A mountainous area super-large bridge comprehensive analysis method based on intelligent inspection unmanned plane

Abstract

The application provides a mountainous large bridge comprehensive analysis method based on an intelligent inspection unmanned aerial vehicle, and belongs to the technical field of bridge health management. The method comprises the following steps: performing spatial overlap degree matching on each bridge disease entity identified in the current inspection and the same-position disease entity in a historical inspection archive, and determining the continuous evolution sequence of each bridge disease; determining the visual rendering coefficient of the corresponding bridge disease in the preset bridge three-dimensional model from the time sequence variation of the disease characteristic parameters in the continuous evolution sequence, and constructing a disease density distribution hot area; when the time sequence variation of the bridge disease extracted in the disease density distribution hot area exceeds the allowable evolution interval of the corresponding disease type, performing difference analysis on the neighborhood structure surface information of the bridge disease in the bridge three-dimensional model and the original design parameters of the mountainous large bridge; and updating the allowable evolution interval according to the difference analysis result. The technical scheme provided by the application can dynamically distinguish the normal damage development and abnormal accelerated degradation in the mountainous large bridge.

Description

Technical Field

[0001] This application relates to the field of bridge health management technology, and more specifically, to a comprehensive analysis method for large mountain bridges based on intelligent inspection drones. Background Technology

[0002] As crucial nodes in transportation arteries, the structural safety of major bridges in mountainous areas is of paramount importance. However, traditional manual inspection methods are limited by complex terrain and the risks of working at heights, resulting in inherent shortcomings such as low efficiency, incomplete coverage, and numerous blind spots. These methods fail to meet the refined requirements of modern bridge maintenance. In recent years, intelligent inspection drones have been widely used in daily inspections, but the current comprehensive analysis of inspection data has significant deficiencies: First, inspection data is stored in a scattered manner, without correlation analysis of multiple types of defects, making it impossible to uncover the inherent connections between defects and resulting in insufficient analysis. Second, defect location and spatial distribution are disconnected, making it difficult to intuitively grasp the distribution patterns of defects and hindering precise treatment. Third, there is a lack of complete digital archives of defects, making it impossible to automatically track the development process of the same defect and making it difficult to predict trends. Fourth, inspection results require manual processing, which is inefficient, and the lack of real-time monitoring of drones, sensors, and other equipment makes it easy for equipment failures to affect the continuity of inspections. Existing technologies are insufficient to meet the needs of comprehensive utilization of inspection data and intelligent bridge health management for major bridges in mountainous areas.

[0003] Chinese patent CN109060281A discloses an integrated bridge inspection system based on unmanned aerial vehicles (UAVs), belonging to the field of UAV bridge inspection technology. It includes a bridge modeling UAV A, a bridge surface data acquisition UAV B, and a ground-based integrated information processing and control system. The ground-based integrated information processing and control system includes a 3D coordinate modeling system, a UAV flight path planning system, a bridge deck defect detection and annotation system, and a bridge quality inspection report generation system. UAV A is used for photographing the bridge under inspection and its surrounding terrain environment; the 3D coordinate modeling system is used to establish a 3D coordinate model of the bridge and its surrounding terrain environment; the UAV flight path planning system is used to plan the flight path of UAV B; and the bridge deck defect detection and annotation system is used to detect defects at corresponding locations on the bridge deck, calculate defect severity indicators, and mark them in the model. This invention achieves full automation of the bridge inspection process, significantly improving inspection efficiency and quality. This scheme establishes a bridge model with a spatial reference through oblique photography and 3D coordinate modeling. Based on this, it automatically calculates and generates a cruise path that fits the bridge surface, drives an inspection drone to collect high-resolution surface images, and then uses a visual defect detection algorithm to automatically identify cracks and other defects and calculate the degree of damage. However, in the long-term health management of extra-long-span bridges in mountainous areas, due to the disturbance of canyon wind fields, satellite signal blockage, and the accumulation of navigation errors of the drone itself, the spatial coordinates of defects obtained in different inspection cycles exhibit non-stationary random drift. This uncertainty in the data layer makes it impossible to determine the location of cracks or concrete spalling in the same structure as the same source defect entity in spatial semantics. Existing technical solutions often rely on manual experience for fuzzy alignment or treat defects as isolated point events, which cannot effectively extract the long-term continuous temporal variation of defect morphological characteristic parameters under a unified spatiotemporal reference. This makes it difficult to distinguish whether the variation is caused by the actual creep effect of concrete materials or fatigue crack propagation, thus making it impossible to dynamically analyze the problem of normal damage development and abnormal accelerated deterioration in extra-long-span bridges in mountainous areas. Therefore, how to dynamically distinguish between normal damage development and abnormal accelerated deterioration in extra-long-span bridges in mountainous areas has become a difficult problem for the industry. Summary of the Invention

[0004] This application provides a comprehensive analysis method for major bridges in mountainous areas based on intelligent inspection drones, which can dynamically distinguish between normal damage development and abnormal accelerated deterioration in major bridges in mountainous areas.

[0005] In the first aspect, this application provides a comprehensive analysis method for major bridges in mountainous areas based on intelligent inspection drones, including the following steps: Acquire data on various types of defects and their corresponding location data collected by intelligent inspection drones during this inspection of a major bridge in a mountainous area. Based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archive. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined. The visualization rendering coefficient of the corresponding bridge disease in the preset bridge 3D model is determined based on the temporal change of the disease characteristic parameters in the continuous evolution sequence, and a disease density distribution hot zone is constructed in the bridge 3D model based on the visualization rendering coefficient. Extract the disease entities corresponding to the target density region in the disease density distribution hot zone. When the temporal change of any extracted bridge disease exceeds the allowable evolution range of its corresponding disease type, perform an overlay analysis between the neighborhood structural surface information of the bridge disease in the bridge three-dimensional model and the original design parameters of the mountain super bridge. The allowable evolution range is updated based on the results of the overlay analysis, and the updated allowable evolution range is used as the anomaly judgment benchmark when inspecting the same bridge defects again.

[0006] Furthermore, the acquisition of various types of damage data and corresponding damage location data collected by intelligent inspection drones during this inspection of major bridges in mountainous areas specifically includes: The system controls an intelligent inspection drone equipped with multi-source sensors to acquire multi-source sensing data from the surfaces of various components of the mountain bridge, and performs timestamp alignment and sensor extrinsic parameter calibration and fusion on the multi-source sensing data to generate a synchronous sensing data stream with spatial reference consistency. The synchronous sensing data stream is subjected to disease feature extraction and classification to obtain multi-type disease data identified in this inspection. When extracting disease features, airborne real-time dynamic differential positioning and inertial measurement unit are used to perform tightly coupled pose calculation to reconstruct the time-varying pose trajectory of the intelligent inspection drone. Based on the time-varying pose trajectory and the depth information of the pixels corresponding to the defects in the synchronous sensing data stream, the multi-type defect data are back-projected onto the global coordinate system to obtain the defect location data corresponding to each bridge defect.

[0007] Furthermore, based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archives. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined, specifically including: Using the spatial coordinate attributes of each bridge defect in the defect location data as anchor points, a set of candidate co-located defect entities is retrieved from the historical inspection archives according to a preset spatial neighborhood. The spatial overlap is obtained by performing a three-dimensional intersection-union ratio calculation on the point cloud of the defect contour corresponding to each bridge defect entity identified in the current inspection and the historical defect contour point cloud of the candidate co-located defect entity in the candidate co-located defect entity set. The pairing relationship between the current defect entity and the candidate co-located defect entity with the spatial overlap exceeding the preset threshold is selected, and the bridge defects belonging to the same pairing relationship are linked together according to the inspection timestamp to form a continuous evolution sequence of each bridge defect.

[0008] Furthermore, the temporal changes in the disease characteristic parameters in the continuous evolution sequence are determined using the following steps: Extract the characteristic parameters of the same bridge defect recorded during each inspection from the continuous evolution sequence, and construct a time-series vector of the characteristic parameters of the defect according to the inspection timestamp order; The time-series vector is subjected to a difference operation on the disease characteristic parameters of adjacent timestamps to obtain the time-series change of the disease characteristic parameters.

[0009] Furthermore, determining the visualization rendering coefficients of the corresponding bridge defects in the preset bridge 3D model based on the temporal changes of defect characteristic parameters in the continuous evolution sequence specifically includes: Using the temporal changes of the defect characteristic parameters in the continuous evolution sequence as independent variables, and substituting them into a preset nonlinear mapping function, the initial rendering intensity factor of each bridge defect is calculated. The initial rendering intensity factors of all bridge defects in this inspection were statistically analyzed, a histogram of rendering intensity distribution of the entire bridge was constructed, and the intensity normalization benchmark was determined based on the data quantiles of the histogram of rendering intensity distribution of the entire bridge. The initial rendering intensity factor of each bridge defect is normalized and compressed using the intensity normalization benchmark to obtain the visualization rendering coefficient of the corresponding bridge defect in the preset bridge 3D model.

[0010] Furthermore, when the temporal variation of any extracted bridge defect exceeds the allowable evolution range of its corresponding defect type, an overlay analysis is performed between the neighborhood structural surface information of the bridge defect in the bridge's three-dimensional model and the original design parameters of the mountainous mega-bridge. This specifically includes: Bridge defects that exceed the allowable evolution range are marked as out-of-limit defects. Using the defect location data of the out-of-limit defects as seed points, a region growth is performed along the topological connectivity direction of the structural surface in the preset three-dimensional bridge model to extract the neighborhood structural surface of the out-of-limit defects. The design reference geometric information corresponding to the neighboring structural surface is retrieved from the original design parameters of the mountain super-large bridge, and the design reference geometric information is spatially registered to the coordinate system of the preset bridge three-dimensional model; The three-dimensional deviation of the design reference geometric information after spatial registration and the measured geometric information of the neighboring structural surface is calculated to obtain the overlap distribution of the neighboring structural surface of the over-limit disease.

[0011] Secondly, this application provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones.

[0012] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones.

[0013] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones provided in this application first acquires multi-type defect data and corresponding defect location data collected by the intelligent inspection drone during the current inspection of the mountainous super-large bridge; secondly, based on the spatial coordinate attributes of the bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archives. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined; furthermore, the corresponding bridge defect is determined in the preset bridge 3D model based on the temporal change of defect characteristic parameters in the continuous evolution sequence. The visualization rendering coefficients are used to construct a fault density distribution hot zone in the 3D model of the bridge based on the visualization rendering coefficients. Then, the fault entities corresponding to the target density areas in the fault density distribution hot zone are extracted. When the temporal change of any extracted bridge fault exceeds the allowable evolution range of its corresponding fault type, the neighborhood structural surface information of the bridge fault in the 3D model of the bridge is subjected to an overlay analysis with the original design parameters of the mountain bridge. Finally, the allowable evolution range is updated based on the overlay analysis results, and the updated allowable evolution range is used as the anomaly judgment benchmark for the next inspection of the same bridge fault.

[0014] Therefore, this application can dynamically distinguish between normal damage development and abnormal accelerated deterioration in large mountain bridges. Firstly, acquiring multi-type defect data and corresponding defect location data collected by intelligent inspection drones during this inspection of large mountain bridges provides a data foundation for upgrading defect detection from two-dimensional images to three-dimensional indexable objects, eliminating the incomparability of defect locations caused by changes in the drone's shooting perspective. Secondly, based on the spatial coordinate attributes of bridge defects in the defect location data, spatial overlap matching is performed to determine the continuous evolution sequence, which can improve the determination of the identity of cross-temporal defects from relying on image feature matching. The method transforms into geometric entity overlap discrimination based on the intersection-union ratio of 3D contour point clouds, solving the problem of unreliable correlation due to inconsistent features of the same disease at the image level caused by changes in lighting and viewing angle. This establishes the longitudinal tracking of diseases on a stable and geometrically interpretable spatial benchmark, effectively distinguishing the causes of variation. Furthermore, based on the temporal changes in disease feature parameters in the continuous evolution sequence, visualization rendering coefficients are determined and disease density distribution hot zones are constructed. This involves converting disease evolution intensity into visual rendering intensity through a two-level mechanism of nonlinear mapping and full-bridge statistical normalization, and using this as a weight to reconstruct the structure. The weighted scalar field of disease density enables the distribution of hot zones on the 3D model to simultaneously reflect the clustering characteristics and individual evolution risks of diseases, achieving automatic highlighting of critical diseases and comparable allocation of rendering resources across batches. This effectively distinguishes between normal damage development and abnormal accelerated deterioration in major bridges. Furthermore, when the temporal variation of any bridge disease exceeds the allowable evolution range, it triggers the overlay analysis of neighboring structural surfaces. This automatically correlates local anomalies of the disease with the geometric deviation quantification at the component level, revealing derivative structural anomalies such as component deformation and settlement that may result from disease development. This provides a basis for updating the anomaly judgment criteria. This provides structural-level decision-making support beyond the parameters of the defects themselves. Finally, the allowable evolution range is updated based on the results of the differential analysis and used as the anomaly judgment benchmark for the next inspection. This is achieved by extracting the structural deviation vector through principal component analysis and using it to drive the adaptive correction of the boundary of the allowable evolution range. This allows the anomaly judgment threshold to evolve from a static empirical value to a dynamic benchmark that reflects the actual structural state of the component where the defect is located, avoiding false alarms and missed alarms caused by the insensitivity of fixed thresholds to local structural difference analysis. In summary, the technical solution provided in this application can dynamically distinguish between normal damage development and abnormal accelerated deterioration in large mountain bridges. Attached Figure Description

[0015] Figure 1 This is an exemplary flowchart of the comprehensive analysis method for major mountain bridges based on intelligent inspection drones, as shown in this application. Figure 2 This is an exemplary flowchart for determining time-series changes as shown in this application; Figure 3This is a schematic diagram of the computer device used to implement a comprehensive analysis method for large mountain bridges based on intelligent inspection drones, as shown in this application. Detailed Implementation

[0016] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] refer to Figure 1 The figure is an exemplary flowchart of the comprehensive analysis method for major bridges in mountainous areas based on intelligent inspection drones, as shown in this application. The figure mainly includes the following steps: In step S101, the intelligent inspection drone collects various types of damage data and corresponding damage location data during this inspection of the mountain bridge.

[0018] It should be noted that intelligent inspection drones refer to highly autonomous aerial vehicle systems designed for the structured inspection of large bridges in mountainous areas. They typically adopt quadcopter or hybrid wing configurations and integrate heterogeneous sensor payloads such as high-precision airborne RTK positioning modules, laser rangefinders, multispectral and high-resolution visible light gimbal cameras, inertial measurement units, and front-end AI inference chips. Through a preset three-dimensional flight path planning engine, they can achieve close-to-wall flight and hovering observation under the turbulent conditions of canyons by relying on vision-inertial-satellite tightly coupled fusion navigation. They can automatically avoid slender obstacles such as main cables and slings, and accurately collect images of defects and perform initial edge screening of key stress-bearing parts with limited accessibility, such as bridge towers and beam bottoms, thereby replacing manual high-altitude close-range inspections.

[0019] The following steps were used to acquire data on various types of defects and their corresponding location data collected by intelligent inspection drones during this inspection of a major bridge in a mountainous area: The system controls an intelligent inspection drone equipped with multi-source sensors to acquire multi-source sensing data from the surfaces of various components of the mountain bridge, and performs timestamp alignment and sensor extrinsic parameter calibration and fusion on the multi-source sensing data to generate a synchronous sensing data stream with spatial reference consistency. The synchronous sensing data stream is subjected to disease feature extraction and classification to obtain multi-type disease data identified in this inspection. When extracting disease features, airborne real-time dynamic differential positioning and inertial measurement unit are used to perform tightly coupled pose calculation to reconstruct the time-varying pose trajectory of the intelligent inspection drone. Based on the time-varying pose trajectory and the depth information of the pixels corresponding to the defects in the synchronous sensing data stream, the multi-type defect data are back-projected onto the global coordinate system to obtain the defect location data corresponding to each bridge defect.

[0020] In practice, firstly, an intelligent inspection drone equipped with a multi-source sensor integrated payload, including a visible light camera, an infrared thermal imager, and a lidar, is autonomously flown along a pre-planned three-dimensional route. Simultaneously, it collects high-resolution images, thermal infrared radiation temperature fields, and three-dimensional laser point clouds of the surfaces of various components of the mountain bridge, such as the bridge towers, main beams, and cables, forming multi-source sensing data covering the entire bridge. This multi-source sensing data refers to a set of heterogeneous observation data acquired in parallel by different measurement sensors during the same inspection flight. Secondly, the data collected by each sensor is synchronized and aligned according to the unified timestamp added by the GNSS second pulse timing module. Then, based on the installation extrinsic parameter matrix of each sensor relative to the center of the airborne inertial measurement unit, obtained through pre-calibration at the inspection field, the data is further processed. Multi-source data is uniformly transformed into a carrier system with the IMU as the origin, generating a synchronous sensing data stream that is both temporally and spatially aligned. This synchronous sensing data stream refers to a structured data sequence where observations from various sensors at the same time for the same target in the same field of view have been spatiotemporally registered. Furthermore, the synchronous sensing data stream is input into a semantic segmentation model based on a deep convolutional encoder-decoder network. The visible light image branch uses a ResNet backbone network to extract multi-scale feature maps, the thermal infrared image branch processes temperature distribution maps in parallel using the same network structure, and the lidar point cloud branch extracts three-dimensional geometric features using PointNet++. These three branches of features undergo cross-modal attention fusion in the feature pyramid network layer, followed by Softmax processing. The classifier predicts the semantic categories of defects such as cracks, spalling, exposed reinforcement, and honeycomb surface defects pixel by pixel and segments the defect contour regions, outputting multi-type defect data. This multi-type defect data refers to a structured record containing defect semantic category labels, pixel region masks, and their two-dimensional coordinate indices in the original image. Then, while the semantic segmentation model performs feature extraction, the airborne GNSS receiver continuously receives real-time dynamic differential corrections broadcast from the base station. These corrections are then tightly coupled with the three-axis angular rate and acceleration output by the IMU (Integrated Device Unit) using extended Kalman filtering for pose estimation. The IMU's high-frequency pre-integrated values ​​are used for state prediction, and GNSS carrier phase double-difference observations and visual odometry inter-frame pose estimation are used for measurement updates. This recursively estimates the UAV's global position. The six-degree-of-freedom pose at each sampling moment in the geocentric coordinate system is used to reconstruct a continuous and smooth time-varying pose trajectory. The time-varying pose trajectory refers to the sequence of position and attitude parameters arranged by timestamps during the movement of the UAV. Finally, for each disease instance in the multi-type disease data, the lidar depth channel ranging value corresponding to all pixels in its semantic segmentation mask in the synchronous perception data stream is taken, and the three-dimensional line-of-sight vector corresponding to the center pixel of the disease area is calculated. Combined with the camera pose parameters at that sampling moment in the time-varying pose trajectory, the two-dimensional image coordinates of the center pixel of the disease are back-projected to the global coordinate system through the collinearity equation. At the same time, the edge pixels of the disease contour are back-projected one by one to obtain the circumscribed contour point set of the disease in three-dimensional space, that is, the disease localization data.

[0021] It should be noted that the defect location data in this application refers to the three-dimensional spatial coordinates and outline range description of each bridge defect entity in the global coordinate system.

[0022] In step S102, based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archive. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined.

[0023] Based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archives. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined by the following steps: Using the spatial coordinate attributes of each bridge defect in the defect location data as anchor points, a set of candidate co-located defect entities is retrieved from the historical inspection archives according to a preset spatial neighborhood. The spatial overlap is obtained by performing a three-dimensional intersection-union ratio calculation on the point cloud of the defect contour corresponding to each bridge defect entity identified in the current inspection and the historical defect contour point cloud of the candidate co-located defect entity in the candidate co-located defect entity set. The pairing relationship between the current defect entity and the candidate co-located defect entity with the spatial overlap exceeding the preset threshold is selected, and the bridge defects belonging to the same pairing relationship are linked together according to the inspection timestamp to form a continuous evolution sequence of each bridge defect.

[0024] In specific implementation, firstly, the spatial coordinate attributes of each bridge defect in the defect location data obtained from this inspection are used as query anchor points. The spatial coordinate attributes refer to the three-dimensional center coordinates and outer envelope range parameters of the defect entity in the global coordinate system. A three-dimensional KD-tree index is constructed in the historical inspection archive database with this query anchor point as the center and according to a preset spatial neighborhood radius for range querying. This retrieves historical defect entities with spatially adjacent locations as candidate co-located defect entities, thus forming a set of candidate co-located defect entities. The set of candidate co-located defect entities refers to the set of historical inspection archives constructed with the spatial coordinate attributes of a bridge defect entity identified in the current inspection as the anchor point and according to a preset spatial neighborhood radius. Let the set of historical defect entity records retrieved by the spatial neighborhood radius be the set to be matched. For example, if the anchor point coordinates of a crack on the surface of a bridge tower are (100.5, 200.3, 50.2), and the neighborhood radius is set to 0.5 meters, then all historical crack records in the historical archives with a 3D Euclidean distance of less than 0.5 meters from this anchor point are included in the candidate set. Then, for each bridge defect entity identified in the current inspection, the outer contour point set formed by back-projecting the edge pixels of the defect from its defect location data to the global coordinate system is extracted. Poisson surface reconstruction is performed on this outer contour point set to generate the current defect contour point cloud. At the same time, each candidate defect entity is retrieved from the candidate co-located defect entity set. Historical disease contour point clouds stored in their respective inspection batches are selected for the same disease entities. The current disease contour point cloud is registered with each historical disease contour point cloud separately. The optimal rigid body transformation matrix between the two point clouds is solved using the iterative nearest point algorithm to complete the fine registration and alignment. The 3D intersection-union ratio of the two point clouds is calculated in the unified coordinate system after registration, which is the ratio of the intersection volume of the convex hulls of the two point clouds to the union volume. This ratio is used as the spatial overlap, which refers to the degree of volume overlap between the current disease entity and the historical disease entity in 3D space. Finally, a preset threshold of 0.6 is set, and the current disease entities with a spatial overlap exceeding this threshold are matched with the candidate co-located diseases. Entities identified as repeated observations of the same defect in different inspection batches are paired with the candidate co-located defect entity and the current corresponding defect entity to form a pairing relationship. This pairing relationship is then established, and the bridge defect records belonging to the same pairing relationship are linked together in ascending order of inspection timestamps to form a continuous evolution sequence of the bridge defect from its first discovery to the current inspection. The candidate co-located defect entity in the same pairing relationship refers to a specific historical defect entity in the set of candidate co-located defect entities whose spatial overlap with the current defect entity exceeds a preset threshold after the 3D intersection-union ratio of the defect contour point cloud is calculated. This entity is identified as a repeated observation record of the same bridge defect in previous inspection batches.

[0025] It should be noted that the continuous evolution sequence in this application refers to the tracking chain in which the defect feature parameters of the same bridge defect entity are arranged in chronological order in each inspection. This scheme uses the three-dimensional spatial coordinate attributes of the defect location data back-projected to the global coordinate system through the collinear equation as anchor points for spatial neighborhood retrieval, and regards the defect entity as a geometric object with three-dimensional volume. The three-dimensional intersection-union ratio is calculated by comparing the point cloud of the defect outline of the current inspection with the historical defect outline point cloud of the same defect in the historical inspection archive. The volume overlap is used as a quantitative index of spatial overlap to determine the identity of defects across time. This method upgrades from two-dimensional image feature similarity comparison to three-dimensional geometric entity overlap discrimination, effectively solving the technical problem that the same defect cannot be reliably associated due to the inconsistency of features at the image level caused by changes in the drone shooting perspective and differences in lighting conditions. This allows the cross-time tracking of defects to be carried out directly in the three-dimensional space of the bridge.

[0026] In step S103, the visualization rendering coefficient of the corresponding bridge disease in the preset bridge three-dimensional model is determined according to the temporal change of the disease characteristic parameters in the continuous evolution sequence, and a disease density distribution hot zone is constructed in the bridge three-dimensional model based on the visualization rendering coefficient.

[0027] refer to Figure 2 As shown, this figure is an exemplary flowchart for determining the amount of time-series change according to this application. The amount of time-series change of the disease characteristic parameters in the continuous evolution sequence described in this application is determined by the following steps: In step S1031, the characteristic parameters of the same bridge defect recorded during each inspection are extracted from the continuous evolution sequence, and a time-series vector of the characteristic parameters of the defect is constructed according to the inspection timestamp order. In step S1032, the disease characteristic parameters of adjacent timestamps in the time series vector are differentially calculated to obtain the time series change of the disease characteristic parameters.

[0028] In specific implementation, firstly, the defect feature parameters stored at the time of each identification of the same bridge defect are extracted one by one from the inspection records included in the continuous evolution sequence. These defect feature parameters refer to quantifiable geometric and morphological description indicators of the defect, such as crack length, crack width, spalling area, and exposed rebar area. The extracted feature parameters from each inspection are arranged in ascending order according to the inspection timestamp, constructing a time-series vector with timestamp as the index and feature parameter values ​​as elements. For example, the length records of a certain crack defect in three inspections are 12.3 mm, ... 14.7 mm and 18.2 mm correspond to timestamps of January 2024, June 2024, and January 2025, respectively, so their time series vector is [12.3, 14.7, 18.2]. First-order backward difference operations are performed sequentially on the feature parameter values ​​of two adjacent timestamps in the constructed time series vector. That is, the feature parameter value of the later timestamp is subtracted from the feature parameter value of the earlier timestamp to obtain the change in disease feature parameters within each adjacent inspection interval. These changes are then arranged sequentially according to their corresponding time intervals to form the time series change of disease feature parameters.

[0029] It should be noted that, in this application, the time-series change refers to the net change sequence of the same type of characteristic parameter of the same disease within the time interval between two adjacent inspections, which is used to quantify the evolution rate and magnitude of the disease.

[0030] The visualization rendering coefficients of corresponding bridge defects in the preset bridge 3D model are determined based on the temporal changes of defect characteristic parameters in the continuous evolution sequence using the following steps: Using the temporal changes of the defect characteristic parameters in the continuous evolution sequence as independent variables, and substituting them into a preset nonlinear mapping function, the initial rendering intensity factor of each bridge defect is calculated. The initial rendering intensity factors of all bridge defects in this inspection were statistically analyzed, a histogram of rendering intensity distribution of the entire bridge was constructed, and the intensity normalization benchmark was determined based on the data quantiles of the histogram of rendering intensity distribution of the entire bridge. The initial rendering intensity factor of each bridge defect is normalized and compressed using the intensity normalization benchmark to obtain the visualization rendering coefficient of the corresponding bridge defect in the preset bridge 3D model.

[0031] In practical implementation, firstly, the temporal changes of the determined disease characteristic parameters are used as independent variables input into a preset nonlinear mapping function. This nonlinear mapping function adopts a Sigmoid variant function form, and its expression is as follows: ,in Let θ be the temporal variation, k be the steepness control parameter, and θ be the offset parameter, taking the median value of the allowable evolution range for this type of defect. The function's output value accelerates towards 1.0 when the temporal variation approaches the upper limit of the allowable evolution range, and remains in a flat zone below 0.3 when there are small fluctuations within the range. This function maps the temporal variation of each bridge defect to an initial rendering intensity factor within the range (0,1). The initial rendering intensity factor refers to the original mapped value of the severity of defect evolution; a larger value indicates that the defect's evolution is closer to or exceeds the allowable boundary. Secondly, the initial rendering intensity factors calculated by the nonlinear mapping function for all bridge defects in this inspection are statistically analyzed. The initial rendering intensity factor values ​​are divided into intervals with a group interval of 0.05, and the frequency of each interval is statistically analyzed. A graph is constructed with the horizontal axis representing rendering intensity and the vertical axis representing... The histogram of the rendering intensity distribution of the number of bridge defects is used. After sorting the histogram data from low to high, the rendering intensity values ​​corresponding to the 90th and 98th percentiles are taken as the lower and upper bounds of the intensity normalization benchmark, respectively. The intensity normalization benchmark refers to the upper and lower bound reference values ​​used to linearly compress the original rendering intensity factor to the standard rendering range. Then, the min-max normalization method is used, with the lower and upper bounds of the intensity normalization benchmark as the lower and upper limits of the normalization target range, to linearly compress and map the initial rendering intensity factor of each bridge defect. Specifically, the initial rendering intensity factor value is subtracted from the normalization lower bound and then divided by the difference between the normalization upper and lower bounds. The result is truncated to the [0,1] interval, and finally, the visualization rendering coefficient of each bridge defect in the preset bridge 3D model is obtained.

[0032] It should be noted that the visualization rendering coefficient in this application refers to the normalized driving parameter used to control the color density and transparency of the disease in the 3D model. In this embodiment, the temporal variation is first converted into an initial rendering intensity factor through a preset Sigmoid variant nonlinear mapping function. This function has a gradient amplification characteristic when the temporal variation approaches the boundary of the allowable evolution interval, so that the disease close to or exceeding the safety threshold receives a significantly enhanced rendering intensity, while maintaining a low intensity output within the safety interval. This achieves automatic highlighting of critical diseases at the rendering level. On this basis, the initial rendering intensity factor of all diseases in the whole bridge is statistically analyzed using a distribution histogram. The intensity normalization benchmark is dynamically determined based on the data quantiles. Then, normalization compression is performed to obtain the final visualization rendering coefficient. This global statistical normalization step solves the rendering threshold mismatch problem caused by the difference in the overall evolution level of diseases between different inspection batches. This makes the hot zone distribution map generated by each inspection comparable across batches, ensuring that the allocation of rendering resources is always concentrated on the subset of diseases with the most drastic evolution in the current batch.

[0033] Based on the aforementioned visualization rendering coefficients, the following steps are used to construct the defect density distribution heatmap in the 3D model of the bridge: Using the spatial coordinate attributes in the location data of each bridge defect as the field source center, and the visualization rendering coefficient of the corresponding bridge defect as the field source intensity weight, a weighted defect density scalar field is constructed on the structural surface of the preset bridge three-dimensional model. The weighted disease density scalar field is subjected to isosurface extraction on the structural surface mesh of the preset bridge three-dimensional model to obtain the boundary contour of the disease density distribution hot zone; The boundary contour of the disease density distribution hot zone is mapped onto the preset bridge 3D model, and the mesh patches within the boundary contour are filled with color levels according to the field value gradient of the weighted disease density scalar field to construct the disease density distribution hot zone.

[0034] In specific implementation, firstly, the three-dimensional spatial coordinate attributes recorded in the defect location data of each bridge defect are used as the center point of the field source. The spatial coordinate attributes refer to the three-dimensional center coordinates of the defect entity in the global coordinate system. The corresponding visualization rendering coefficient is used as the intensity weight of this field source point. On the structural surface of the preset bridge three-dimensional model, a Gaussian kernel function is used to spatially diffuse and superimpose the intensity weight of each field source point. The bandwidth parameter of the Gaussian kernel function is between 0.3 meters and 0.8 meters and is positively correlated with the radius of the circumcircle of the defect outline. By traversing all defect field source points, the weighted intensity is accumulated at each vertex of the triangular mesh on the structural surface, thus constructing... A weighted scalar field of disease density covering the surface of the bridge structure is constructed. This weighted scalar field refers to the continuous scalar distribution of the spatial density of diseases on the structural surface, weighted by visualization rendering coefficients. Then, the moving cube algorithm is used to extract isosurfaces from this weighted scalar field on the structural surface mesh of a pre-defined 3D bridge model. Specifically, the isosurface extraction threshold is set to the median value of the scalar field. Grid vertices above this threshold are marked as internal points, and those below are marked as external points. Linear interpolation along the boundary of the triangular mesh determines the intersection points of the isosurfaces between internal and external points. Connecting these intersection points forms a closed triangular facet. A set of triangular facets is used, the outer boundary of which forms the boundary contour of the disease density distribution hot zone. This boundary contour refers to the dividing line between the area where the disease density is significantly higher than the average level of the entire bridge and the surrounding area. Finally, the extracted boundary contour of the disease density distribution hot zone is mapped back to the texture space of the preset bridge 3D model using UV coordinates. Within the structural surface mesh area enclosed by the boundary contour, color gradation rendering is performed based on the gradient direction and amplitude of the weighted disease density scalar field in that area. Color is assigned to each facet using a gradient mapping table where the minimum field value corresponds to a cool blue tone and the maximum field value corresponds to a warm red tone. The rendered colored area is the hot zone of disease density distribution. The gradient mapping table is constructed using hue gradient based on the HSL color space. Blue corresponds to the minimum scalar field value, cyan corresponds to the low to medium field value, yellow corresponds to the high to medium field value, and red corresponds to the maximum field value. Hue is linearly interpolated at equal intervals along the direction of increasing field value. At the same time, saturation is kept at 100% and brightness is kept at 50% throughout to eliminate the influence of brightness interference on visual perception, forming a four-segment gradient mapping table of blue-cyan-yellow-red. This ensures that the rendering color gradation of the density hot zone on the 3D model can strictly correspond to the field value gradient. This will not be elaborated further here.

[0035] It should be noted that, in this application, the disease density distribution hot zone refers to the highlighted area on the three-dimensional model that intuitively expresses the density and severity of disease evolution using color gradations. In this application, the disease density distribution hot zone is constructed by superimposing Gaussian kernel function spatial diffusion on the structural surface using the visualization rendering coefficient of each disease as the field source intensity weight, so that the delineation of the hot zone boundary contour depends not only on the spatial aggregation degree of the disease, but also on the modulation of the evolution intensity of each disease point itself. That is, the disease near the boundary of the allowable evolution interval will form a wider range and higher level of field value contribution due to its high rendering coefficient, while the field value contribution of minor diseases is suppressed. Thus, a composite hot zone that can reflect the clustering characteristics and individual evolution risk of diseases is formed on the three-dimensional model of the bridge, realizing the accurate identification and automatic highlighting of the weak areas of bridge structural safety.

[0036] In step S104, the disease entities corresponding to the target density region in the disease density distribution hot zone are extracted. When the temporal change of any extracted bridge disease exceeds the allowable evolution range of its corresponding disease type, the neighborhood structure surface information of the bridge disease in the bridge three-dimensional model is subjected to overlap analysis with the original design parameters of the mountain super bridge.

[0037] In specific implementation, in the pre-defined 3D bridge model with completed hot zone construction of disease density distribution, the maximum value of the weighted disease density scalar field is used as the retrieval anchor point. Gradient descent search is performed along the topological connectivity direction of the structural surface mesh. The set of all mesh patches covered by the target density region is extracted with the field value decreasing to 60% of the maximum field value as the cutoff boundary. The spatial coordinate attributes of each bridge disease are queried one by one from the disease location data to see if they fall within the projection range of the set of mesh patches in the global coordinate system. The bridge disease entities that fall within the range are output as the disease entities corresponding to the target density region. The target density region refers to the subset of structural surface mesh patches defined in the weighted disease density scalar field with the maximum field value as the starting center, through gradient descent search, and with 60% of the maximum field value as the cutoff boundary. The field value in this region is significantly higher than the median level of the weighted disease density scalar field of the whole bridge, representing the weak part of the bridge structure where the double superposition effect of disease evolution intensity and spatial aggregation is most concentrated.

[0038] It should be noted that the disease entities corresponding to the target density region mentioned in this application refer to the set of disease objects that make a dominant contribution.

[0039] When the temporal variation of any extracted bridge defect exceeds the allowable evolution range of its corresponding defect type, the following steps are used to perform an overlay analysis between the neighborhood structural surface information of the bridge defect in the bridge's three-dimensional model and the original design parameters of the mountainous mega-bridge: Bridge defects that exceed the allowable evolution range are marked as out-of-limit defects. Using the defect location data of the out-of-limit defects as seed points, a region growth is performed along the topological connectivity direction of the structural surface in the preset three-dimensional bridge model to extract the neighborhood structural surface of the out-of-limit defects. The design reference geometric information corresponding to the neighboring structural surface is retrieved from the original design parameters of the mountain super-large bridge, and the design reference geometric information is spatially registered to the coordinate system of the preset bridge three-dimensional model; The three-dimensional deviation of the design reference geometric information after spatial registration and the measured geometric information of the neighboring structural surface is calculated to obtain the overlap distribution of the neighboring structural surface of the over-limit disease.

[0040] In specific implementation, firstly, bridge defects whose temporal changes exceed the upper or lower limit of the allowable evolution range for their corresponding defect type are marked as "excessive defects." Excessive defects refer to defect entities whose evolution has exceeded the safety boundary of the defect type. Using the three-dimensional spatial coordinates in the defect location data as seed points, a region growth algorithm based on mesh topological connectivity is executed on the triangular mesh of the pre-set bridge three-dimensional model's structural surface. Specifically, starting with the triangular facet containing the seed point, expansion occurs outward along the shared edge direction of adjacent facets. The expansion terminates when the facet curvature change exceeds a pre-set curvature threshold or the geodesic distance from the seed point reaches a radius threshold. The set of connected facets meeting these conditions is extracted as the neighborhood structural surface of the excessive defect. The neighborhood structural surface refers to the local component geometric description surface extending along the structural surface centered on the excessive defect. Then, from the original design parameter files of the mountainous super-large bridge, i.e., the BIM model or the digitized design drawings, the design reference geometric information corresponding to the neighborhood structural surface is retrieved based on both component coding and spatial location conditions. The reference geometric information refers to the three-dimensional geometric shape, size, and spatial pose parameters of bridge components in the design state. A two-stage point cloud registration process is adopted, consisting of coarse registration using the four-point method and fine registration using iterative nearest point. Initial matching point pairs are constructed by using the boundary corners of the neighboring structural surfaces and the corresponding corners of the design reference geometric information to solve the rotation and translation matrices for coarse registration. Then, iterative nearest point fine registration is performed using all vertices of the neighboring structural surfaces as the source point set and the sampling points of the design reference geometric information as the target point set, spatially registering the design reference geometric information to the global coordinate system of the preset bridge 3D model. Finally, in a unified coordinate system, the three-dimensional deviation between the spatially registered design reference geometric information and the measured geometric information of the neighboring structural surfaces is calculated. Specifically, each grid vertex of the neighboring structural surface is used as a sampling point, and a ray is drawn along the normal direction of the vertex to the corresponding surface of the design reference geometric information to find the intersection. The Euclidean distance between the intersection point and the sampling point is calculated as the deviation value at the sampling point. After traversing all grid vertices of the neighboring structural surfaces, a set of deviation values ​​with spatial distribution characteristics is obtained, namely the overlapping distribution of the neighboring structural surfaces with excessive defects.

[0041] It should be noted that the overlap distribution in this application refers to the three-dimensional spatial deviation field between the actual geometric shape of the structural surface in the vicinity of the over-limit defect and the geometric shape of the design reference. Unlike the conventional handling method in the prior art, which only records the deformation parameters of the over-limit defect itself in isolation or directly triggers alarms with fixed rules, this solution does not take the defect itself as the analysis endpoint. Instead, it uses the spatial coordinate attributes of the over-limit defect as the seed point, performs region growth along the topological connectivity direction of the structural surface, extracts the vicinity structural surface, and performs three-dimensional deviation calculation on a vertex-by-vertex basis with the measured geometric information of the vicinity structural surface and the design reference geometric information retrieved from the original design parameters and spatially registered. This yields an overlap distribution with spatial distribution characteristics. This approach identifies the local evolution of the defect within the overall geometric deviation field of the component it is attached to, enabling the quantitative display of derivative structural anomalies such as component deformation, settlement, or displacement that may result from the development of the defect. This provides a decision-making basis that goes beyond the dimension of the defect's own characteristic parameters and can reflect changes in the structural state at the component level for subsequent updates to the allowable evolution range.

[0042] In step S105, the allowable evolution range is updated based on the overlap analysis results, and the updated allowable evolution range is used as the anomaly judgment benchmark when inspecting the same bridge defect again.

[0043] The allowable evolution interval is updated based on the overlap analysis results. The updated allowable evolution interval is then used as the anomaly judgment criterion for the next inspection of the same bridge defect. This is achieved through the following steps: Principal component analysis was performed on the results of the overlap analysis to extract the maximum deviation direction and magnitude of the structural surface of the neighborhood of the over-limit disease, and a structural deviation vector was constructed. The structural deviation vector is correlated with the current allowable evolution interval of the disease type corresponding to the over-limit disease, and the interval boundary correction increment is calculated. The interval boundary correction increment is positively correlated with the deviation magnitude and expands unidirectionally along the direction of the maximum deviation. The current permissible evolution interval is shifted by the interval boundary correction increment to obtain the updated permissible evolution interval. The updated permissible evolution interval is then bound and stored with the defect identifier of the over-limit defect, and used as the anomaly judgment benchmark when inspecting the same bridge defect again.

[0044] In specific implementation, firstly, a data matrix is ​​constructed from the overlapping distribution data of the neighborhood structural surfaces of the obtained over-limit disease. This matrix uses each grid vertex of the neighborhood structural surface as a sample and the three-dimensional deviation components of each vertex as variables. Principal component analysis is performed on this matrix to calculate the covariance matrix of the deviation components and solve for the eigenvalues ​​and eigenvectors of this covariance matrix. The eigenvector corresponding to the largest eigenvalue is extracted as the direction of maximum deviation, and the range of the projected coordinates of all sample points along this direction is used as the magnitude of the deviation. The direction of maximum deviation and the magnitude of the deviation are combined to construct a structural deviation vector. This structural deviation vector refers to the vectorized representation of the dominant direction and magnitude of the geometric deviation of the neighborhood structural surface of the over-limit disease. Then, the structural deviation vector is... The current permissible evolution interval of the disease type corresponding to the quantity and the disease exceeding the limit is correlated and calculated. The current permissible evolution interval is defined as a numerical interval with the time-series change of disease characteristic parameters as the horizontal axis and the upper and lower limits of the permissible range as the boundaries. When calculating the interval boundary correction increment, the deviation magnitude is converted into a correction magnitude value in units of change through a preset linear mapping coefficient. This linear mapping coefficient is the empirical conversion ratio of the change of disease characteristic parameters corresponding to the unit deviation of the neighboring structural surface. For example, when the deviation magnitude of the neighboring structural surface in the direction of the maximum deviation is 5 mm, the empirical statistical value of the correction magnitude value of the time-series change of crack width is 0.2 mm, then the linear mapping coefficient is taken as 0.04 (i.e. 0.2 mm ÷ The value of 5 mm indicates that for every 1 mm of geometric deviation in the neighboring structural surface, the allowable evolution interval boundary of the crack width is adjusted by 0.04 mm. In practical applications, this linear mapping coefficient is categorized according to disease type: 0.02 to 0.05 for cracks, 0.08 to 0.15 for spalling, and 0.10 to 0.20 for exposed reinforcement. After each round of inspection, the coefficient is updated using least-squares regression based on newly added samples of overlap-disease changes. The correction magnitude is directly proportional to the magnitude of the deviation. Simultaneously, the direction of the maximum deviation is projected from the three-dimensional spatial direction to the one-dimensional direction of the change in disease characteristic parameters to determine the positive or negative polarity of the correction increment. The interval boundary is then adjusted along this polarity direction. One-way correction is performed to obtain the interval boundary correction increment, which refers to the magnitude and direction parameters of the allowable evolution interval boundary movement caused by structural deviation. Finally, the upper and lower bounds of the current allowable evolution interval are algebraically summed with the interval boundary correction increment. If the projection of the maximum deviation direction on the dimension of the change in the defect characteristic parameters is positive, the upper bound is moved up and the lower bound remains unchanged. If the projection is negative, the lower bound is moved down and the upper bound remains unchanged. The new interval after translation is the updated allowable evolution interval. The updated allowable evolution interval and the defect identifier of the defect exceeding the limit are bound and stored in the historical inspection archive database as the anomaly judgment benchmark when inspecting the same bridge defect again.

[0045] It should be noted that the abnormality judgment benchmark in this application refers to the threshold range standard for judging whether the temporal change of the disease is abnormal in the next inspection cycle.

[0046] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones.

[0047] In some embodiments, reference Figure 3 The figure is a schematic diagram of the structure of a computer device for implementing a comprehensive analysis method for major mountain bridges based on intelligent inspection drones, according to some embodiments of this application. The comprehensive analysis method for major mountain bridges based on intelligent inspection drones in the above embodiments can be implemented through... Figure 3 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0048] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the comprehensive analysis method for mountain bridges based on intelligent inspection drones in this application.

[0049] The communication bus 302 can be used to transmit information between the aforementioned components.

[0050] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0051] The memory 303 stores program code for executing the scheme of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones can be achieved by the processor 301 and one or more software modules in the program code in the memory 303.

[0052] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0053] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0054] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0055] In addition, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned comprehensive analysis method for mountain super-large bridges based on intelligent inspection drones.

[0056] Although preferred embodiments of this application have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0057] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application.

Claims

1. A comprehensive analysis method for extra-large bridges in mountainous areas based on intelligent inspection drones, characterized in that, Includes the following steps: Acquire data on various types of defects and their corresponding location data collected by intelligent inspection drones during this inspection of a major bridge in a mountainous area. Based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archive. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined. The visualization rendering coefficient of the corresponding bridge disease in the preset bridge 3D model is determined based on the temporal change of the disease characteristic parameters in the continuous evolution sequence, and a disease density distribution hot zone is constructed in the bridge 3D model based on the visualization rendering coefficient. Extract the disease entities corresponding to the target density region in the disease density distribution hot zone. When the temporal change of any extracted bridge disease exceeds the allowable evolution range of its corresponding disease type, perform an overlay analysis between the neighborhood structural surface information of the bridge disease in the bridge three-dimensional model and the original design parameters of the mountain super bridge. The allowable evolution range is updated based on the results of the overlay analysis, and the updated allowable evolution range is used as the anomaly judgment benchmark when inspecting the same bridge defects again.

2. The method as described in claim 1, characterized in that, The data collected by the intelligent inspection drone during this inspection of a major bridge in the mountainous area includes various types of defects and their corresponding location data. The system controls an intelligent inspection drone equipped with multi-source sensors to acquire multi-source sensing data from the surfaces of various components of the mountain bridge, and performs timestamp alignment and sensor extrinsic parameter calibration and fusion on the multi-source sensing data to generate a synchronous sensing data stream with spatial reference consistency. The synchronous sensing data stream is subjected to disease feature extraction and classification to obtain multi-type disease data identified in this inspection. When extracting disease features, airborne real-time dynamic differential positioning and inertial measurement unit are used to perform tightly coupled pose calculation to reconstruct the time-varying pose trajectory of the intelligent inspection drone. Based on the time-varying pose trajectory and the depth information of the pixels corresponding to the defects in the synchronous sensing data stream, the multi-type defect data are back-projected onto the global coordinate system to obtain the defect location data corresponding to each bridge defect.

3. The method as described in claim 1, characterized in that, Based on the spatial coordinate attributes of bridge defects in the defect location data, the spatial overlap of each bridge defect entity identified in the current inspection is matched with the corresponding defect entities in the historical inspection archives. When the matching degree exceeds a preset threshold, the continuous evolution sequence of each bridge defect is determined, specifically including: Using the spatial coordinate attributes of each bridge defect in the defect location data as anchor points, a set of candidate co-located defect entities is retrieved from the historical inspection archives according to a preset spatial neighborhood. The spatial overlap is obtained by performing a three-dimensional intersection-union ratio calculation on the point cloud of the defect contour corresponding to each bridge defect entity identified in the current inspection and the historical defect contour point cloud of the candidate co-located defect entity in the candidate co-located defect entity set. The pairing relationship between the current defect entity and the candidate co-located defect entity with the spatial overlap exceeding the preset threshold is selected, and the bridge defects belonging to the same pairing relationship are linked together according to the inspection timestamp to form a continuous evolution sequence of each bridge defect.

4. The method as described in claim 1, characterized in that, The temporal changes in the disease characteristic parameters in the continuous evolution sequence are determined using the following steps: Extract the characteristic parameters of the same bridge defect recorded during each inspection from the continuous evolution sequence, and construct a time-series vector of the characteristic parameters of the defect according to the inspection timestamp order; The time-series vector is subjected to a difference operation on the disease characteristic parameters of adjacent timestamps to obtain the time-series change of the disease characteristic parameters.

5. The method as described in claim 1, characterized in that, Determining the visualization rendering coefficients of corresponding bridge defects in a preset 3D bridge model based on the temporal changes of defect characteristic parameters in the continuous evolution sequence specifically includes: Using the temporal changes of the defect characteristic parameters in the continuous evolution sequence as independent variables, and substituting them into a preset nonlinear mapping function, the initial rendering intensity factor of each bridge defect is calculated. The initial rendering intensity factors of all bridge defects in this inspection were statistically analyzed, a histogram of rendering intensity distribution of the entire bridge was constructed, and the intensity normalization benchmark was determined based on the data quantiles of the histogram of rendering intensity distribution of the entire bridge. The initial rendering intensity factor of each bridge defect is normalized and compressed using the intensity normalization benchmark to obtain the visualization rendering coefficient of the corresponding bridge defect in the preset bridge 3D model.

6. The method as described in claim 1, characterized in that, When the temporal variation of any extracted bridge defect exceeds the allowable evolution range of its corresponding defect type, an overlay analysis is performed between the neighborhood structural surface information of the bridge defect in the bridge's 3D model and the original design parameters of the mountainous mega-bridge. Specifically, this includes: Bridge defects that exceed the allowable evolution range are marked as out-of-limit defects. Using the defect location data of the out-of-limit defects as seed points, a region growth is performed along the topological connectivity direction of the structural surface in the preset three-dimensional bridge model to extract the neighborhood structural surface of the out-of-limit defects. The design reference geometric information corresponding to the neighboring structural surface is retrieved from the original design parameters of the mountain super-large bridge, and the design reference geometric information is spatially registered to the coordinate system of the preset bridge three-dimensional model; The three-dimensional deviation of the design reference geometric information after spatial registration and the measured geometric information of the neighboring structural surface is calculated to obtain the overlap distribution of the neighboring structural surface of the over-limit disease.

7. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the comprehensive analysis method for mountainous super-large bridges based on intelligent inspection drones as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the comprehensive analysis method for mountain super-large bridges based on intelligent inspection drones as described in any one of claims 1 to 6.

Images