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these 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 defect detection method for aluminum-magnesium-manganese metal roofs, characterized in that: include: Read the design documents of the target roof and establish test benchmark data, which includes roof panel type parameters, vertical edge height parameters, purlin axis parameters, fixing seat arrangement parameters, moisture-proof layer overlap parameters, insulation layer laying parameters, and threshold library; Images of the exposed state of the base layer during the construction phase are collected, visible light images during the sealing phase are collected, and auxiliary detection data during the sealing phase are collected. The auxiliary detection data is one or both of infrared temperature images and elevation data. Establish a roof coordinate system and unify the construction stage data and the slab sealing stage data into the roof coordinate system; Based on the aforementioned detection benchmark data, roof panel outline templates, vertical joint centerline templates, and fixed seat position templates are established, and the templates are mapped to the unified data. Extract measured parameters of the base layer from the construction phase data, and generate abnormal areas of the base layer according to the threshold library; Extract the actual sealing parameters from the sealing stage data, and generate sealing anomaly areas according to the threshold library; Project the abnormal area of ​​the base layer onto the coordinates of the sealing stage, and calculate the overlap rate between the abnormal area of ​​the base layer and the abnormal area of ​​the sealing stage after projection. Perform a consistency check, which includes at least the corresponding check of the fixed seat offset and the continuity of the vertical joint; when infrared temperature images are available, perform a corresponding check of the overlap anomaly and the temperature anomaly zone; when elevation data is available, perform a corresponding check of the insulation anomaly and the elevation gradient connected area. The defect type is determined based on the parameter deviation and overlap rate, and the comprehensive deviation score is calculated. The defect level and defect coordinates are then output based on the comprehensive deviation score.

2. The defect detection method for aluminum-magnesium-manganese metal roofs according to claim 1, characterized in that: The roof coordinate system is determined by the eaves baseline, gable baseline and ridge direction line. The unified processing adopts control point registration and perspective correction. The control points include the eaves corner point, gable corner point, ridge intersection point and gutter corner point.

3. The defect detection method for aluminum-magnesium-manganese metal roofs according to claim 2, characterized in that: The threshold library is established through calibration samples, which include roof samples without defects and roof samples with known defects. The threshold items include the center offset threshold of the fixing seat, the exposed length threshold of the fastener, the coverage threshold of the thermal insulation pad, the overlap width threshold of the moisture-proof layer, the proportion threshold of the length of the continuous overlap section, the opening width threshold of the insulation layer joint, the local settlement threshold, the vertical joint height threshold, the vertical joint overlap width threshold, and the vertical edge tilt angle threshold.

4. The defect detection method for aluminum-magnesium-manganese metal roofs according to claim 3, characterized in that: The measured parameters of the base layer include at least the center offset of the fixing seat, the exposed length of the fasteners, the coverage of the heat insulation pad, the overlap width of the moisture-proof layer, the proportion of the length of the continuous overlap section, the coverage integrity rate of the insulation layer, the width of the joint opening, and the local settlement; wherein, the center offset of the fixing seat is the vertical distance from the center point of the fixing seat to the axis of the corresponding purlin.

5. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 4, characterized in that: The measured parameters of the sealing plate include at least the vertical joint height, vertical joint overlap width, vertical edge inclination angle, and vertical joint centerline continuity; the vertical joint centerline continuity is determined by the ratio of the vertical joint centerline interruption length to the corresponding total vertical joint length.

6. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 5, characterized in that: When infrared temperature images are available, temperature anomaly zones are calculated by temperature gradients along the vertical seam direction and perpendicular to the vertical seam direction, and temperature anomaly regions are formed by connected component extraction.

7. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 6, characterized in that: When elevation data exists, the elevation gradient connected region is obtained by calculating the gradient magnitude of the elevation data and performing connected component extraction.

8. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 7, characterized in that: The overlap rate is calculated using the following formula: Overlap rate = intersection area or area of ​​abnormal area in the base layer after projection; The intersection area is the area of ​​the intersection between the abnormal area of ​​the base layer and the abnormal area of ​​the sealing plate after projection.

9. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 8, characterized in that: The defect confirmation criteria are: at least one parameter in the measured parameters of the base layer exceeds the corresponding threshold, at least one parameter in the measured parameters of the sealing plate exceeds the corresponding threshold, and the overlap rate is not less than the overlap rate threshold.

10. A defect detection method for aluminum-magnesium-manganese metal roofs according to claim 9, characterized in that: The comprehensive deviation score is calculated by the deviation of the base parameters, the deviation of the sealing parameters, and the overlap rate according to the preset weights. The defect level is determined based on the three-level threshold range corresponding to the comprehensive deviation score. The output results include construction stage identifier, roof zoning identifier, panel type template number, defect type, defect coordinates, measured parameter values, threshold items, and defect level.

Images