A bridge member identification method based on unmanned aerial vehicle point cloud reconstruction and three-dimensional synthetic data

Abstract

A bridge component identification method based on unmanned aerial vehicle point cloud reconstruction and three-dimensional synthetic data, point cloud is generated by unmanned aerial vehicle video acquisition, combined with improved PointNet++ model to realize accurate identification of bridge components, and a technical process of "data acquisition-point cloud generation-model training-component identification" is constructed. The specific implementation steps are as follows: ① using unmanned aerial vehicle to collect bridge component field video; ② building a three-dimensional reconstruction framework to generate dense point cloud of bridge component; ③ batch generation of diversified bridge point cloud synthetic data. ④ construct a labeled point cloud semantic segmentation dataset, which is used for model training after preprocessing and enhancement; ⑤ realize point cloud recognition based on improved PointNet++ model; ⑥ post-processing of the recognition result, extracting the geometric parameters of the component for bridge structure state evaluation. The present application can realize efficient generation and high-precision identification of bridge component point cloud, and provide reliable data support for bridge structure quality evaluation.

Description

Technical Field

[0001] This invention relates to a method for assessing the quality of bridge structures, specifically a method for reconstructing and identifying three-dimensional point clouds of bridge components based on UAV data acquisition and computer vision technology, belonging to the field of structural engineering. Background Technology

[0002] Bridges, as a vital component of modern transportation networks, play an irreplaceable role in ensuring the integrity and convenience of these networks. As of 2023, my country had 1.07 million highway bridges in service, and their safe operation is directly related to public safety. However, with the increasing service life of bridges, problems such as natural aging, cracking, and damage inevitably arise, leading to structural damage and a decline in load-bearing capacity. To address the issue of bridge structural aging, structural health testing has become an effective means of ensuring the safe operation of bridges.

[0003] Traditional inspection methods, such as manual visual inspection and telescopic inspection, suffer from low efficiency and poor accuracy, making them unsuitable for practical applications. In recent years, with the rise of machine vision technology, two-dimensional close-up image analysis has been widely used in bridge structure inspection. However, due to the lack of three-dimensional spatial context information, this method struggles to correlate with specific structural components, leading to inaccurate damage localization and hindering the assessment of the overall health of the bridge. Three-dimensional point cloud technology can clearly present the three-dimensional spatial information, surface condition, and structural parameters of components, providing a reliable basis for accurate component classification and feature extraction. However, current point cloud datasets for bridges are not only limited in number and coverage of bridge types, but also suffer from high data acquisition costs and insufficient annotation accuracy, resulting in low accuracy of trained point cloud semantic recognition models. Therefore, there is an urgent need to develop a high-precision, low-cost point cloud semantic recognition method to extract features from bridge components through spatial segmentation and category determination, providing technical support for the long-term maintenance and operational safety of bridges. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method for generating and recognizing 3D point clouds of bridge components based on unmanned aerial vehicles (UAVs), thereby improving the feasibility of point cloud-based bridge structure inspection in practical applications. The specific content includes:

[0005] A method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data, characterized by the following steps:

[0006] A. Use drones to collect videos of bridge components on site;

[0007] B. Establish a framework for a three-dimensional reconstruction method of bridge components, and use the video of the bridge components to reconstruct a three-dimensional point cloud model of the bridge components;

[0008] B1. Extract keyframes from the video of the bridge components using FFmpeg software;

[0009] B2. Process the keyframe images using COLMAP software to output a three-dimensional point cloud of the bridge components;

[0010] C. Generation of large-scale 3D point cloud synthetic data for bridges;

[0011] C1. Set bridge component parameters (such as piers, beams, etc.), perform random parameter perturbation to generate diverse models, and convert them into mesh models. Use uniform sampling to generate triangular facet sampling points; finally, add noise simulation (Gaussian noise, block occlusion, missing data, etc.) to enhance data diversity.

[0012] C2. Perform UAV-simulated laser scanning on the triangulated model. By planning a virtual flight path, simulate the laser emission angle, position, and reception time, calculate the time of flight (ToF) to reconstruct the spatial position of the point cloud and generate a 3D point cloud of the bridge.

[0013] C3. Repeat the above operations to obtain a large amount of diverse bridge point cloud composite data;

[0014] D. Construct a semantic recognition dataset of point clouds of bridge components;

[0015] D1. Point Cloud Data Preprocessing: The bridge point cloud synthetic data is subjected to three-dimensional normalization processing, the centroid of the point cloud is calculated to achieve centering, and then the point cloud is scaled to a unit sphere according to the maximum radius to ensure the geometric consistency of the data; a nine-dimensional input feature vector is constructed, which includes normalized XYZ coordinates (core geometric features), normalized RGB channel values ​​(color features), and normal vector features (key spatial features), and the feature matrix is ​​reorganized into the "9×N" format required by the model (N is the number of points);

[0016] D2. Data Labeling and Classification: For the main components of the bridge, professional 3D labeling tools are used for manual and detailed labeling to clarify their category classification; for a small number of components, their inherent shape features are first used for automatic pre-labeling by algorithms, and then rigorously verified and corrected manually to ensure the accuracy of the labeling results to the greatest extent; the point cloud dataset is divided into training set and test set according to a preset ratio, adhering to the basic principle that the proportion of points of each component category in the training set and test set is consistent, with the training set accounting for 80% of the total data and the test set accounting for 20% of the total data;

[0017] D3. Data Augmentation: Various data augmentation operations are performed on the training set to expand the sample diversity. Basic augmentation methods include master rotation (simulating different observation angles), XY plane rotation (simulating horizontal changes), coordinate jitter (introducing slight positional perturbations), random rotation (all-around viewpoint changes), scale scaling (simulating distance changes), and local translation (simulating local positional shifts). For small sample component categories with few points, additional and more intense targeted augmentations are performed to significantly increase the diversity of their features.

[0018] E. Build an improved PointNet++ point cloud recognition model for bridge components;

[0019] E1. Network Structure Design and Optimization: Based on the PointNet++ architecture, the model is built by integrating the classification, segmentation and utils modules. The geometric relationship learning ability is enhanced by multi-dimensional feature fusion, which improves the model's adaptability to bridge scenarios.

[0020] E2. Training Strategy and Execution Process: During training, a balanced sampler is used to dynamically adjust sample weights, and an enhanced weighted loss function is used to impose additional penalties on small sample misprediction and low IoU categories; an initial learning rate is set, and a scheduling method combining linear warm-up and cosine annealing is adopted, along with a gradient pruning strategy to ensure training stability; the test set mIoU is used as the core indicator, and an early stopping mechanism is triggered when there is no improvement for several consecutive rounds, and finally the category prediction result of the optimal model is output to complete the point cloud recognition of bridge components;

[0021] F. Post-processing and application of identification results: Calculate geometric parameters to provide data support for the assessment of the health status of bridge structures.

[0022] Furthermore, the method for acquiring bridge component video in step A is as follows: using a drone to perform a 360° omnidirectional surround of the target component and acquire video of the target component. The video acquisition requirements include information of the component at different heights from top to bottom, and finally collect the bridge component video.

[0023] Furthermore, in step B1, FFmpeg will be used to extract key frame images of the bridge components from the video. By selecting 0.25 as the time scaling factor, it will be ensured that the output key frames contain as much feature information of the bridge components as possible.

[0024] Furthermore, in step B2, the output of the three-dimensional point cloud of the bridge component is achieved using COLMAP software: COLMAP software uses the SIFT algorithm to extract image features, uses KNN, LoweRatio, and RANSAC algorithms to perform feature matching, combines the SfM algorithm to reconstruct the sparse point cloud of the target component, and finally uses the MVS algorithm to reconstruct the dense point cloud.

[0025] Furthermore, in step C1, the parameters are randomly perturbed to generate diverse models: by extracting the basic geometric parameters of bridge components (such as pier diameter / height, box girder cross-sectional dimensions) and applying random perturbations to the parameters, model diversification is achieved; by controlling the perturbation intensity (dimension ±5% tolerance, position offset <10cm, angle deflection <3°), the degree of morphological variation of the generated model can be adjusted; by configuring a reasonable perturbation threshold range, the generated component model can cover common deviation morphologies in actual engineering while maintaining structural rationality, so as to output training-level point cloud data with strong generalization ability.

[0026] Furthermore, the data augmentation operations in step D3 include basic augmentation and small-sample targeted augmentation; basic augmentation is performed on all samples and includes operations such as rotation, scaling, and Gaussian noise; while small-sample targeted augmentation doubles the augmentation intensity for small sample categories (such as cables and anchors), including:

[0027] Δ~U(-0.05λ,0.05λ)(2)

[0028] Where xyz′ represents the enhanced 3D coordinates, xyz i For the original 3D coordinates, m i s is a local deformation mask (randomly selected 40% of the points), s is the scaling factor along the principal axis (0.8-1.2, calculated by principal component analysis PCA), and Δ is the translation amount, which enhances the morphological diversity of small samples.

[0029] Furthermore, the multi-dimensional feature fusion mechanism in step E1 includes local coordinate normalization, RGB feature standardization, and coordinate gradient feature extraction. Local coordinate normalization enhances the expression of local geometric features through decentralization and scale scaling, and can be expressed as:

[0030]

[0031] Among them, xyz i The original 3D coordinates, xyz center Let max be the coordinates of the center point of the point cloud. dist xyz represents the distance from the farthest point in the point cloud to the center point (used for scale normalization). local These are the normalized local coordinates (mapped to the unit sphere); RGB feature normalization can normalize color features in the 0-1 range, handling the influence of illumination changes, expressed as:

[0032]

[0033] Among them, rgb iThe original RGB values, rgb min , rgb max These represent the minimum and maximum values ​​of the RGB channels in the point cloud, respectively. norm The standardized RGB features; coordinate gradient feature extraction calculates the local continuity features of the point cloud through neighborhood difference, simulating normal vector information, and can be expressed as follows:

[0034]

[0035] Among them, grad i Let N(i) be the coordinate gradient of point i, and let N(i) be the set of neighboring points of point i (obtained by the K-nearest neighbor algorithm, K=16), where K is the number of neighboring points, used to balance the gradient magnitude.

[0036] Furthermore, the BalancedSampler in step E2 solves the class imbalance problem by first determining the class of small samples, then dynamically adjusting the weights, and finally performing stratified sampling. The determination of the small sample class is based on a dynamically determined threshold based on the class distribution of the training set, using the following formula:

[0037] T = max(1000, min(median(C)) valid ),1.5×median(C valid ))) (7)

[0038] Among them, C valid ={c k |c k >0} represents the set of points for the effective category (sample count > 0), median() is the median function, and T is the small sample threshold (ensuring no less than 1000 points, suitable for small bridge components such as cables and connectors); dynamic weight calculation allocates sampling weights based on the number of dominant categories in the samples, with higher weights for smaller sample counts, and the formula is:

[0039]

[0040] Among them, w i c is the sampling weight for the i-th sample. dom(i) The total number of points in the dominant class of the sample (the dominant class is defined as the class that appears most frequently in the sample) is used. The weight array is normalized to ensure that the sum of probabilities is 1. The stratified sampling strategy allocates sampling quotas according to the class for the multi-class point cloud in each sample. The dominant class (non-small sample) has a maximum proportion of ≤30%, and the small sample class has a minimum proportion of ≥20%. Small sample classes are sampled with replacement to ensure that each small sample class contains at least 200 points in the training batch. Insufficient points are supplemented by repeated sampling, but the number of repetitions is limited to ≤3 times (to avoid feature redundancy).

[0041] Furthermore, the enhanced weighted loss in step E2 constructs a total loss function by fusing cross-entropy loss, IoU loss, and a dynamic weighting mechanism. The cross-entropy loss improves the prediction accuracy of smaller classes by introducing a few-sample error penalty, and its formula is:

[0042]

[0043] in, To predict the probability, y i For real labels, s(y) i ) is a small sample mask (y i (Takes 1 for small sample classes, otherwise takes 0); IoU loss can dynamically adjust class weights, focusing on optimizing low IoU classes, and the formula is:

[0044]

[0045] Among them, C present For the set of categories appearing in the current batch, IoU k Let w be the crossover ratio of the k-th class. k = 2.0 (small sample class) or 1.0 (non-small sample class), and when IoU k <0.3 w k Multiply by an additional 1.5; the total loss function is calculated as follows:

[0046] L total =α·L CE +(1-α)·L IoU (11)

[0047] Where α is the cross-entropy loss weight and 1-α is the IoU loss weight.

[0048] The advantages of this invention are:

[0049] (1) Compared with the multi-view images of bridge components taken by humans, the key frames of bridge component videos extracted by FFmpeg software can provide more feature information of bridge components, and realize high-precision three-dimensional point cloud reconstruction of bridge components.

[0050] (2) A large number of synthetic point clouds were generated by random parameter perturbation, noise simulation and virtual laser scanning, which greatly improved the diversity of training data, alleviated the problem of insufficient point cloud training data, and made the model's generalization ability significantly better than that of models trained with only real-world data.

[0051] (3) Compared with the dataset construction method that relies solely on manual annotation, the model of “manual fine annotation of main components - algorithm pre-annotation - manual verification of small sample components” ensures the accuracy of annotation while reducing the cost of small sample component annotation and solves the problems of low efficiency and difficulty in small sample annotation in traditional annotation.

[0052] (4) Compared with ordinary point cloud recognition models, the improved PointNet++ model enhances geometric learning through multi-dimensional feature fusion, and combines a balanced sampler and an enhanced weighted loss to achieve far superior accuracy in small sample component recognition compared with traditional models, with better overall mIoU index. Attached Figure Description

[0053] Figure 1 This is a flowchart of the method of the present invention;

[0054] Figure 2 This is a sample drawing of the digital model of the highway bridge in an embodiment of the present invention;

[0055] Figure 3 These are keyframes of the target components extracted using FFmpeg in part of this invention;

[0056] Figure 4 This is a three-dimensional point cloud diagram of the reconstructed bridge components according to the present invention;

[0057] Figures 5(a)-5(d) Figure 5(a) is a schematic diagram of the bridge three-dimensional point cloud data synthesized by the present invention, Figure 5(b) is a schematic diagram of the bridge deck point cloud synthesis model, Figure 5(c) is a schematic diagram of the bridge and pier combined point cloud synthesis model, and Figure 5(d) is a schematic diagram of the whole bridge point cloud synthesis model. Detailed Implementation

[0058] The following is a detailed description of the method for generating and recognizing three-dimensional point clouds of bridge components based on unmanned aerial vehicles (UAVs) according to the present invention, with reference to the accompanying drawings.

[0059] The implementation process of the bridge component identification method based on UAV point cloud reconstruction and 3D synthetic data of the present invention is as follows: Figure 1 As shown, the specific steps include:

[0060] A. A digital model of a highway bridge is selected as the instance object. The structure of the model is as follows: Figure 2 As shown, a 360° all-around surround shot was taken of the bridge pier, and the acquisition process covered information from different heights of the bridge pier from top to bottom, obtaining a 6-second on-site video of the target component.

[0061] B. Constructing a three-dimensional reconstruction framework for bridge components:

[0062] B1. Keyframes were extracted from the acquired video using FFmpeg software. A time scaling factor of 0.25 was selected, resulting in 93 keyframe images of the target component. Some of these keyframe images are shown below. Figure 3 As shown;

[0063] B2. Import the keyframe images into COLMAP software, extract image features using the SIFT algorithm, perform feature matching using KNN, Lowe's Ratio, and RANSAC algorithms, reconstruct sparse point clouds using the SfM algorithm, and then generate dense 3D point clouds using the MVS algorithm; the reconstructed point cloud is shown below. Figure 4 As shown;

[0064] C. Generate large-scale 3D point cloud composite data of the bridge:

[0065] C1. Set the parameters of bridge components such as piers and beams, randomly perturb the parameters (dimension ±5% tolerance, position offset <10cm, angle deflection <3°), generate diverse models and convert them into mesh models, use uniform sampling method to generate triangular face sampling points, and finally add Gaussian noise, block occlusion and other methods to enhance data diversity.

[0066] C2. Perform UAV-simulated laser scanning on the triangulated model, plan virtual flight routes, simulate laser emission angles and positions, calculate the time of flight (ToF) through the receiving time to reconstruct the spatial position of the point cloud, and generate a 3D point cloud of the bridge. Some point cloud data are shown in Figure 5.

[0067] C3. Repeat the above operations to obtain a large batch of diverse bridge point cloud synthetic data, and then perform dataset annotation and segmentation. Annotation is performed using a "manual-led + automatic-assisted" model: major components are manually marked using professional 3D annotation tools, while smaller sample components are first automatically pre-annotated using shape features and then manually verified.

[0068] C4. The final data consists of 845 npy files, divided into training and test sets in an 8:2 ratio.

[0069] D. Improve PointNet++ model building

[0070] D1. Point Cloud Data Preprocessing and Feature Construction. The synthesized point cloud data undergoes 3D normalization: the centroid is calculated for centering, and the data is scaled to the maximum radius within a unit sphere; a nine-dimensional input feature vector is constructed and recombined into a "9×N" format, containing normalized XYZ coordinates, normalized RGB channel values, and normal vector features. This feature vector will serve as input to the improved PointNet++ model, where multi-dimensional feature fusion will further enhance geometric relationship learning during model training.

[0071] D2. Training Set Data Augmentation. Data augmentation is performed on the training set, including master rotation, XY plane rotation, coordinate jitter, etc.; more intensive targeted augmentation is applied to small sample components to increase their feature diversity and alleviate class imbalance during model training.

[0072] D3. Model Structure Optimization and Training Strategy Adaptation. Based on the PointNet++ architecture, specific optimizations were made for bridge scenarios, including a multi-scale feature fusion network: a new local-global interaction module was added, where local branches extract neighborhood details and global branches acquire contour features, achieving bidirectional fusion through cross-attention; simultaneously, a dynamic balancing training mechanism was designed: a balanced sampler was used to increase the sampling probability of small samples (3-5 times), coupled with an enhanced weighted loss, and combined with a learning rate and gradient clipping of "linear warm-up + cosine annealing" to ensure stable model convergence and focus on difficult-to-identify regions.

[0073] E. Post-processing and application of identification results: Extract point clouds of each component by category, calculate geometric parameters such as length, volume, and axis deviation, and provide data support for bridge structural health assessment.

[0074] This embodiment verifies the method using a bridge digital model, demonstrating that it can efficiently generate and accurately identify point clouds of bridge components, meeting the engineering requirements for bridge structure inspection. The scope of protection of this invention is not limited to the above embodiment; all equivalent modifications made based on the principles of this invention should be included within the scope of protection.

Claims

1. A method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data, characterized in that, Includes the following steps: A. Use drones to collect on-site videos of bridge components; B. Establish a framework for a three-dimensional reconstruction method of bridge components, and use the video of the bridge components to reconstruct a three-dimensional point cloud model of the bridge components; B1. Extract keyframes from the video of the bridge components using FFmpeg software; B2. Process the keyframe images using COLMAP software to output a three-dimensional point cloud of the bridge components; C. Generation of large-scale 3D point cloud synthetic data for bridges; C1. Set bridge component parameters, perform random parameter perturbation to generate diverse models, and convert them into mesh models. Use uniform sampling method to generate triangular facet sampling points; finally, add noise simulation to enhance data diversity; noise simulation includes Gaussian noise, block occlusion, and missing data. C2. Perform UAV-simulated laser scanning on the triangulated model, calculate the flight time (ToF) by planning virtual flight routes, simulating laser emission angle, position, and reception time, and reconstruct the spatial position of the point cloud to generate a 3D point cloud of the bridge. C3. Repeat the above operations to obtain a large amount of diverse bridge point cloud composite data; D. Construct a semantic segmentation dataset of bridge component point clouds; D1. Point cloud data preprocessing: The bridge point cloud synthetic data is subjected to three-dimensional normalization processing, the centroid of the point cloud is calculated to achieve centering, and then the point cloud is scaled to a unit sphere according to the maximum radius to ensure the geometric consistency of the data; a nine-dimensional input feature vector is constructed, which includes the normalized XYZ coordinates, normalized RGB channel values ​​and normal vector features, and the feature matrix is ​​reorganized into the "9×N" format required by the model, where N is the number of points; D2. Data Labeling and Classification: For the main components of the bridge, professional 3D labeling tools are used for manual and detailed labeling to clarify their category classification; for a small number of components, their inherent shape features are first used for automatic pre-labeling by algorithms, and then they are strictly verified and corrected manually to ensure the accuracy of the labeling results to the greatest extent; the point cloud dataset is divided into training set and test set according to a preset ratio, and the basic principle of keeping the proportion of points of each component category in the training set and test set consistent is followed. D3. Data Augmentation: Various data augmentation operations are performed on the training set to expand the sample diversity. Basic augmentation methods include simulating master rotation from different observation angles, simulating XY plane rotation with horizontal changes, introducing coordinate jitter with slight positional perturbations, random rotation with all-around viewpoint changes, scaling with distance changes, and local translation with local positional shifts. For small sample component categories with few points, additional and more targeted augmentations are performed to significantly increase the diversity of their features. E. Build an improved PointNet++ point cloud recognition model for bridge components; E1. Network Structure Design and Optimization: Based on the PointNet++ architecture, the model is built by integrating the classification, segmentation and utils modules. The geometric relationship learning ability is enhanced by multi-dimensional feature fusion, which improves the model's adaptability to bridge scenarios. E2. Training Strategy and Execution Process: During training, a balanced sampler is used to dynamically adjust sample weights, and an enhanced weighted loss function is used to impose additional penalties on small sample misprediction and low IoU categories. Set an initial learning rate, adopt a scheduling method that combines linear warm-up and cosine annealing, and use gradient pruning strategy to ensure training stability; Using the test set mIoU as the core indicator, an early stopping mechanism is triggered when there is no improvement for several consecutive rounds, and finally the category prediction result of the optimal model is output to complete the point cloud recognition of bridge components. F. Post-processing and application of identification results: Calculate geometric parameters to provide data support for the assessment of the health status of bridge structures.

2. The bridge component identification method based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The method for acquiring bridge component video in step A is as follows: using a drone to surround the target component in a 360° omnidirectional manner to acquire video of the target component. The video acquisition requirements include information of the component at different heights from top to bottom, and finally, the bridge component video is collected.

3. The bridge component identification method based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: In step B1, FFmpeg will be used to extract key frame images of the bridge components from the video. By selecting 0.25 as the time scaling factor, it will be ensured that the output key frames contain as much feature information of the bridge components as possible.

4. The method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: In step B2, the output of the three-dimensional point cloud of the bridge component is achieved using COLMAP software. COLMAP software uses the SIFT algorithm to extract image features, KNN, LoweRatio, and RANSAC algorithms to perform feature matching, and combines the SfM algorithm to reconstruct the sparse point cloud of the target component. Finally, the MVS algorithm is used to reconstruct the dense point cloud.

5. The bridge component identification method based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The random perturbation of parameters in step C1 to generate diverse models includes: extracting the basic geometric parameters of bridge components and applying random perturbations to the parameters to achieve model diversification; adjusting the degree of morphological variation of the generated model by controlling the perturbation intensity; and configuring a reasonable perturbation threshold range so that the generated component model can cover common deviation morphologies in actual engineering while maintaining structural rationality, in order to output training-level point cloud data with strong generalization ability.

6. The method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The data augmentation operations in step D3 include basic augmentation and small-sample targeted augmentation. The basic augmentation is performed on all samples, including rotation, scaling, and Gaussian noise operations; Small sample-specific augmentation doubles the augmentation intensity for small sample categories, including: in, To enhance the post-3D coordinates, Original 3D coordinates For local deformation masking, The scaling factor is the main axis, and Δ is the translation amount, which enhances the morphological diversity of small samples.

7. The bridge component identification method based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The multi-dimensional feature fusion mechanism in step E1 includes local coordinate normalization, RGB feature standardization, and coordinate gradient feature extraction. Local coordinate normalization enhances the expression of local geometric features through decentralization and scale scaling, and can be expressed as: in, Original 3D coordinates Here are the coordinates of the center point of the point cloud. The distance from the farthest point in the point cloud to the center point is used for scale normalization. This maps the normalized local coordinates to the unit sphere; RGB feature normalization normalizes color features, expressed as: in, The original RGB values. , These represent the minimum and maximum values ​​of the RGB channels in the point cloud, respectively. The standardized RGB features; coordinate gradient feature extraction calculates the local continuity features of the point cloud through neighborhood difference, simulating normal vector information, and can be expressed as: in, Let i be the gradient of the coordinates of point i. Let K be the set of neighborhood points of point i, and K be the number of neighborhood points, used to balance the gradient magnitude.

8. The method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The BalancedSampler in step E2 solves the class imbalance problem by first determining the class of small samples, then dynamically adjusting the weights, and finally performing stratified sampling. The determination of the small sample class is based on a dynamically determined threshold based on the distribution of the training set classes, using the following formula: in, The valid category is the set of points where the number of samples is greater than 0. The median function is used, where T is the small sample threshold. Dynamic weighting is calculated by allocating sampling weights based on the number of samples belonging to the dominant class; the fewer the samples, the higher the weight. The formula is as follows: in, Let i be the sampling weight of the i-th sample. The total number of points in the dominant class of the sample is defined as the class that appears most frequently in the sample. The weight array is normalized to ensure that the sum of probabilities is 1. The stratified sampling strategy allocates sampling quotas according to the class for the multi-class point cloud in each sample. The maximum proportion of non-small sample classes is ≤30%, and the minimum proportion of small sample classes is ≥20%. Small sample classes use sampling with replacement to ensure that each small sample class contains at least 200 points in the training batch.

9. A method for identifying bridge components based on UAV point cloud reconstruction and 3D synthetic data according to claim 1, characterized in that: The enhanced weighted loss in step E2 constructs a total loss function by fusing cross-entropy loss, IoU loss, and a dynamic weighting mechanism. The cross-entropy loss improves the accuracy of small-class predictions by introducing a few-sample error penalty, and its formula is as follows: in, To predict probabilities, For real labels, For small sample masks, Set the value to 1 if it represents a small sample category, otherwise set it to 0; The loss function can dynamically adjust class weights, focusing on low-level categories. Category optimization, the formula is: in, The set of categories appearing in the current batch. The intersection-union ratio (CUNR) of the k-th class is given by the small sample class. For non-small sample classes , and when hour Multiply by an additional 1.5; the total loss function is calculated as follows: in, For cross-entropy loss weights, for Loss weights.

Images