A method for multi-period image comparison and deterioration trend prediction of external riser risk points

Abstract

This invention relates to the field of image processing technology and discloses a method for comparing and predicting the degradation trend of risk points in external risers across multiple periods. The method includes: acquiring multi-period datasets; establishing unified spatial coordinates and fitting a cylindrical surface model based on facade control points and external riser components to obtain standardized images; performing cross-period registration on the standardized images and unfolding the external riser region into a surface unfolding plane to obtain cross-period co-location images; processing the cross-period co-location images to obtain cross-period comparable images; extracting quantitative indicators within the surface unfolding plane to obtain a set of risk point indicators; constructing a spatiotemporal coupling graph and obtaining the cross-period trajectory of risk points; fusing the risk point indicator set to form a degradation state representation; obtaining the estimated time to reach the maintenance threshold based on the degradation state representation and generating maintenance recommendations. This application improves the accuracy of external riser risk point location and the reliability of trend prediction.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to a method for comparing multi-period images of risk points of external risers and predicting their degradation trends. Background Technology

[0002] External risers are constantly exposed to wind, rain, and temperature fluctuations, making them susceptible to corrosion, leakage, deformation, and loosening. With the increasing scale and aging of urban infrastructure, maintenance departments need to conduct high-frequency, standardized, and comparable remote inspections and predictive maintenance of external risers without affecting their usability. While image-based automated diagnostics offer advantages such as wide coverage, low cost, and traceability, external risers, being cylindrical curved surfaces, suffer from incomparability across different time periods due to changes in viewing angle, lighting, and thermal radiation. Furthermore, failure evolution exhibits pathways such as axial dripping, propagation along supports, and penetration into walls, making single-period assessments insufficient to characterize the physical coupling relationships between the affected components.

[0003] Existing solutions mostly focus on single-phase defect identification or equipment-level inspection organization, lacking cross-phase registration and physical constraint modeling for cylindrical curved surfaces. For example, CN112541887B provides a leakage defect detection method based on thermal imaging and morphological features, which can locate anomalies in a single operation, but it does not establish a unified spatial coordinate and cylindrical curved surface model for cylindrical curved surfaces, nor does it perform cross-phase registration and surface unfolding under multi-phase conditions. It is difficult to achieve co-location comparison of the same physical part in different acquisition periods, and it does not construct a spatiotemporal coupling diagram to describe the path relationships of axial dripping, support connection, and wall penetration. The suspended tunnel inspection device, system, and method described in CN112729405A emphasizes circumferential stitching and equipment accessibility, enabling long-distance image coverage in a single phase. However, it primarily focuses on acquiring and stitching images for structural inspection, neglecting issues such as the alignment of structural anchor points on external risers, inter-phase radiation standardization, and thermal imaging physical correction. Furthermore, it does not introduce edge weight determination methods based on capillary diffusion, gravity dripping, and material moisture absorption mechanisms to achieve inter-phase correlation and unified numbering. In summary, existing technologies exhibit weak inter-phase comparability and a high misjudgment rate.

[0004] Therefore, it is necessary to design a method for comparing multi-period images of external riser risk points and predicting deterioration trends to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a method for comparing and predicting the degradation trend of risk points of external risers in multiple periods, aiming to solve the problems of poor inter-period comparability and high false detection rate caused by the lack of inter-period isotope, inter-period radiation standardization and thermal image physical correction under unified spatial coordinates and cylindrical surface model, as well as spatiotemporal coupling map correlation based on physical mechanism in the existing technology.

[0006] This invention proposes a method for comparing multi-period images of risk points in external risers and predicting their degradation trends, including:

[0007] Multiple images were acquired, and the acquisition time, acquisition posture, imaging parameters and environmental parameters were recorded. A unified time reference and sliding time window were established to obtain a multi-phase dataset. The multi-phase images include visible light images and thermal images of the external riser.

[0008] Based on the facade control points and external pipe components, a unified spatial coordinate system is established and a cylindrical surface model is fitted. Imaging distortion correction and perspective correction are performed on the multi-period dataset to obtain standardized images.

[0009] Under the constraints of the cylindrical curved surface model, the standardized image is registered across periods, and the outer riser area is unfolded into a curved unfolded plane. The structural anchor points are used for co-alignment and cyclic consistency verification to obtain the cross-period co-aligned image.

[0010] Based on the imaging parameters and environmental parameters, the intertemporal isotopic images are subjected to intertemporal radiometric normalization, and the thermal imaging images are physically corrected to obtain intertemporal comparable images.

[0011] Risk points are generated within the surface development plane for corrosion, leakage, deformation and loosening. Quantitative indicators are extracted based on appearance changes, morphological changes, thermal anomalies and geometric deviations to obtain a set of risk point indicators.

[0012] A spatiotemporal coupling graph is constructed using the risk points as nodes and the axial dripping path, the support connection path, and the wall penetration path as edges. The edge weights are determined based on capillary diffusion, gravity dripping, and the material moisture absorption mechanism. The risk points are then correlated across periods and assigned a unified number to obtain the cross-period trajectory of the risk points.

[0013] Within a sliding time window, the risk point indicator set is weighted and fused to form a degradation state representation. The degradation state representation is then subjected to trend fitting and consistency verification is performed in conjunction with time-related information to obtain the expected time to reach the maintenance threshold.

[0014] Maintenance recommendations are generated based on the estimated time to reach the maintenance threshold.

[0015] Furthermore, when obtaining multi-period datasets, this includes:

[0016] The internal and external parameters of the data acquisition equipment are calibrated and synchronized with a unified time reference.

[0017] The visible light image and thermal image are acquired at the same time during each acquisition period.

[0018] A uniform exposure time, sensitivity, aperture value, and white balance mode are set for the visible light images, and a uniform emissivity, background temperature, and reflection compensation are set for the thermal images, with on-site correction performed using a temperature reference surface.

[0019] The imaging parameters include exposure time, ISO, aperture value, lens focal length, focus distance, white balance mode, image resolution, image frame rate, and noise suppression level. The imaging parameters also include thermal imaging emissivity setting, background temperature setting, reflection compensation parameters, and non-uniformity correction status. The environmental parameters include ambient temperature, air humidity, sunlight intensity, and wind speed.

[0020] Furthermore, when establishing a unified spatial coordinate system and fitting a cylindrical surface model, the following steps are included:

[0021] No fewer than three facade control points are set on the facade and used as permanent markers. The scale and direction of the unified spatial coordinates are determined by the relative spacing and included angle of the facade control points.

[0022] The window sill line, corner line, and eaves line are used as auxiliary geometric constraints to define the planar position;

[0023] Stable boundaries of clamp edges, support edges, joint profiles and elbow profiles are extracted within the external riser area. The axial direction and cross-sectional radius of the external riser are estimated based on the geometric fitting method to obtain the cylindrical surface model.

[0024] Furthermore, when obtaining trans-period isotopic images, this includes:

[0025] A baseline for expansion, starting from the structural anchor point, is selected and remains unchanged throughout the data collection period;

[0026] Coarse registration of axial and circumferential directions is completed under unified spatial coordinates, and fine registration is performed under the constraints of cylindrical surface model, limiting axial displacement, circumferential displacement and local deformation to not exceed preset position limits.

[0027] The outer riser area is mapped to the curved surface unfolding plane according to the unfolding baseline, and the relative order of the structural anchor points in the circumferential direction is kept unchanged during the mapping process; the co-alignment is performed in the neighborhood of the structural anchor points, the co-alignment includes boundary shape consistency verification and texture stable region consistency verification, and abnormal areas are removed by occlusion detection rules.

[0028] Cyclic consistency checks are performed across multiple consecutive periods. If the check fails, the process is rolled back and the expansion baseline is redefined. Once the check is met, the cross-period co-positional image is obtained.

[0029] Furthermore, when obtaining comparable images across periods, this includes:

[0030] Determine the baseline acquisition period and use visible light images and thermal images as the radiometric reference;

[0031] Set up a reference gray card and keep it in the same position and orientation in each acquisition period. Perform lens vignetting correction, zone gain correction and white balance unification on the visible light image. Use the reference gray card to perform brightness mapping on the same material area. Detect and mark areas with strong reflection, shadow and raindrop pollution and remove them in the cross-period comparison.

[0032] For thermal imaging images, emissivity is set for steel pipes, coated pipes and insulation layers according to the external pipe material library. Two-point correction is performed using a temperature reference surface, background temperature is recorded and reflection compensation and non-uniformity correction are performed.

[0033] Brightness consistency is verified in the stable area of ​​the external riser component to ensure that the brightness difference of the same component in different acquisition periods does not exceed the preset brightness difference limit, thereby obtaining the cross-period comparable image.

[0034] Furthermore, when obtaining the set of risk point indicators, it includes:

[0035] Based on the comparable images across different periods, candidate regions are retrieved in the gravity direction near the structural anchor point and downstream of the joint; corrosion candidates are generated for regions with coating peeling, abnormal color, or increased surface roughness along the circumferential direction; leakage candidates are generated for regions with stripes along the gravity direction and temperature differences in the thermal imaging images; deformation candidates are generated for regions with offset between the axial and circumferential positions of the external riser; and loosening candidates are generated for regions with increased clamping gap between the clamp and the bracket or increased relative displacement of the hole.

[0036] A local benchmark is established for candidate regions based on stable regions of the same material. Candidate regions that exist or show an increasing trend in adjacent collection periods are retained as risk points. Quantitative indicators are extracted based on the appearance changes, morphological changes, thermal anomalies and geometric deviations. The risk point index set is formed by combining the unified spatial coordinates of the risk points with the quantitative indicators.

[0037] Furthermore, the construction of the spatiotemporal coupling graph includes:

[0038] In the curved surface unfolding plane, axial dripping paths, support connection paths, and wall penetration paths are generated based on the structural anchor points and joint positions. The axial dripping path traces downwards from the structural anchor point or joint along the axial direction of the external riser and in the direction of gravity, with the path direction determined by stain extension and thermal anomaly connectivity. The support connection path connects the center positions of adjacent supports and clamps, with the path direction determined by the component sequence. The wall penetration path extends from the junction of the external riser and the external facade towards the external facade, with the path direction determined by appearance changes and thermal anomaly connectivity. For each risk point on the path, the nearest path position is determined and an initial edge is established, resulting in a spatiotemporal coupling graph containing nodes, edges, and edge weights.

[0039] Furthermore, when obtaining the intertemporal trajectory of risk points, it includes:

[0040] Within the sliding time window, the extension speed and continuity of the stain along the direction of gravity are extracted based on capillary diffusion, the strip shape and downstream connectivity are extracted based on gravity dripping, the area growth rate and boundary expansion rate of the same material region are extracted based on material moisture absorption, and the edge weights are calculated according to the preset weight rules in combination with component constraints, geometric proximity and path connectivity.

[0041] On the spatiotemporal coupling graph, risk points in adjacent acquisition periods are paired. When spatial proximity, path connectivity, similar quantitative indicators with consistent growth trends, and both positive and negative matching are met, cross-period associations are established. When pairing conflicts occur, priority is given to pairings that are connected through three types of paths and have higher edge weights. Trajectories with short-term occlusion are compensated for missing data. Trajectories that do not meet the conditions for multiple consecutive periods are terminated. New trajectories are established for newly emerging risk points that meet the initial conditions, and a unified number is assigned after the association is completed.

[0042] Furthermore, when obtaining the estimated time to reach the maintenance threshold, the following are included:

[0043] Using four failure modes—corrosion, leakage, deformation, and loosening—as the weighting source, and combining index stability, intertemporal consistency, and path connectivity, weights are assigned to various indicators in the risk point index set to obtain the degradation state representation.

[0044] The degradation state representation is trend-fitted based on robust regression, and short-term fluctuations caused by rainfall, sunshine, and temperature cycles are corrected according to time-related information. The fitting results are then subjected to consistency verification, and a re-examination suggestion is output if the consistency verification fails. If the consistency verification passes, the estimated time to reach the maintenance threshold is calculated based on the difference between the degradation state representation and the maintenance threshold, as well as the growth rate of the degradation state representation.

[0045] Furthermore, when generating maintenance recommendations based on the estimated time to reach the maintenance threshold, the process includes:

[0046] Based on the relationship between the estimated time to reach the maintenance threshold and the sliding time window, risk points are classified into three levels: urgent, priority, and routine. A suggested operation time window is determined by combining time-related information. Operation positioning data is output based on the unified spatial coordinates and structural anchor points, including the unified spatial coordinate position, the curved surface unfolding plane position, and the nearest structural anchor point identifier. Based on the spatiotemporal coupling diagram, risk points connected to the target risk point are retrieved along the axial drip path, support connection path, and wall seepage path. Risk points with an estimated time to reach the maintenance threshold within the buffer period are merged to generate a linked maintenance suggestion. A maintenance action list is provided according to the risk point type, including at least one of the following: anti-corrosion surface treatment and coating repair, leakage sealing and diversion treatment, component repositioning and replacement, clamp tightening, and support reinforcement.

[0047] The final output includes cross-period comparison images, a summary of change indicators, the estimated time to reach the maintenance threshold, recommended operation time windows, and a maintenance action list.

[0048] Compared with existing technologies, the beneficial effects of this invention are as follows: By organizing multi-period data and establishing unified spatial coordinates for facade control points and external pipe components, and combining a cylindrical curved surface model to implement inter-period registration and surface unfolding with surface constraints, consistent mapping of the same physical part in different acquisition periods is achieved under the constraints of structural anchor point alignment and cyclic consistency verification; inter-period radiometric standardization and thermal imaging physical correction are completed based on imaging parameters and environmental parameters, reducing the inter-period incomparability caused by differences in viewing angle, illumination, and thermal radiation; a set of risk point indicators is constructed within the surface unfolding plane around appearance changes, morphological changes, thermal anomalies, and geometric deviations, and then axial dripping is used. A spatiotemporal coupling diagram is established using the path, support connection path, and wall penetration path as edges. Edge weights are determined according to capillary diffusion, gravity dripping, and material moisture absorption mechanisms, achieving robust intertemporal correlation and unified numbering of risk points, reducing spurious changes and false correlations. Within a sliding time window, the risk point indicator set is weighted and fused to form a degradation state representation. The estimated time to reach the maintenance threshold is output through trend fitting and time correlation consistency verification, and maintenance suggestions are generated in conjunction with this. This forms a closed loop of data collection—location—standardization—mapping—prediction—maintenance, improving the intertemporal comparability of external risers, the accuracy of risk point location, and the reliability of trend prediction, while reducing false alarms and duplicate inspections. Attached Figure Description

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0050] Figure 1The flowchart illustrates the method for comparing multi-period images of risk points and predicting deterioration trends for external risers provided in this embodiment of the invention. Detailed Implementation

[0051] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0052] In traditional remote inspections of external risers, the risers, being cylindrical curved surfaces, are constantly exposed to wind, rain, and thermal cycles. Multiple image acquisitions are inevitably affected by changes in viewing angle, lighting, and thermal radiation, leading to incomparability between images from different periods. Furthermore, failure evolution exhibits pathological characteristics such as axial dripping, conduction along supports, and penetration into walls, making single-period assessments unable to characterize the physical coupling relationships between the affected components. This problem hinders image-based automated diagnosis from achieving high-frequency, standardized, and comparable remote inspections, thus affecting the accuracy and reliability of predictive maintenance. Maintenance departments are unable to track and analyze trends in risks such as corrosion, leakage, deformation, and loosening.

[0053] For example, in the scenario of inspecting the external pipes of a high-rise residential building in a city, maintenance personnel used drones equipped with visible light cameras and thermal imagers to collect data in multiple phases. Specifically, during the spring collection period, the solar altitude angle is low, resulting in ample sunlight on the east side of the external pipes while the west side shows obvious shadows; during the summer collection period, direct sunlight leads to increased overall brightness and rising ambient temperature. The same leak point appears as a localized wet area in the spring images, but is difficult to identify in the summer images due to the high-temperature evaporation effect; at the same time, loosening points at the support connections propagate downwards along the axis, but a single-phase image can only capture the current state and cannot reflect its evolution path from top to bottom. Furthermore, thermal imaging data is affected by fluctuations in ambient temperature, and thermal anomaly signals show inconsistent changes in different collection periods, resulting in the inability to align risk point identification results across periods. Thus, the incomparability of images across periods causes deviations in risk point location, and the lack of physical coupling relationships makes it impossible to accurately model the failure evolution path, ultimately affecting the scientific and timely nature of maintenance decisions.

[0054] If the above problems are not addressed, the incomparability of cross-period images will lead to a continuous increase in the misjudgment rate of risk points. The lack of failure evolution paths will render the degradation trend prediction model ineffective, and the operation and maintenance department will be unable to achieve dynamic monitoring and quantitative assessment of external riser risks. Specifically, predictive maintenance systems will struggle to distinguish between actual degradation and environmental interference, resulting in improper allocation of maintenance resources, increasing the probability of sudden facility failures, and consequently affecting the safe operation and service life of urban infrastructure.

[0055] For this, please refer to Figure 1 As shown, this application proposes a method for multi-period image comparison and degradation trend prediction of risk points of external risers, including:

[0056] S100: Acquires multiple images and records acquisition time, acquisition posture, imaging parameters and environmental parameters, establishes a unified time reference and sliding time window, and obtains multiple datasets. The multiple images include multiple visible light images and thermal images of the external riser.

[0057] S200: Based on the facade control points and external pipe components, a unified spatial coordinate system is established and a cylindrical surface model is fitted. Imaging distortion correction and perspective correction are performed on multi-period datasets to obtain standardized images.

[0058] S300: Under the constraints of the cylindrical surface model, standardized images are registered across periods, and the external riser area is unfolded into a surface unfolding plane. The structural anchor points are used for co-alignment and cyclic consistency verification to obtain cross-period co-aligned images.

[0059] S400: Perform cross-period radiometric normalization on cross-period co-location images based on imaging parameters and environmental parameters, and perform physical correction on thermal imaging images to obtain cross-period comparable images.

[0060] S500: In the curved surface unfolding plane, risk points are generated for corrosion, leakage, deformation and loosening. Quantitative indicators are extracted around appearance changes, morphological changes, thermal anomalies and geometric deviations to obtain a set of risk point indicators.

[0061] S600: A spatiotemporal coupling diagram is constructed with risk points as nodes and axial dripping paths, support connection paths and wall penetration paths as edges. Edge weights are determined based on capillary diffusion, gravity dripping and material moisture absorption mechanisms. Risk points are correlated across periods and assigned a unified number to obtain the cross-period trajectory of risk points.

[0062] S700: Within a sliding time window, the risk point indicator set is weighted and fused to form a degradation state representation. The degradation state representation is then trend-fitted and combined with time-related information for consistency verification to obtain the expected time to reach the maintenance threshold.

[0063] S800: Generates maintenance recommendations based on the estimated time to reach the maintenance threshold.

[0064] This application relates to a method for multi-period image comparison and degradation trend prediction of external risers. It addresses the problem that external risers, exposed to wind, rain, and thermal cycling for extended periods, are prone to corrosion, leakage, deformation, and loosening. Existing technologies suffer from incomparability of images across different periods due to variations in viewing angle, illumination, and thermal radiation. Furthermore, single-period analysis struggles to characterize the pathogenesis of failure evolution, including axial dripping, support conduction, and wall penetration. In practical applications, establishing a unified spatial coordinate system and fitting a cylindrical surface model refers to determining the spatial representation of the external riser through geometric constraints. This can be achieved using surface reconstruction methods based on LiDAR point clouds. For example, iterative nearest-point algorithms can be used to register multi-view scan data and fit cylindrical parameters, or structured light scanning can be used to obtain three-dimensional point clouds, and principal component analysis can be employed to estimate the axial direction and cross-sectional radius. The primary goal is to achieve accurate representation of the external riser's geometry and eliminate geometric distortion caused by lens distortion and viewing angle differences. Furthermore, cross-period registration of standardized images under the constraints of a cylindrical surface model refers to the process of mapping images from different periods to a unified geometric framework. This can be achieved using deep learning-based feature matching methods, such as using convolutional neural networks to extract stable texture feature points in the images and calculate the spatial transformation matrix, or using phase correlation algorithms to estimate translation and rotation parameters in the frequency domain. The main purpose is to achieve accurate alignment of the external riser area at different times, ensuring that the same physical part remains consistent in position across different periods. Cross-period radiometric standardization of co-located images based on imaging and environmental parameters refers to the process of eliminating differences in illumination and thermal radiation to make the images comparable in physical dimensions. This can be achieved using radiometric correction methods based on physical models, such as compensating for radiation deviations caused by changes in ambient temperature and solar radiation intensity according to the atmospheric radiative transfer equation, or correcting the spectral response of different surfaces by establishing a database of material reflectance characteristics. The main purpose is to eliminate the influence of environmental parameter fluctuations on image brightness and temperature values, making visible light images and thermal images comparable under cross-period conditions. Specifically, constructing a spatiotemporal coupled graph with risk points as nodes refers to the process of abstracting risk points and their evolution paths into a graph structure. This can be achieved using path generation methods based on fluid dynamics models. For example, the moisture diffusion rate during the material's moisture absorption process can be calculated based on Darcy's law to determine the direction of the wall's penetration path, or the liquid dripping trajectory under gravity can be simulated by particle tracking to generate an axial dripping path. The main purpose is to characterize the path dependence driven by physical mechanisms during the failure evolution process.As a preferred implementation, determining edge weights based on capillary diffusion, gravitational dripping, and material moisture absorption mechanisms refers to the calculation process of quantifying the correlation strength between risk points. This can be achieved using empirical formulas combined with field measurement data. For example, the Washburn equation can be used to calculate the capillary rise height in porous media to determine the edge weights between leakage risk points, or Stokes' law can be used to estimate the velocity of droplets under gravity to characterize the connectivity of the axial dripping path. The main purpose is to reflect the degree of influence of physical failure mechanisms on the intertemporal correlation of risk points, ensuring that the trajectory construction conforms to the actual evolution law. Therefore, this application achieves geometric distortion correction by establishing a unified spatial coordinate system and fitting a cylindrical surface model. It combines intertemporal registration and surface unfolding to solve the intertemporal alignment problem of the same physical part, and constructs a spatiotemporal coupling diagram based on edge weights driven by physical mechanisms. This solves the problem of intertemporal incomparability of multi-period images of external riser risk points caused by changes in viewing angle, illumination, and thermal radiation. Simultaneously, it characterizes the physical coupling relationship along a specific path in the failure evolution, ultimately achieving reliable prediction of intertemporal correlation and deterioration trends of risk points.

[0065] When implementing the method of multi-phase image comparison and degradation trend prediction of the risk points of the external riser, the visible light image and thermal image of the external riser are first acquired through multi-phase image acquisition equipment. At the same time, the acquisition time, acquisition posture, imaging parameters and environmental parameters are recorded. The acquisition time is synchronized with a unified time reference, and the sliding time window is set to three months according to the operation and maintenance cycle, thereby forming a multi-phase dataset with time series correlation. Based on the control points of the permanent markings on the facade and the geometric features of the external riser components, a unified spatial coordinate system was established. Specifically, three reflective markers were used as permanent markers for the facade control points. The relative spacing and angle between these markers were determined by total station measurements to determine the coordinate scale and direction. Window sill lines, corner lines, and eaves lines served as auxiliary geometric constraints to define the planar position. The edges of clamps, supports, joints, and bends within the external riser area were extracted as stable boundaries. The axial direction and cross-sectional radius of the external riser were estimated using a geometric fitting method, and a cylindrical surface model was then fitted. Under the constraints of this model, lens distortion correction and perspective correction were performed on the multi-period datasets to ensure that the projections of the facade control points in each period's imagery coincided with the unified spatial coordinate system, thereby outputting standardized images. Furthermore, under the constraints of the cylindrical surface model, standardized images are registered across periods. The edge of the clamp is selected as the structural anchor point and the unfolding baseline is set. After coarse registration in the axial and circumferential directions under unified spatial coordinates, the fine registration process limits the axial displacement to no more than 5 mm, the circumferential displacement to no more than 2 degrees, and the local deformation to within the preset limits. The outer riser area is mapped to the surface unfolding plane while keeping the relative order of the structural anchor points in the circumferential direction unchanged. In the neighborhood of the structural anchor points, the alignment is achieved through boundary shape consistency verification and texture stable region consistency verification. Cyclic consistency verification is performed in multiple consecutive periods to ensure the reliability of registration, thereby obtaining cross-period co-position images. Based on imaging and environmental parameters, intertemporal radiometric standardization uses a reference gray card for visible light images to correct vignetting, perform zone gain correction, and unify white balance. The reference gray card is placed in the same position on the external riser and maintains a fixed orientation. Brightness mapping of areas with the same material is completed based on the reflectivity of the gray card. For thermal imaging images, the emissivity is set to 0.85 for steel pipes, 0.92 for coated pipes, and 0.95 for insulation layers according to the external riser material library. Two-point correction is performed using a temperature reference surface to compensate for background temperature. The physical correction process eliminates areas with strong reflections, shadows, and raindrop contamination. Ultimately, the brightness difference of the same component in different acquisition periods is controlled within the preset limit, obtaining comparable intertemporal images.

[0066] Specifically, within the curved surface unfolding plane, risk point generation focuses on the gravity direction of the structural anchor point neighborhood and downstream of the joint. For example, corrosion candidates are generated for areas with coating peeling along the circumferential direction within 10 cm downstream of the clamp; leakage candidates are generated for areas in thermal imaging where the temperature difference exceeds 2 degrees Celsius and forms strips along the gravity direction; areas where the axial position of the external riser shifts by more than 3 mm are marked as deformation candidates; and areas where the clamping gap increases by more than 1 mm are generated as loosening candidates. Candidate areas are screened using stable areas of the same material as the local benchmark, retaining only areas that exist in adjacent acquisition periods or show an increasing trend as risk points. Quantitative indicators are extracted around appearance changes, morphological changes, thermal anomalies, and geometric deviations. Color differences are calculated using the HSV color space, thermal anomaly amplitude is quantified based on temperature difference, and axial position changes are converted using pixel displacement, forming a risk point indicator set that includes unified spatial coordinates and quantitative indicators for risk points. When constructing a spatiotemporal coupling diagram with risk points as nodes, the axial dripping path traces downwards along the axial direction of gravity from the structural anchor point along the external riser. The path direction is determined by the connection between stain extension and thermal anomaly. The support connection path connects the center positions of adjacent support components. The wall penetration path extends from the junction of the external riser and the external facade to the external facade. The path direction is jointly determined by the connection between appearance changes and thermal anomaly. The edge weights are calculated based on the capillary diffusion mechanism to extract the stain extension rate, the gravity dripping mechanism to analyze the strip morphology, and the material moisture absorption mechanism to calculate the area growth rate. Combined with component constraints and geometric proximity, the weights are applied according to preset rules. The risk points are paired across periods based on spatial proximity, path connectivity, similar quantitative indicators, and consistent growth trends. Short-term occlusion trajectories are assigned a unified number after missing measurement compensation, forming the risk point cross-period trajectory. Within the sliding time window, the risk point index set is weighted and fused according to the weights of four failure modes: corrosion, leakage, deformation, and loosening. The degradation state is fitted through robust regression trend, and short-term fluctuations caused by rainfall, sunshine, and temperature cycles are corrected using time correlation information. Minimum period verification, monotonicity verification, and window consistency verification ensure the reliability of the fit. The difference and growth rate between the degradation state and the maintenance threshold are used to calculate the expected time to reach the maintenance threshold. The final maintenance recommendation is divided into emergency, priority, and routine levels based on this time. The operation positioning data outputs a unified spatial coordinate position and the nearest structural anchor point identifier. The linked maintenance recommendation is generated by retrieving associated risk points on the axial drip path, support connection path, and wall seepage path through the spatiotemporal coupling diagram. The maintenance action list includes specific operations such as anti-corrosion surface treatment, leakage sealing, component repositioning, and clamp tightening.

[0067] By establishing a unified spatial coordinate system and cylindrical surface model, geometric distortion caused by changes in perspective is eliminated. Intertemporal radiation standardization offsets image deviations caused by fluctuations in illumination and thermal radiation, making multi-period images comparable in physical dimensions. The construction of the spatiotemporal coupling diagram transforms capillary diffusion, gravitational dripping, and material moisture absorption mechanisms into edge weights, accurately depicting the path evolution of failure along axial dripping, support conduction, and wall penetration, and achieving precise and unified numbering of risk points across periods. The trend fitting of degradation state representation, combined with time correlation information, corrects short-term interference, ensuring the reliability of degradation trend prediction. This solves the core problems of incomparability of multi-period images of external riser risk points and lack of failure evolution path modeling, providing operable technical support for predictive maintenance.

[0068] Traditional methods for comparing and predicting the degradation trend of risk points in external risers suffer from incomparability due to inconsistent acquisition conditions. Specifically, external risers are exposed to complex environments for extended periods. If equipment calibration, acquisition time periods, imaging parameters, and environmental parameters are not standardized when acquiring multiple images, image geometric distortion, brightness differences, and temperature measurement errors can occur due to perspective deviations, illumination fluctuations, thermal radiation interference, and time reference misalignments. This makes it difficult to achieve physical consistency in cross-period registration and radiation standardization, thereby affecting the reliability of risk point index extraction and degradation trend prediction.

[0069] In this regard, this application further proposes methods for obtaining multi-period datasets, including:

[0070] The internal and external parameters of the data acquisition equipment are calibrated and synchronized with a unified time reference.

[0071] Acquire visible light images and thermal images during the same time period in each acquisition phase;

[0072] For visible light images, set uniform exposure time, sensitivity, aperture value and white balance mode; for thermal images, set uniform emissivity, background temperature and reflection compensation; and perform on-site correction using a temperature reference surface.

[0073] Imaging parameters include exposure time, ISO, aperture, lens focal length, focus distance, white balance mode, image resolution, image frame rate, and noise suppression level. Imaging parameters also include thermal imaging emissivity settings, background temperature settings, reflection compensation parameters, and non-uniformity correction status. Environmental parameters include ambient temperature, air humidity, sunlight intensity, and wind speed.

[0074] In practical applications, intrinsic parameter calibration refers to calibrating the distortion parameters and optical characteristics of the acquisition device's lens. This can be achieved using a checkerboard calibration board combined with the Zhang Zhengyou calibration method or feature point-based self-calibration techniques, aiming to eliminate geometric distortion caused by lens distortion. Extrinsic parameter calibration can be understood as determining the pose relationship of the acquisition device in a unified spatial coordinate system. This can be accomplished using a fixed reference point array or a laser tracker, aiming to establish a precise mapping between the device's coordinate system and physical space. Unified time reference synchronization specifically involves precisely aligning the timestamps of multiple acquisition devices, for example, achieving microsecond-level synchronization through GPS timing modules or network time protocols. Its purpose is to ensure the reliability of the temporal sequence of images from different periods, providing a foundation for constructing sliding time windows. Acquisition within the same time period can be understood as performing image acquisition operations within a fixed daily time window, such as choosing two hours before and after noon to avoid drastic changes in the angle of sunlight. Its purpose is to reduce fluctuations in visible light image brightness and thermal imaging temperature drift caused by differences in natural lighting conditions. Standardized exposure time and other visible light parameter settings specifically fix optical parameters to preset values, avoiding variables introduced by the equipment's automatic adjustment mechanism. The aim is to maintain stable appearance characteristics such as coating color and surface roughness across different image periods. Standardized emissivity and other thermal imaging parameter settings refer to setting preset thermal radiation characteristic parameters based on the material type of the external pipe. For example, different emissivity thresholds are used for steel pipes and coated pipes. The purpose is to eliminate atmospheric absorption and reflection interference, ensuring the physical accuracy of temperature measurements. On-site temperature reference surface calibration specifically utilizes a known temperature reference surface for real-time compensation, such as using a constant-temperature blackbody reference source or a standard temperature patch. The purpose is to correct for the dynamic effects of ambient temperature and humidity on thermal imaging data.

[0075] By systematically constraining the physical benchmarks and operational procedures of the acquisition process, the geometric structure of the images faithfully reflects the true morphology of the external riser, avoiding cross-period registration failures caused by equipment deviations. Simultaneously, it avoids interference from changes in natural lighting on image features, preserving coupled information on appearance and thermal condition, laying a data foundation for multimodal risk identification. Furthermore, fixed optical parameters suppress ambient light fluctuations, and thermal radiation measurements are corrected based on material characteristics, ensuring that image differences only reflect the true deterioration state of the external riser. Clearly defined imaging and environmental parameters provide traceable evidence for cross-period radiation standardization, ensuring that processing is based on consistent physical benchmarks for comparison. These measures collectively construct a high-fidelity multi-period dataset, fundamentally guaranteeing the generation quality of cross-period co-location images and the physical rationality of risk point trajectory prediction.

[0076] As a specific implementation method, the scheme of this application is implemented as follows: The acquisition equipment adopts a combination of an industrial-grade visible light camera and an infrared thermal imager. The internal parameter calibration is completed in the laboratory using a standard checkerboard calibration board, and the external parameter calibration is carried out on-site using an array of permanent marker points on the exterior facade. Time synchronization is achieved by a built-in GPS module to ensure that the timestamp error of each period of image is less than 10 milliseconds. The acquisition period is limited to 11:00 to 13:00 every day to reduce the impact of changes in the angle of sunlight. The visible light image parameters are fixed as an exposure time of 1 / 500 second, an ISO sensitivity of 100, an aperture value of f / 8, and a manual white balance mode. The thermal imaging image parameters are set according to the material library with emissivity (0.85 for steel pipes and 0.92 for coated pipes). The background temperature is taken from the actual measured value of the environment, and a constant-temperature blackbody reference source is placed near the pipe wall for on-site calibration. Environmental parameters are recorded synchronously through a portable weather station, including a temperature and humidity sensor and a sunmeter.

[0077] The above technical solution eliminates image geometric distortion and brightness differences caused by inconsistent acquisition conditions, ensures the physical consistency of images across periods, and provides a reliable data foundation for risk point index extraction and degradation trend prediction, thereby improving the accuracy of external riser risk identification and the reliability of prediction results.

[0078] Traditional existing methods for correcting external pipe images suffer from abnormal observation interference caused by obstructions, strong reflections, and moving targets when establishing unified spatial coordinates and fitting cylindrical surface models. Furthermore, they lack a dynamic verification mechanism for the correction results, resulting in insufficient accuracy of the unified spatial coordinates and cylindrical surface models. This leads to cross-period image registration deviations and affects the reliability of risk point trajectory tracking.

[0079] In response, this application further proposes the following steps for establishing a unified spatial coordinate system and fitting a cylindrical surface model: setting no fewer than three facade control points as permanent markers on the facade; determining the scale and direction of the unified spatial coordinate system by the relative spacing and included angle of the facade control points; using window sill lines, corner lines, and eaves lines as auxiliary geometric constraints to define the planar position; extracting stable boundaries of clamp edges, support edges, joint profiles, and elbow profiles within the external riser area; and estimating the axial direction and cross-sectional radius of the external riser based on a geometric fitting method to obtain a cylindrical surface model. During the fitting process, a random consistency elimination method is used to exclude abnormal observations caused by occlusions, strong reflections, and moving targets. Vegetation, scaffolding, and temporary marker areas are marked and eliminated according to occlusion detection rules. Lens distortion correction is performed on multi-period images based on the intrinsic and extrinsic parameters of the acquisition equipment, and perspective correction is implemented under unified spatial coordinate constraints to ensure that the projection of the facade control points in each period of images coincides with the unified spatial coordinates. Three-phase cyclic consistency verification and multi-component consistency verification are used as passing conditions. When the deviation exceeds the preset error limit, the system automatically backs down and re-estimates the cylindrical surface model or updates the perspective correction parameters. Standardized images are output under the condition of meeting the consistency verification and error limit, which are used for cross-period registration and surface unfolding.

[0080] Among them, facade control points refer to permanent reference markers set on the building facade. These can be implemented using physical markers such as reflective patches, metal embedded markers, or laser-etched dot matrices. Their purpose is to provide a stable geometric benchmark and avoid coordinate drift caused by changes in the acquisition posture. Window sill lines, corner lines, and eaves lines can be understood as inherent structural feature lines of the building. They can be extracted using edge detection algorithms or deep learning segmentation models. Their purpose is to enhance the stability of the coordinate system by utilizing the inherent geometric features of the building and reduce the impact of errors from a single control point on the overall positioning. Clamp edges, support edges, joint contours, and elbow contours refer to stable geometric boundaries in the external pipe structure. They can be extracted using Hough transform or active contour models. Their purpose is to select structural features as the fitting basis and avoid interference from surface coating changes or temporary stains on model accuracy. The random consistency elimination method is specifically a robust statistical estimation technique, which can be implemented using the RANSAC algorithm or the M-estimator. Its purpose is to eliminate abnormal observation points caused by non-structural factors through iterative screening. Occlusion detection rules can be understood as a set of logical rules for identifying temporary obstacles. These rules can be constructed based on region growing algorithms or semantic segmentation networks, aiming to mark and remove temporary interference areas such as vegetation and scaffolding. Three-phase cyclic consistency verification refers to a closed-loop verification mechanism in the time dimension, which can be achieved through bidirectional registration error comparison of three phases of imagery. Its purpose is to ensure the continuity of the correction results in the time series. Multi-component consistency verification refers to a cross-validation mechanism in the structural dimension, which can be achieved through comparison of independent correction results of different external riser components. Its purpose is to verify the reliability of the correction results in the spatial structure.

[0081] Specifically, this application's solution establishes a dual-benchmark system by setting permanently marked facade control points on the facade, determining the scale and direction of a unified spatial coordinate system based on the relative spacing and angle of these control points, and using window sill lines, corner lines, and eaves lines as auxiliary geometric constraints. Based on this, stable boundaries such as clamp edges, support edges, joint contours, and elbow contours in the riser structure are used as the fitting basis. Anomalies are filtered using a random consistency elimination method, and temporary interference areas are eliminated through occlusion detection rules, ensuring that the cylindrical surface model is based solely on the riser body structure. Subsequently, lens distortion correction is performed based on the intrinsic and extrinsic parameters of the acquisition equipment, and perspective correction is implemented under unified spatial coordinate constraints, ensuring that the projected positions of the facade control points strictly coincide. Finally, a dual verification mechanism of three-phase cyclic consistency verification and multi-component consistency verification is used. When the deviation exceeds the preset error limit, automatic backtracking adjustment is implemented, and standardized images are output only when all verification conditions are met, thus forming a complete correction closed loop. This process achieves an organic unity of the stability of the spatial benchmark, the robustness of the model fitting, and the verifiability of the calibration results, thus solving the accuracy problems caused by environmental interference and lack of verification.

[0082] As a specific implementation method, the solution of this application is implemented as follows: Stainless steel reflective markers are permanently installed on the building facade as facade control points; a laser scanner is used to obtain the spatial positions of window sills, corner lines, and eaves lines; the Canny edge detection algorithm is used to extract the geometric boundaries of clamps and supports in the external pipe area, and the cross-sectional radius is estimated by combining Hough circle transformation; during the fitting process, the RANSAC algorithm is used to iteratively remove outliers, and the vegetation and scaffolding areas are segmented based on the U-Net network; distortion correction is performed according to the calibrated camera intrinsic parameter matrix and extrinsic parameter rotation and translation vector, and the control point projections are made to coincide through the perspective transformation matrix; the cyclic registration error threshold of the three phases of images is set to 2 pixels, and the correction deviations of the three types of components, clamps, supports, and joints, are required to be lower than the preset limit. If the standard is not met, the cylindrical surface model parameters are automatically adjusted and recalculated.

[0083] Through the above technical solutions, this application eliminates the interference of occlusions, strong reflections and moving targets on model fitting, establishes a correction mechanism with dynamic verification capabilities, improves the accuracy of unified spatial coordinates and cylindrical surface models, and thus ensures the accuracy of cross-period image registration.

[0084] Specifically, in some of the embodiments described above in this application, cross-period registration and surface unfolding are proposed to obtain cross-period co-position images. However, in the implementation process, due to the geometric characteristics of the cylindrical surface of the outer riser and external interference, if the unfolding baseline shifts during each acquisition period, the unfolded surface plane will not be able to align with the same physical position. Coarse registration and fine registration lack displacement constraints and are easily affected by changes in viewing angle and local deformation, resulting in cumulative errors. The structural anchor points are disordered in sequence during circumferential mapping, causing geometric relationship distortion. Co-position alignment is not combined with boundary shape and texture stability verification, which is easily affected by occlusion and texture changes. The lack of cyclic consistency verification makes the multi-period registration results unreliable, thus making it difficult to support the accurate correlation of risk point cross-period trajectories.

[0085] In this regard, this application further proposes steps for obtaining intertemporal isotopic images, including:

[0086] A baseline for expansion, starting from the structural anchor point, is selected and remains unchanged throughout the data collection period;

[0087] Coarse registration of axial and circumferential directions is completed under unified spatial coordinates, and fine registration is performed under the constraints of cylindrical surface model, limiting axial displacement, circumferential displacement and local deformation to not exceed preset position limits.

[0088] The outer riser area is mapped to the curved surface development plane based on the unfolding baseline, and the relative order of the structural anchor points in the circumferential direction is kept unchanged during the mapping process; the co-alignment is performed in the neighborhood of the structural anchor points, including boundary shape consistency verification and texture stable region consistency verification, and abnormal areas are removed by occlusion detection rules.

[0089] Cyclic consistency checks are performed across multiple consecutive periods. If the check fails, the process is rolled back and the baseline is redefined. Once the check is met, cross-period co-positional images are obtained.

[0090] Among them, the unfolding baseline refers to the reference line starting from the structural anchor point. It can use stable geometric feature points on the outer riser as the reference, such as the edge of the clamp or the connection point of the bracket. Its purpose is to fix the starting position of the surface unfolding and avoid reference drift caused by differences in the acquisition posture. Coarse registration and fine registration refer to the registration process in stages. Coarse registration can use a fast matching algorithm based on feature points to achieve overall contour alignment. Fine registration can use an optimization algorithm based on the cylindrical surface model to achieve local accurate matching. Its purpose is to first quickly reduce the computational complexity and then accurately suppress perspective distortion caused by changes in viewing angle and mismatch caused by local occlusion. Surface unfolding mapping refers to the mapping of the cylindrical surface outer riser. The process of converting the riser region into a planar representation can be achieved using equidistant or conformal unfolding methods. Its purpose is to faithfully preserve the circumferential geometric topology of the outer riser and avoid misalignment of structural anchor points. Co-alignment refers to high-precision matching within the neighborhood of structural anchor points. This can be achieved by combining boundary shape feature matching with texture region correlation analysis. Its purpose is to improve the anti-interference capability of matching through geometric contour stability verification and surface feature consistency confirmation. Cyclic consistency verification refers to a closed-loop verification mechanism between multiple image phases. This can be achieved using forward and backward matching consistency checks. Its purpose is to detect cumulative registration errors and dynamically adjust baseline parameters to eliminate deviations.

[0091] Specifically, the proposed solution achieves accurate cross-period co-location of external riser images across multiple periods through a collaborative mechanism of fixed unfolding baseline, phased registration, order-preserving mapping, multimodal alignment, and closed-loop verification. First, the unfolding baseline, starting from the structural anchor points, remains unchanged throughout each acquisition period, ensuring consistency in the initial position of the surface unfolding. Then, coarse registration is performed axially and circumferentially under a unified spatial coordinate system to quickly align the overall contour. Fine registration is then performed based on cylindrical surface model constraints, limiting the displacement range and suppressing the effects of viewpoint changes and local deformation. Next, the external riser region is mapped to the surface unfolding plane according to the unfolding baseline, while maintaining the relative order of the structural anchor points in the circumferential direction to preserve the geometric topology. Then, co-location alignment is achieved through dual verification of boundary shape and texture stability within the neighborhood of the structural anchor points, and occlusion anomalies are eliminated. Finally, through continuous multi-period cyclic consistency verification, the unfolding baseline is dynamically adjusted until the consistency requirements are met, thereby obtaining high-precision cross-period co-location images.

[0092] As a specific implementation method, the solution of this application is implemented as follows: When acquiring images of the external riser, the edge of the clamp is selected as the structural anchor point, and the unfolding baseline is set with the anchor point as the starting point; under a unified spatial coordinate system, coarse registration in the axial and circumferential directions is first completed by feature point matching, and then fine registration is performed based on the cylindrical surface model to limit the displacement within the preset position limit; when mapping the external riser area to the surface unfolding plane, it is ensured that the arrangement order of all structural anchor points in the circumferential direction is consistent with the physical position; within the neighborhood of the structural anchor point, the boundary shape consistency is verified by edge detection, and the stability is confirmed by local texture correlation analysis, and areas blocked by vegetation are automatically removed; cyclic consistency verification is performed between consecutive images, and when the matching error exceeds the threshold, the unfolding baseline is re-determined until consistent cross-period co-position images are obtained.

[0093] The above solution solves the problem of physical misalignment caused by the offset of the unfolding baseline, avoids the cumulative error in coarse and fine registration, prevents geometric distortion caused by disordered structural anchor point sequence, improves the anti-interference ability of co-alignment, and ensures the reliability of multi-period registration results, thus providing a spatial consistency basis for the accurate correlation of risk point cross-period trajectories.

[0094] In some of the embodiments described above in this application, it is proposed to obtain comparable images across periods to eliminate radiation differences between images from multiple periods. However, in the process of implementation, due to the lack of specific implementation of radiation benchmark setting, fine correction of visible light images, and physical correction methods for thermal imaging, the brightness of the images is inconsistent due to the influence of ambient light fluctuations, material characteristics differences, and thermal radiation interference. This leads to misjudgments during cross-period comparison and makes it impossible to accurately identify the real changes in risk points.

[0095] In this regard, this application further proposes that when obtaining comparable images across different periods, the following steps should be taken:

[0096] Determine the baseline acquisition period and use visible light images and thermal images as the radiometric reference;

[0097] Set up a reference gray card and keep it in the same position and orientation in each acquisition period. Perform lens vignetting correction, zone gain correction and white balance unification on visible light images. Use the reference gray card to perform brightness mapping on areas of the same material. Detect and mark areas with strong reflection, shadow and raindrop contamination and remove them in cross-period comparison.

[0098] For thermal imaging images, emissivity is set for steel pipes, coated pipes and insulation layers according to the external pipe material library. Two-point correction is performed using a temperature reference surface, background temperature is recorded and reflection compensation and non-uniformity correction are performed.

[0099] Brightness consistency is verified in stable areas of external riser components to ensure that the brightness difference of the same component in different acquisition periods does not exceed the preset brightness difference limit, thus obtaining comparable images across periods.

[0100] In practical applications, a radiation reference refers to a technical benchmark used to establish a unified radiation scale reference. It can be achieved by using visible light and thermal images from the reference acquisition period as absolute physical dimension calibration sources. The purpose is to avoid error accumulation in multi-period relative calibration and ensure that all images are calibrated to consistent radiation physical dimensions. A reference gray card can be understood as a physical reference object with known reflectivity. It can be implemented using a standard grayscale calibration plate or a ceramic calibration sheet with stable reflectivity. Its purpose is to provide a correlation benchmark for the inherent reflectivity characteristics of the material in visible light images, ensuring the accuracy of brightness mapping across periods. Specifically, lens vignetting correction refers to a technique for compensating for edge attenuation in the optical system. It can be achieved using mathematical correction based on the lens distortion model or pixel-level gain matrix adjustment based on the calibration image. Its purpose is to eliminate image edge brightness attenuation caused by lens optical characteristics. In practical applications, two-point calibration can be understood as a linear response calibration method for thermal imaging equipment. It can be achieved by using two-point reference sources with known temperatures for slope and intercept calibration. Its purpose is to improve the absolute temperature measurement accuracy of thermal imaging and avoid temperature distortion caused by sensor nonlinearity.

[0101] Specifically, the proposed solution achieves comparable image acquisition across different periods by constructing a closed-loop radiation calibration system: First, a baseline acquisition period is determined as the radiation baseline, providing a unified radiation scale reference for all images. Then, a reference gray card is used to maintain a fixed position and orientation across acquisition periods. Lens vignetting correction is performed on visible light images to compensate for optical attenuation, zone gain correction is applied to eliminate illumination inhomogeneity, and white balance is unified to eliminate color temperature drift. Brightness mapping of areas of the same material is also performed using the reference gray card to ensure consistent material radiation response. Simultaneously, environmental interference areas are detected and marked for removal. For thermal imaging images, emissivity is set for different materials based on an external riser material library. Two-point correction is performed using a temperature reference surface to improve temperature accuracy, and background temperature is recorded for reflection compensation and non-uniformity correction to eliminate environmental reflected heat source interference. Finally, brightness consistency verification of stable areas of external riser components verifies the correction effect and ensures comparability across periods. This solution systematically solves the incomparability problem caused by environmental and equipment factors in multiple image periods through the coordinated operation of radiation baseline setting, fine visible light correction, thermal image physical correction, and consistency verification.

[0102] As a specific implementation method, the scheme of this application is implemented as follows: In a certain inspection of the external riser, the first collection during a sunny spring day is selected as the baseline collection period, and the visible light image and thermal imaging image of that period are used as the radiation reference; during each collection period, a standard grayscale calibration plate is installed on a fixed bracket near the external riser to ensure that its position and orientation are completely consistent with the baseline period; for the visible light image, a polynomial model based on lens calibration parameters is used to correct the lens vignetting, the image is divided into grid areas to perform partition gain correction, and uniformly set to daylight white balance mode; the grayscale calibration plate is used to perform brightness mapping on the steel pipe area, and the strongly reflective areas in the image are marked as invalid data; for thermal imaging, the emissivity is set for the steel pipe according to the material library, and the corresponding emissivity is set for the coated pipe, a constant temperature metal plate is used as a temperature reference surface for two-point correction, the ambient background temperature is recorded and reflection compensation is performed; finally, the clamp area is selected as a stabilizing component, and its brightness difference is checked to see if it is lower than the preset threshold. After passing the check, comparable images across periods are output.

[0103] Through the above solution, this application eliminates the radiation differences between multiple images, ensuring that the image brightness of the external riser remains consistent in different acquisition periods. This avoids misjudgments caused by fluctuations in ambient light, differences in material properties, and thermal radiation interference, thereby enabling accurate identification of the true changes in risk points.

[0104] In some of the embodiments described above in this application, a set of risk point indicators is proposed to quantify the deterioration status of risk points in external risers. However, in the process of implementation, candidate areas in comparable images across periods are easily affected by temporary environmental interference, material differences, or short-term fluctuations, which may lead to the misjudgment of non-deterioration changes (such as temporary stains, residual light, or temporary shading) as real risk points. Furthermore, there is a lack of refined quantitative basis for the four types of failure modes: corrosion, leakage, deformation, and loosening, making it difficult to accurately distinguish the physical nature and deterioration trend of risk points.

[0105] In this regard, this application further proposes the following steps for creating a risk point indicator set:

[0106] Based on comparable images across different periods, candidate regions are searched in the gravity direction near the structural anchor point and downstream of the joint; corrosion candidates are generated for regions with coating peeling, abnormal color, or increased surface roughness along the circumferential direction; leakage candidates are generated for regions with stripes along the gravity direction and temperature differences in the thermal imaging; deformation candidates are generated for regions with offset between the axial and circumferential positions of the external riser; and loosening candidates are generated for regions with increased clamping gap between the clamp and the bracket or increased relative displacement of the hole.

[0107] Local benchmarks were established for candidate areas based on stable regions of the same material. Candidate areas that existed or showed an increasing trend in adjacent collection periods were retained as risk points. Quantitative indicators were extracted based on appearance changes, morphological changes, thermal anomalies, and geometric deviations. A risk point indicator set was formed by combining the unified spatial coordinates of the risk points with the quantitative indicators. The quantitative indicators included color difference, brightness difference, surface roughness change, the length of the stain along the direction of gravity, the length of the boundary gap, the strip length ratio, the amplitude of thermal anomalies, the duration of thermal anomalies, the proportion of anomaly areas, the amount of axial position change, the amount of circumferential position change, roundness deviation, the amount of clamping gap change, and the amount of relative displacement of the support hole position.

[0108] In practical applications, the structural anchor point neighborhood refers to the local area centered on the structural anchor point. This can be achieved using a fixed-radius circular area or an annular area determined according to the proportion of the external riser diameter, aiming to provide a stable reference position and reduce spatial positioning errors. The gravity direction downstream of the joint can be understood as the area along the external riser axis and in the direction of gravity. This can be determined by the projection direction of the gravity vector onto the external riser axis, aiming to conform to the law of leakage and other risks developing along physical paths. Coating peeling, abnormal color, or increased surface roughness refer to changes in the physical or optical properties of the coating on the external riser surface. This can be achieved by detecting surface texture changes through image segmentation algorithms or identifying color differences through multispectral analysis, aiming to accurately characterize the corrosion and deterioration state. A stripe forming along the gravity direction with a temperature difference in the thermal imaging image can be understood as an area with stripe morphological characteristics and accompanied by temperature anomalies. This can be achieved by detecting the stripe structure through morphological analysis combined with thermal imaging temperature field analysis, aiming to use dual features to pinpoint the actual leakage. External riser axial position... The offset of the circumferential position refers to the change in the geometric position of the external riser. This can be achieved by calculating the displacement vector through feature point matching or by assessing the degree of deformation through surface fitting, with the aim of directly relating to the physical manifestation of geometric deformation. The increase in the clamping gap between the clamp and the bracket or the increase in the relative displacement of the hole position can be understood as the change in the relative position of the mechanical connection components. This can be achieved by measuring the gap width through edge detection or by calculating the displacement through feature point matching, with the aim of focusing on the quantitative threshold of mechanical connection failure. Establishing a local benchmark in a stable region of the same material means selecting a stable region of the same material as a reference. This can be achieved by identifying homogeneous regions through material classification algorithms and calculating their average optical properties, with the aim of eliminating interference from global illumination or material differences. Retaining candidate regions that exist or show an increasing trend in adjacent acquisition periods can be understood as candidate regions with temporal continuity. This can be achieved by detecting the trend of change through time series analysis or by screening persistent regions through threshold comparison, with the aim of filtering temporary noise and ensuring that only effective risk points with deterioration evolution characteristics are retained.

[0109] Specifically, the proposed solution first retrieves candidate regions in the vicinity of structural anchor points and downstream of joints along the gravity direction based on comparable images from different periods. The structural anchor points serve as stable spatial reference points to ensure consistent retrieval locations, while the gravity direction constraint aligns with the physical path development of risks such as leakage. Subsequently, differentiated candidate generation rules are designed for different failure modes. Corrosion candidates are generated based on circumferential coating peeling, color abnormalities, or increased surface roughness; leakage candidates are generated based on the stripe morphology along the gravity direction and thermal imaging temperature differences; deformation candidates are generated based on axial and circumferential positional offsets; and loosening candidates are generated based on increased clamping gaps or hole displacement. The physical rationality of candidate generation is improved by combining spatial distribution patterns and multimodal image features. Next, a local benchmark is established for candidate regions based on stable regions of the same material to eliminate interference from global illumination or material differences. Candidate regions that exist or show an increasing trend in adjacent acquisition periods are retained as risk points, and temporal continuity is used to filter temporary noise. Finally, quantitative indicators were extracted based on changes in appearance, morphology, thermal anomalies, and geometric deviations. The deterioration state of risk points was decomposed into measurable dimensions. A set of indicators was formed by combining the unified spatial coordinates of risk points with quantitative indicators, which not only retained the ability to locate spatial locations but also provided multi-dimensional deterioration characterization.

[0110] As a specific implementation method, the solution of this application is implemented as follows: structural anchor points are set at the clamp positions of the external riser, and candidate leakage areas are searched along the gravity direction within a certain range downstream of the joint; for external risers made of steel pipe, the pipe section between adjacent supports is selected as a stable area of ​​the same material to establish a local benchmark; when a certain area is detected to exist in two consecutive images and the length of the stain extension increases, it is identified as a leakage risk point, and its thermal anomaly amplitude and duration are calculated as quantitative indicators.

[0111] The above technical solution avoids misjudging temporary environmental disturbances as real risk points and achieves accurate identification and quantitative characterization of four types of failure modes: corrosion, leakage, deformation, and loosening.

[0112] In practical applications, some of the embodiments described above in this application propose a set of risk point indicators to quantify appearance changes, morphological changes, thermal anomalies, and geometric deviations. However, in the implementation process, risk points are isolated and cannot reflect the physical path relationship of failure evolution along axial dripping, support connection, and wall penetration. This leads to the neglect of mechanisms such as capillary diffusion, gravity dripping, and material moisture absorption when correlated across periods. As a result, the construction of risk point trajectories lacks physical constraints, has a high misjudgment rate, and is difficult to accurately predict the deterioration trend.

[0113] In response, this application further proposes that when constructing a spatiotemporal coupling graph, the following steps are taken: in the surface unfolding plane, an axial dripping path, a support connection path, and a wall penetration path are generated based on the positions of the structural anchor points and joints; wherein, the axial dripping path traces downward from the structural anchor point or joint along the axial direction of the external riser and in the direction of gravity, and the path direction is determined by the extension of stains and the connection of thermal anomalies; the support connection path connects the center positions of the components of adjacent supports and clamps, and the path direction is determined by the component sequence; the wall penetration path extends from the junction of the external riser and the external facade towards the external facade, and the path direction is determined by the connection of appearance changes and the connection of thermal anomalies; the nearest path position is determined for each risk point on the path and an initial edge is established to obtain a spatiotemporal coupling graph containing nodes, edges, and edge weights.

[0114] Among them, structural anchor points refer to stable reference points used for spatial positioning in the external riser system. These can be achieved using geometric feature points such as clamp edges, support edges, joint profiles, or elbow profiles, aiming to provide a unified spatial benchmark to ensure the accuracy of path generation. Axial drip path refers to the failure propagation path formed by gravity along the axial direction of the external riser. It can be determined based on the continuity of stain extension and the connectivity of thermal anomaly areas, aiming to realistically recreate the drip evolution process of leakage failure along the direction of gravity. Support connection path refers to the path connecting the external riser support structure, which can be defined based on the geometric center sequence of the supports and clamps, aiming to... The system is designed to depict the transmission path of loosening or deformation risks within the support system; the wall penetration path refers to the path of outward diffusion from the junction of the external riser and the wall, which can be determined by combining information from both appearance changes and thermal anomalies, with the aim of simulating the penetration mechanism of corrosion or leakage spreading to the wall; the nearest path location refers to the projection point of the risk point on the physical path, which can be achieved using the Euclidean distance minimization algorithm, with the aim of accurately anchoring the risk point to the physical path; the initial edge refers to the edge connecting the risk point node and the path, which can be represented using the adjacency relation in graph theory, with the aim of forming a graph structure foundation with risk points as nodes and physical paths as edges.

[0115] Specifically, the solution in this application generates axial dripping paths, support connection paths, and wall penetration paths in the curved surface unfolding plane based on the positions of structural anchor points and joints, ensuring that path generation is strictly based on the physical structural characteristics of the external riser. The axial dripping path is determined by the extension of stains and the connectivity of thermal anomalies, utilizing the natural extension of stains under gravity and the connectivity of thermal anomaly areas to realistically recreate the dripping evolution process of leakage failure along the direction of gravity. The support connection path is determined by the sequence of components, accurately depicting the transmission path of loosening or deformation risks between support systems through the serialized association of mechanical connection points. The wall penetration path is determined by the connectivity of appearance changes and thermal anomalies, combining the dual connectivity of visual and thermal anomalies to simulate the penetration mechanism of corrosion or leakage spreading to the wall. The nearest path position is determined for risk points on the path and an initial edge is established, accurately mapping the risk points to the physical path, forming a spatiotemporal coupling graph containing nodes, edges, and edge weights, so that the risk point trajectory is tightly coupled with the physical mechanism of failure.

[0116] As a specific implementation method, the solution of this application is implemented as follows: the structural anchor point can be a permanent marker point set on the facade, such as the intersection of the window sill line and the corner line; the axial drip path can be tracked based on the connectivity of temperature difference areas in thermal imaging, such as a strip-shaped thermal anomaly area extending downward along the direction of gravity; the support connection path can connect the geometric centers of adjacent supports, and the path direction is determined by the order of the support sequence; the wall penetration path can be combined with the connectivity of color change areas and thermal anomaly areas in visible light images, such as a rust area spreading outward from the junction of the external riser and the wall; the risk point can be anchored to the nearest path position, such as the leakage risk point located on the axial drip path, and the deformation risk point located on the support connection path.

[0117] Through the above technical solutions, the construction of risk point trajectories has physical constraints, which improves the accuracy of cross-period correlation, reduces the false judgment rate, and improves the reliability of deterioration trend prediction.

[0118] Traditional methods for monitoring risk points in external risers, when constructing spatiotemporal coupling maps to describe the physical coupling relationships between risk points, suffer from issues such as trajectory breaks, mismatches, or path connectivity distortions when linking risk points across different collection periods due to interference from obstructions, fluctuations in environmental parameters, or instantaneous changes in quantitative indicators. This makes it impossible to accurately reflect the deterioration evolution process driven by physical mechanisms such as capillary diffusion, gravity dripping, and material moisture absorption, thereby affecting the continuity and reliability of deterioration trend prediction.

[0119] In this regard, this application further proposes that when obtaining the intertemporal trajectory of risk points, the following methods should be included:

[0120] Within the sliding time window, the extension speed and continuity of the stain along the direction of gravity are extracted based on capillary diffusion, the strip morphology and downstream connectivity are extracted based on gravity dripping, and the area growth rate and boundary expansion rate of the same material region are extracted based on material moisture absorption. In combination with component constraints, geometric proximity and path connectivity, the edge weights are calculated according to the preset weight rules.

[0121] Risk points in adjacent data collection periods are paired on the spatiotemporal coupling graph. Cross-period associations are established when spatial proximity, path connectivity, similar quantitative indicators with consistent growth trends, and both positive and negative matching are met. When pairing conflicts occur, priority is given to pairings that are connected by all three types of paths and have higher edge weights. Trajectories with short-term occlusion are compensated for missing data. Trajectories that do not meet the conditions for multiple consecutive periods are terminated. New trajectories are established for newly emerging risk points that meet the initial conditions, and a unified number is assigned after the association is completed.

[0122] In practical applications, capillary diffusion extraction refers to quantifying the natural permeation characteristics of liquids in material pores. This can be achieved by analyzing the pixel displacement rate and morphological continuity index of contaminants along the gravity direction in continuously acquired images, aiming to avoid misjudgments caused by instantaneous interference from relying solely on image grayscale changes. Gravity drip extraction can be understood as identifying the axial stripe features formed by contaminants under gravity. This can be achieved by calculating the stripe length ratio, downstream region connectivity index, and morphological similarity parameters, aiming to reliably identify the physical continuity of the leakage path and reduce false positive associations caused by fluctuations in ambient light. Material hygroscopic extraction specifically refers to characterizing the impact of material hygroscopicity on corrosion diffusion, which can be achieved by monitoring areas of the same material... The temperature gradient change rate and boundary expansion rate in thermal imaging are used to capture the true growth trend of risk points and overcome the instantaneous apparent interference caused by rain or temperature changes in a single image. Edge weight calculation can be understood as a reliability assessment of association based on comprehensive physical constraints. It can be implemented using a weighted fusion algorithm based on weight factors of component geometric constraints, spatial distance decay function and path connectivity confidence, with the aim of making the association results more consistent with the actual degradation law. Pairing conflict handling specifically refers to solving the decision problem of multi-candidate matching. It can be implemented using a strategy that combines a voting mechanism of three types of path connectivity verification with edge weight threshold comparison, with the aim of prioritizing strong association evidence that conforms to the physical mechanism.

[0123] This application's solution extracts dynamic evolution features by integrating three physical mechanisms—capillary diffusion, gravitational dripping, and material moisture absorption—within a sliding time window. It then constructs a multi-dimensional edge weight system based on component constraints, geometric proximity, and path connectivity, forming a physical mechanism-driven correlation reliability assessment mechanism. A multi-condition verification pairing strategy is implemented on the spatiotemporal coupling graph, requiring simultaneous satisfaction of spatial proximity, path connectivity, quantitative index similarity, and growth trend consistency. A bidirectional verification mechanism of forward and reverse matching ensures correlation robustness. When pairing conflicts occur, the system prioritizes pairing schemes with higher edge weights that simultaneously pass verification through axial dripping paths, support connection paths, and wall penetration paths, avoiding path distortion caused by random matching. For short-term occlusion scenarios, the system performs missing measurement compensation based on historical trajectory data and the physical evolution model to maintain trajectory continuity. When a risk point fails to meet the correlation conditions for multiple consecutive periods, the trajectory is automatically terminated to prevent invalid predictions. For newly emerging risk points, a new trajectory is established through initial condition verification to ensure timely capture of new risk sources. Finally, a unified numbering mechanism provides a unique identifier for each risk point trajectory, enabling precise tracking of the risk point evolution process.

[0124] As a specific implementation method, this application is implemented as follows: Within the curved surface unfolding plane, for leakage risk points of steel pipe external risers, the system first extracts the extension speed of stains along the gravity direction based on the morphological characteristics of temperature difference areas in thermal imaging images. Gravity dripping parameters are determined by calculating the change rate of strip elongation ratio and the growth rate of downstream connected area in images acquired continuously. Simultaneously, the system monitors the color difference trend of areas of the same material in visible light images and extracts the material's moisture absorption index based on environmental humidity data. On the spatiotemporal coupling map, the system pairs and verifies current risk points with previous candidate points. When the spatial distance between two points is less than a preset threshold, both are located on the same axial dripping path, the thermal anomaly amplitude changes in the same direction, and the displacement deviations of both forward and reverse matching are less than the allowable range, a cross-period association is established. If a risk point lacks single-period data due to scaffolding obstruction, the system performs linear interpolation compensation based on the extension speed and direction of historical trajectories. When a risk point fails to meet the association conditions for three consecutive periods, the trajectory is automatically terminated. Newly emerging risk points, after verification that they meet the initial conditions, establish a new trajectory and assign a unique number.

[0125] Through the above technical solutions, this application solves the problems of breakage and mismatch in the cross-period trajectory correlation of risk points, realizes accurate tracking of the degradation evolution process based on physical mechanisms, and ensures the continuity and reliability of degradation trend prediction.

[0126] Specifically, in some of the embodiments described above in this application, a method is proposed to obtain the expected time to reach the maintenance threshold in order to predict the deterioration trend of the risk point of the external riser. However, in this process, short-term fluctuations caused by environmental factors interfere with the trend fitting process, the weight setting does not fully combine the differences in failure modes and the characteristics of indicators, resulting in distortion of the degradation state representation, and the lack of multi-dimensional verification standards in the consistency verification mechanism leads to insufficient reliability of the prediction results, thereby affecting the accuracy of maintenance decisions.

[0127] In this regard, this application further proposes methods for obtaining the expected time to reach the maintenance threshold, including:

[0128] Using four failure modes—corrosion, leakage, deformation, and loosening—as the source of weights, and combining index stability, intertemporal consistency, and path connectivity, weights are assigned to various indicators in the risk point indicator set to obtain a representation of the degradation state.

[0129] Robust regression is used to fit the trend of the degradation state representation, and short-term fluctuations caused by rainfall, sunshine, and temperature cycles are corrected based on time-related information. Consistency checks are performed on the fitting results; if the consistency check fails, a re-check suggestion is output. If the consistency check passes, the expected time to reach the maintenance threshold is calculated based on the difference between the degradation state representation and the maintenance threshold, as well as the growth rate of the degradation state representation. Consistency checks include minimum period count check, monotonicity check, forward and backward fit consistency check, residual limit check, and window consistency check.

[0130] Among them, the degradation status representation refers to the core indicator that comprehensively quantifies the degree of deterioration of the risk points of the external riser. It can be achieved by using a weighted fusion of quantitative indicators such as color difference and thermal anomaly amplitude from the risk point indicator set. The purpose is to more accurately reflect the actual deterioration process under different failure modes. Robust regression can be understood as a trend fitting method that is robust to outliers. Specifically, it can be implemented using the M-estimator or the least squares regression. Its purpose is to suppress image noise and outlier interference and ensure that the fitting results focus on the continuous deterioration trend. Consistency verification refers to the multi-dimensional verification mechanism implemented on the trend fitting results. It can be implemented by setting a minimum period threshold, monotonicity direction constraints, and bidirectional time series verification. The purpose is to build a strict quality control system and intercept unreliable prediction results.

[0131] Specifically, this application's solution uses four failure modes—corrosion, leakage, deformation, and loosening—as the weighting source, allocating index weights based on the unique influence of different failure physical mechanisms on the degradation process, thus avoiding bias caused by a single index. It combines index stability to screen for long-term reliable quantitative data, utilizes intertemporal consistency to verify the comparability of multi-period images, and introduces path connectivity to reflect physical coupling relationships between risk points, such as axial dripping and support conduction, forming a comprehensive representation of the degradation state. When performing trend fitting based on robust regression, its robustness against outliers suppresses random fluctuations, and it identifies and eliminates short-term effects caused by rainfall, sunshine, and temperature cycles by associating environmental parameters with time-related information. The time fluctuations allow the fitting to focus on the true degradation trend; in the subsequent multi-dimensional consistency verification, the minimum number of periods verification ensures that the amount of data supports the trend analysis, the monotonicity verification verifies the reasonable direction of the degradation state, the forward and backward fitting consistency verification verifies the model stability through bidirectional time series, the residual limit verification controls the fitting error range, and the window consistency verification ensures the continuity of prediction results in different time windows. These verifications together constitute a closed-loop verification mechanism; when the verification fails, a re-examination suggestion is output to trigger manual review, and after the verification passes, the remaining lifespan is scientifically quantified based on the difference between the degradation state representation and the maintenance threshold and the growth rate, thus forming a complete technical chain from data fusion to trend prediction.

[0132] As a preferred embodiment, the specific implementation of this application's solution is as follows: For a corrosion risk point on the external riser, a higher weight can be assigned based on the corrosion failure mode. Long-term stable indicators such as surface roughness changes are screened in conjunction with indicator stability. Based on path connectivity, associated risk points on the axial dripping path are included in the weight calculation to obtain a degradation state representation. Robust regression is used to perform trend fitting on the degradation state representation, while short-term brightness fluctuations are corrected based on rainfall event information in the collected records. If the fitting result passes the minimum number of periods verification (e.g., no less than three periods of data within the sliding time window), monotonicity verification (the degradation state shows a non-decreasing trend), and forward and backward fitting consistency verification (bidirectional error is below the threshold), the estimated time to reach the maintenance threshold is calculated based on the difference between the degradation state representation and the maintenance threshold and the growth rate. If the window consistency verification fails, a re-inspection suggestion is output and data is collected again.

[0133] Through the above scheme, this application eliminates the interference of short-term fluctuations caused by environmental factors on trend fitting, making the degradation state representation more consistent with the actual deterioration mechanism. At the same time, the reliability of the prediction results is improved through a multi-dimensional consistency verification mechanism, thereby providing accurate time basis for external riser maintenance decisions.

[0134] Specifically, in some of the embodiments described above in this application, maintenance suggestions are proposed to guide maintenance work based on the expected arrival time of the maintenance threshold. However, in the implementation process, the maintenance suggestions lack a dynamic risk point priority classification mechanism, collaborative optimization of operation time windows and environmental factors, multi-dimensional output of accurate location data, linkage processing of physical path associations between risk points, and targeted matching of categorized maintenance actions. This leads to an imbalance in the allocation of maintenance resources, inappropriate selection of operation timing, difficulty in on-site positioning, omission of chain risks, and generalization of maintenance measures, making it difficult to support efficient and accurate predictive maintenance.

[0135] In response, this application further proposes that when generating maintenance recommendations based on the expected time to reach the maintenance threshold, the following should be included:

[0136] Based on the relationship between the expected time to reach the maintenance threshold and the sliding time window, risk points are classified into three levels: urgent, priority, and routine. A suggested operation time window is determined by combining time-related information. Operation positioning data is output based on unified spatial coordinates and structural anchor points, including the unified spatial coordinate position, the surface unfolding plane position, and the nearest structural anchor point identifier. Based on the spatiotemporal coupling diagram, risk points connected to the target risk point are retrieved along the axial drip path, support connection path, and wall seepage path. Risk points expected to reach the maintenance threshold within the buffer period are merged to generate a coordinated maintenance recommendation. A maintenance action list is provided according to the risk point type, including at least one of the following: anti-corrosion surface treatment and coating repair, leakage sealing and diversion treatment, component repositioning and replacement, clamp tightening, and support reinforcement.

[0137] The final output includes cross-period comparison images, a summary of change indicators, the estimated time to reach the maintenance threshold, recommended operation time windows, and a maintenance action list.

[0138] Among these, risk point classification refers to dynamically prioritizing risk points based on the relationship between the expected time to reach the maintenance threshold and the sliding time window. This can be achieved using a time difference threshold comparison algorithm, aiming to dynamically capture changes in risk urgency and avoid resource allocation imbalances caused by static classification. Recommended operation time window determination refers to determining the optimal time period for maintenance operations by combining time-related information. This can be achieved by building a time series model using historical environmental parameter data, aiming to embed maintenance operations into periods of minimal environmental interference and reduce maintenance interruptions caused by sudden weather changes. Operation positioning data output refers to generating multi-dimensional positioning information based on unified spatial coordinates and structural anchor points. This can be achieved by using a coordinate transformation algorithm to map risk points to unified spatial coordinate positions and surface unfolding plane positions, and associating them with the nearest structural anchor point identifier. This aims to overcome the limitations of cylindrical surfaces. The issues include: distortion of perspective in positioning; risk point linkage retrieval refers to searching for relevant risk points on physical paths based on spatiotemporal coupling graphs. This can be achieved using graph traversal algorithms to search along axial dripping paths, support connection paths, and wall penetration paths, aiming to identify cascading risks driven by capillary diffusion, gravity dripping, and material moisture absorption mechanisms; linkage maintenance suggestion generation refers to merging risk points with similar expected maintenance threshold times within a buffer period. This can be achieved by clustering risk points using time window matching algorithms, aiming to avoid duplicate work and path-related risk omissions caused by isolated processing; and maintenance action list matching refers to providing targeted maintenance actions according to risk point types. This can be achieved by mapping four failure modes—corrosion, leakage, deformation, and loosening—to corresponding maintenance action sets based on a rule base, aiming to ensure that maintenance measures accurately correspond to degradation mechanisms.

[0139] Specifically, the proposed solution first dynamically classifies risk points based on the relationship between the expected time to reach the maintenance threshold and the sliding time window, using this classification as input for subsequent steps. Then, it analyzes historical patterns of environmental parameters such as rainfall, sunshine, and temperature cycles using time-related information to determine the recommended work time window with minimal environmental interference. Simultaneously, it maps risk points to unified spatial coordinates and surface unfolding planes based on unified spatial coordinates and structural anchor points, associating them with the nearest structural anchor point identifier to generate precise work positioning data. Building upon this, it retrieves physically associated risk points based on the spatiotemporal coupling diagram along the axial drip path, support connection path, and wall penetration path. Within the buffer period, it clusters risk points with similar expected maintenance threshold times to generate coordinated maintenance recommendations. Finally, it matches corresponding maintenance action lists from a pre-set action library according to risk point type and integrates multi-dimensional information such as cross-period comparison images and change index summaries to output complete maintenance recommendations. Each step is executed sequentially with information transmitted level by level. Risk point classification guides resource allocation priority, operation time window determination optimizes operation timing, location data solves on-site location problems, linkage suggestions handle path-related risks, and action list ensures that maintenance measures are targeted, ultimately forming a closed-loop decision-making process from risk assessment to maintenance execution.

[0140] As a preferred embodiment, the solution of this application is implemented as follows: When the system detects that the estimated time to reach the maintenance threshold of a certain external riser risk point is short, it automatically classifies it as an emergency level; the system analyzes historical environmental data to identify stable weather periods as suggested operation time windows; using a unified spatial coordinate system established by a cylindrical surface model, the risk point location is accurately mapped to the surface of the external riser and associated with the nearest clamp mark; in the spatiotemporal coupling diagram, another risk point is found downstream along the axial drip path, whose estimated time to reach the maintenance threshold is similar, and the two are merged into a joint maintenance suggestion within the buffer period; for risk points of leakage type, the system recommends sealing and diversion treatment actions at the leakage point; finally, a comprehensive report is generated that includes cross-period image comparison, summary of index changes, and detailed maintenance guidance.

[0141] Through the above solutions, this application can dynamically adjust the priority of risk points in the maintenance of external risers, ensuring that emergency risks are handled in a timely manner without delay; it recommends that the operation time window be synchronized with environmental patterns to reduce operation interruptions caused by sudden weather changes; multi-dimensional positioning data enables maintenance personnel to quickly and accurately locate risk points, improving on-site work efficiency; the linkage maintenance recommendation identifies and handles path-related risks, avoiding repetitive work and the omission of chain risks; targeted maintenance actions ensure that measures are accurately matched with the degradation mechanism, improving maintenance effectiveness, thereby achieving efficient and accurate predictive maintenance.

[0142] In the above embodiments, by organizing multi-period data and establishing a unified spatial coordinate system for facade control points and external pipe components, and combining a cylindrical curved surface model to implement cross-period registration and surface unfolding with surface constraints, consistent mapping of the same physical part in different acquisition periods is achieved under the constraints of structural anchor point alignment and cyclic consistency verification. Cross-period radiometric standardization and thermal imaging physical correction are completed based on imaging parameters and environmental parameters, reducing the cross-period incomparability caused by differences in viewing angle, illumination, and thermal radiation. Within the surface unfolding plane, a set of risk point indicators is constructed around appearance changes, morphological changes, thermal anomalies, and geometric deviations, and then further refined based on axial drip path and support connection... A spatiotemporal coupling map is established using the connection path and the wall penetration path as edges. Edge weights are determined according to capillary diffusion, gravity dripping, and material moisture absorption mechanisms, achieving robust intertemporal correlation and unified numbering of risk points, reducing spurious changes and false correlations. Within a sliding time window, the risk point indicator set is weighted and fused to form a degradation state representation. The estimated time to reach the maintenance threshold is output through trend fitting and time correlation consistency verification, and maintenance suggestions are generated in conjunction with this. This forms a closed loop of data collection—location—standardization—mapping—prediction—maintenance, improving the intertemporal comparability of external risers, the accuracy of risk point location, and the reliability of trend prediction, while reducing false alarms and duplicate inspections.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for multi-period image comparison and degradation trend prediction of external riser risk points, characterized in that, include: Multiple images were acquired, and the acquisition time, acquisition posture, imaging parameters and environmental parameters were recorded. A unified time reference and sliding time window were established to obtain a multi-phase dataset. The multi-phase images include visible light images and thermal images of the external riser. Based on the facade control points and external pipe components, a unified spatial coordinate system is established and a cylindrical surface model is fitted. Imaging distortion correction and perspective correction are performed on the multi-period dataset to obtain standardized images. Under the constraints of the cylindrical curved surface model, the standardized image is registered across periods, and the outer riser area is unfolded into a curved unfolded plane. The structural anchor points are used for co-alignment and cyclic consistency verification to obtain the cross-period co-aligned image. Based on the imaging parameters and environmental parameters, the intertemporal isotopic images are subjected to intertemporal radiometric normalization, and the thermal imaging images are physically corrected to obtain intertemporal comparable images. Risk points are generated within the surface development plane for corrosion, leakage, deformation and loosening. Quantitative indicators are extracted based on appearance changes, morphological changes, thermal anomalies and geometric deviations to obtain a set of risk point indicators. A spatiotemporal coupling graph is constructed using the risk points as nodes and the axial dripping path, the support connection path, and the wall penetration path as edges. The edge weights are determined based on capillary diffusion, gravity dripping, and the material moisture absorption mechanism. The risk points are then correlated across periods and assigned a unified number to obtain the cross-period trajectory of the risk points. Within a sliding time window, the risk point indicator set is weighted and fused to form a degradation state representation. The degradation state representation is then subjected to trend fitting and consistency verification is performed in conjunction with time-related information to obtain the expected time to reach the maintenance threshold. Maintenance recommendations are generated based on the estimated time to reach the maintenance threshold.

2. The external riser risk point multi-phase image correlation and degradation trend prediction method of claim 1, wherein, When obtaining multi-period datasets, the following are included: The internal and external parameters of the data acquisition equipment are calibrated and synchronized with a unified time reference. The visible light image and thermal image are acquired at the same time during each acquisition period. A uniform exposure time, sensitivity, aperture value, and white balance mode are set for the visible light images, and a uniform emissivity, background temperature, and reflection compensation are set for the thermal images, with on-site correction performed using a temperature reference surface. The imaging parameters include exposure time, ISO, aperture value, lens focal length, focus distance, white balance mode, image resolution, image frame rate, and noise suppression level. The imaging parameters also include thermal imaging emissivity setting, background temperature setting, reflection compensation parameters, and non-uniformity correction status. The environmental parameters include ambient temperature, air humidity, sunlight intensity, and wind speed.

3. The external riser risk point multi-phase image correlation and degradation trend prediction method of claim 1, wherein, When establishing a unified spatial coordinate system and fitting a cylindrical surface model, the following steps are included: No fewer than three facade control points are set on the facade and used as permanent markers. The scale and direction of the unified spatial coordinates are determined by the relative spacing and included angle of the facade control points. The window sill line, corner line, and eaves line are used as auxiliary geometric constraints to define the planar position; Stable boundaries of clamp edges, support edges, joint profiles and elbow profiles are extracted within the external riser area. The axial direction and cross-sectional radius of the external riser are estimated based on the geometric fitting method to obtain the cylindrical surface model.

4. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 3, wherein, When obtaining isotopic images across different periods, the following should be included: A baseline for expansion, starting from the structural anchor point, is selected and remains unchanged throughout the data collection period; Coarse registration of axial and circumferential directions is completed under unified spatial coordinates, and fine registration is performed under the constraints of cylindrical surface model, limiting axial displacement, circumferential displacement and local deformation to not exceed preset position limits. The outer riser area is mapped to the curved surface unfolding plane according to the unfolding baseline, and the relative order of the structural anchor points in the circumferential direction is kept unchanged during the mapping process; the co-alignment is performed in the neighborhood of the structural anchor points, the co-alignment includes boundary shape consistency verification and texture stable region consistency verification, and abnormal areas are removed by occlusion detection rules. Cyclic consistency checks are performed across multiple consecutive periods. If the check fails, the process is rolled back and the expansion baseline is redefined. Once the check is met, the cross-period co-positional image is obtained.

5. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 1, wherein, When obtaining comparable images across different periods, the following should be included: Determine the baseline acquisition period and use visible light images and thermal images as the radiometric reference; Set up a reference gray card and keep it in the same position and orientation in each acquisition period. Perform lens vignetting correction, zone gain correction and white balance unification on the visible light image. Use the reference gray card to perform brightness mapping on the same material area. Detect and mark areas with strong reflection, shadow and raindrop pollution and remove them in the cross-period comparison. For thermal imaging images, emissivity is set for steel pipes, coated pipes and insulation layers according to the external pipe material library. Two-point correction is performed using a temperature reference surface, background temperature is recorded and reflection compensation and non-uniformity correction are performed. Brightness consistency is verified in the stable area of ​​the external riser component to ensure that the brightness difference of the same component in different acquisition periods does not exceed the preset brightness difference limit, thereby obtaining the cross-period comparable image.

6. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 5, wherein, When obtaining the set of risk point indicators, it includes: Based on the comparable images across different periods, candidate regions are retrieved in the gravity direction near the structural anchor point and downstream of the joint; corrosion candidates are generated for regions with coating peeling, abnormal color, or increased surface roughness along the circumferential direction; leakage candidates are generated for regions with stripes along the gravity direction and temperature differences in the thermal imaging images; deformation candidates are generated for regions with offset between the axial and circumferential positions of the external riser; and loosening candidates are generated for regions with increased clamping gap between the clamp and the bracket or increased relative displacement of the hole. A local benchmark is established for candidate regions based on stable regions of the same material. Candidate regions that exist or show an increasing trend in adjacent collection periods are retained as risk points. Quantitative indicators are extracted based on the appearance changes, morphological changes, thermal anomalies and geometric deviations. The risk point index set is formed by combining the unified spatial coordinates of the risk points with the quantitative indicators.

7. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 6, wherein, Constructing a spatiotemporal coupling graph includes: In the curved surface unfolding plane, axial dripping paths, support connection paths, and wall penetration paths are generated based on the structural anchor points and joint positions. The axial dripping path traces downwards from the structural anchor point or joint along the axial direction of the external riser and in the direction of gravity, with the path direction determined by stain extension and thermal anomaly connectivity. The support connection path connects the center positions of adjacent supports and clamps, with the path direction determined by the component sequence. The wall penetration path extends from the junction of the external riser and the external facade towards the external facade, with the path direction determined by appearance changes and thermal anomaly connectivity. For each risk point on the path, the nearest path position is determined and an initial edge is established, resulting in a spatiotemporal coupling graph containing nodes, edges, and edge weights.

8. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 7, wherein, When obtaining the intertemporal trajectory of risk points, the following should be included: Within the sliding time window, the extension speed and continuity of the stain along the direction of gravity are extracted based on capillary diffusion, the strip shape and downstream connectivity are extracted based on gravity dripping, the area growth rate and boundary expansion rate of the same material region are extracted based on material moisture absorption, and the edge weights are calculated according to the preset weight rules in combination with component constraints, geometric proximity and path connectivity. On the spatiotemporal coupling graph, risk points in adjacent acquisition periods are paired. When spatial proximity, path connectivity, similar quantitative indicators with consistent growth trends, and both positive and negative matching are met, cross-period associations are established. When pairing conflicts occur, priority is given to pairings that are connected through three types of paths and have higher edge weights. Trajectories with short-term occlusion are compensated for missing data. Trajectories that do not meet the conditions for multiple consecutive periods are terminated. New trajectories are established for newly emerging risk points that meet the initial conditions, and a unified number is assigned after the association is completed.

9. The external riser risk point multi-phase imagery correlation and degradation trend prediction method of claim 1, wherein, When obtaining the estimated time to reach the maintenance threshold, the following are included: Using four failure modes—corrosion, leakage, deformation, and loosening—as the weighting source, and combining index stability, intertemporal consistency, and path connectivity, weights are assigned to various indicators in the risk point index set to obtain the degradation state representation. The degradation state representation is trend-fitted based on robust regression, and short-term fluctuations caused by rainfall, sunshine, and temperature cycles are corrected according to time-related information. The fitting results are then subjected to consistency verification, and a re-examination suggestion is output if the consistency verification fails. If the consistency verification passes, the estimated time to reach the maintenance threshold is calculated based on the difference between the degradation state representation and the maintenance threshold, as well as the growth rate of the degradation state representation.

10. The external riser risk point multi-era imagery correlation and degradation trend prediction method of claim 1, wherein, When generating maintenance recommendations based on the estimated time to reach the maintenance threshold, the following are included: Based on the relationship between the estimated time to reach the maintenance threshold and the sliding time window, risk points are classified into three levels: urgent, priority, and routine. A suggested operation time window is determined by combining time-related information. Operation positioning data is output based on the unified spatial coordinates and structural anchor points, including the unified spatial coordinate position, the curved surface unfolding plane position, and the nearest structural anchor point identifier. Based on the spatiotemporal coupling diagram, risk points connected to the target risk point are retrieved along the axial drip path, support connection path, and wall seepage path. Risk points with an estimated time to reach the maintenance threshold within the buffer period are merged to generate a linked maintenance suggestion. A maintenance action list is provided according to the risk point type, including at least one of the following: anti-corrosion surface treatment and coating repair, leakage sealing and diversion treatment, component repositioning and replacement, clamp tightening, and support reinforcement. The final output includes cross-period comparison images, a summary of change indicators, the estimated time to reach the maintenance threshold, recommended operation time windows, and a maintenance action list.

Images