Method for detecting rusting target of grid structure

Abstract

The invention relates to a grid structure corrosion target detection method. The method comprises the steps of obtaining high-definition image data and point cloud data of the surface of a to-be-detected grid rod piece and thickness data in the grid rod piece; extracting channel features in the high-definition image data for corrosion analysis to obtain a first corrosion detection result; point cloud features are extracted based on the preprocessed point cloud data, and a second corrosion detection result is recognized; comparing the difference between the preprocessed thickness data and the normal thickness data to obtain a third corrosion detection result; fusing the channel features, the point cloud features and the conditional probability distribution of the thickness data based on a Bayesian reasoning multi-modal data fusion strategy to obtain the corrosion probability of each detection area, further determining a corrosion area, and integrating three corrosion detection results of the corrosion area to obtain a corrosion comprehensive evaluation report of the corrosion area; and combining the corrosion comprehensive evaluation report and future corrosion risk prediction based on LSTM model analysis to obtain a final corrosion detection report.

Description

Technical Field

[0001] This application relates to the technical field of rust detection for grid structures, and particularly to a method for detecting rust targets in grid structures. Background Art

[0002] Currently, the existing methods for detecting rust on grid members have the following problems: 1. Traditional rust detection for grid members mainly relies on single data sources such as image data or ultrasonic data, resulting in the inability to comprehensively reflect the complex characteristics of rust (such as the combined evaluation of surface rust and internal corrosion), and being unable to efficiently and accurately identify the entire picture of rust, especially rust in deep or hidden areas.

[0003] 2. Existing image processing methods often perform poorly in complex backgrounds and are easily affected by factors such as lighting, shadows, and stains, leading to inaccurate identification of rust areas. At the same time, traditional algorithms have low accuracy in feature extraction and classification, making it difficult to achieve efficient detection of subtle rust.

[0004] 3. Current rust detection technologies mainly focus on single-modal or single-feature analysis for rust evaluation, and are unable to effectively combine different sensor data (such as images, point clouds, ultrasonic thickness data, etc.) to comprehensively evaluate the morphology, depth of rust and its impact on structural safety.

[0005] 4. Existing detection methods mainly focus on static detection results, lacking dynamic prediction and evaluation of the development trend of rust, being unable to timely discover potential corrosion risks, and also being difficult to formulate effective maintenance and repair strategies based on the development trend of rust.

[0006] 5. Traditional rust detection methods are easily affected by external factors (such as environment, equipment performance, operators, etc.), resulting in low accuracy and reliability of detection results, and thus affecting structural safety assessment and decision-making. Summary of the Invention

[0007] Based on this, it is necessary to provide a method for detecting rust targets in grid structures, which includes: S1: Obtain high-definition image data, point cloud data on the surface of the grid members to be measured, and thickness data inside the grid members, and preprocess the point cloud data and the thickness data respectively; S2: Use a convolutional neural network model based on a deep residual network to extract texture features, color features, and edge features from the high-definition image data, and perform rust analysis based on these three channel features to obtain a first rust detection result including the rust position, shape, and area; The plane fitting algorithm using random sample consensus is used to process the pre - processed point cloud data, extract the point cloud features on the surface of the rod, analyze the local curvature change based on the point cloud features, and identify the second rust detection result including the surface depressions and pits of the rod; By comparing the difference between the pre - processed thickness data and the normal thickness data, the third rust detection result including the degree of internal rust of the rod is obtained; S3: Adopt a multi - modal data fusion strategy based on Bayesian inference to fuse the channel features, the point cloud features and the conditional probability distribution of the thickness data, obtain the rust probability of each detection area, determine the rust area based on the rust probability, and comprehensively obtain the comprehensive rust assessment report of the rust area by combining the first rust detection result, the second rust detection result and the third rust detection result of the rust area; S4: Use the LSTM model to perform time - series analysis on the channel features, point cloud features and thickness data, predict the future rust development trend, and then conduct rust risk assessment to obtain the future rust risk prediction; Combine the comprehensive rust assessment report and the future rust risk prediction to obtain the final rust detection report.

[0008] Beneficial effects: This method obtains the high - definition image data, point cloud data on the surface of the grid frame rod to be measured, and the thickness data inside the grid frame rod; extracts the channel features in the high - definition image data for rust analysis to obtain the first rust detection result; extracts the point cloud features based on the pre - processed point cloud data to identify the second rust detection result; compares the difference between the pre - processed thickness data and the normal thickness data to obtain the third rust detection result; fuses the conditional probability distributions of the channel features, point cloud features and thickness data using a multi - modal data fusion strategy based on Bayesian inference to obtain the rust probability of each detection area, and then determines the rust area. By comprehensively combining the three rust detection results of the rust area, a comprehensive rust assessment report of the rust area is obtained; Combine the comprehensive rust assessment report and the future rust risk prediction analyzed based on the LSTM model to obtain the final rust detection report. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0010] Figure 1 It is a flowchart of the rust target detection method for the grid frame structure in the embodiment of the present application. Detailed Embodiments

[0011] To make the above objects, features, and advantages of the present application more apparent and understandable, the following provides a detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings. Many specific details are set forth in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present application. Therefore, the present application is not limited by the specific embodiments disclosed below.

[0012] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present application, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically and clearly defined.

[0013] As Figure 1 shown, this embodiment provides a method for detecting the rust target of a grid structure, and the method includes: S1: Obtain the high-definition image data, point cloud data on the surface of the grid members to be measured, and the thickness data inside the grid members, and preprocess the point cloud data and the thickness data respectively.

[0014] In this embodiment, the high-definition image data is collected by a high-definition camera, the point cloud data is collected by a three-dimensional laser scanner, and the thickness data is collected by an ultrasonic thickness gauge.

[0015] In this embodiment, the preprocessing of the point cloud data and the thickness data respectively includes: Use statistical filtering or outlier removal algorithms (such as the RANSAC algorithm) to perform noise reduction processing on the point cloud data, including removing isolated points, redundant points, or error points. This processing can remove irrelevant or incorrect points and improve the overall quality of the point cloud data; Use signal filtering techniques (such as Kalman filtering, mean filtering) to perform filtering and signal processing on the thickness data to improve the signal-to-noise ratio of the data, make the processed data cleaner, facilitate subsequent feature extraction and distribution, and improve the accuracy of rust detection.

[0016] S2: Use a convolutional neural network model based on a deep residual network to extract texture features, color features, and edge features from the high-definition image data, and perform rust analysis based on these three channel features to obtain a first rust detection result including the rust position, shape, and area.

[0017] Further, the process of extracting texture features, color features, and edge features from the high-definition image data includes: Step 1: Denoise the high-definition image data using a Gaussian filter; Normalize the RGB values of each pixel in the denoised high-definition image data; Perform data augmentation on the high-definition image data; Step 2: Use ResNet as the backbone network to process the high-definition image data processed in Step 1; ResNet is formed by stacking multiple residual blocks, and each residual block includes two or three convolutional operations; Define a small-sized convolutional kernel in ResNet and slide the convolutional kernel on the high-definition image data, Capture the texture features in the high-definition image data through the local receptive field operation of the convolutional kernel; Extract the color features of the RGB channels in the high-definition image data through the multi-channel operation of the convolutional kernel; Enhance the boundary information of the rust area in the high-definition image data by embedding the Sobel operator into the convolutional kernel to obtain edge features.

[0018] The convolutional neural network model can effectively identify the rust area, separate it from the background area, and output information such as the specific location, area, and shape of the rust area, providing accurate image feature input for subsequent multi-modal data fusion.

[0019] Furthermore, the process of rust analysis based on texture features, color features, and edge features includes: Step 1: For the feature map output by ResNet, denoted as: , where X represents the feature map, H represents the height of the feature map, W represents the width of the feature map, and 3 is the number of channels; the feature map includes texture features, color features, and edge features; Step 2: Perform global averaging on the 3 channels to obtain the global description vector of the channels, and the calculation formula is: ; where, represents the global description vector of the c-th channel; represents the value of the c-th channel at the (i, j) position in the feature map; Step 3: Input the global description vector into the fully connected layer to obtain the channel weights, and the calculation formula is: ; where, represents the channel weight of the c-th channel; represents the weight matrix of the first fully connected layer; represents the weight matrix of the second fully connected layer; represents the ReLU activation function; represents the Sigmoid activation function; Step 4: Apply the weights of each channel to the corresponding channel to obtain a channel-weighted feature map; Step 5: Input the channel-weighted feature map into the classification network to obtain a first rust detection result including the rust location, shape, and area.

[0020] Use the plane fitting algorithm of random sample consensus to process the preprocessed point cloud data, extract the point cloud features on the surface of the rod, analyze the local curvature change based on the point cloud features, and identify a second rust detection result including the surface depression and pit of the rod.

[0021] Furthermore, the process of extracting the point cloud features on the surface of the rod includes: Step 1: Set parameters, including the maximum number of iterations, distance threshold, and inlier ratio threshold; Step 2: Randomly select 3 points from the preprocessed point cloud data in each iteration , and determine a fitting plane according to the 3 points. The equation of the fitting plane is: ; ; ; where a, b, and c are the three components of the normal vector n of the fitting plane respectively; d represents the offset, which is the distance from the fitting plane to the origin and determines the position of the fitting plane in three-dimensional space; x, y, and z are the abscissa, ordinate, and vertical coordinate of any point in the point cloud data respectively; Step 3: Calculate the distance from any point in the point cloud data to the fitting plane. The calculation formula is: ; where represents the distance from the i-th point to the fitting plane; , , represent the abscissa, ordinate, and vertical coordinate of the i-th point respectively; Step 4: Compare the distances from all points in the point cloud data to the fitting plane with the distance threshold , and regard the points with distances less than the distance threshold as inliers, and count the number of inliers; Step 5: If the number of inliers in the current iteration is greater than that in the previous iteration, update the three components a, b, and c and d in the fitting plane equation; Step 6: Repeat steps 2 - 5 until the maximum number of iterations is reached or the inlier ratio exceeds the inlier ratio threshold, to obtain the optimal plane (a, b, c, d) and the set of inliers belonging to this optimal plane; Step 7: Project the point cloud data onto the optimal plane, and use the normal vector of the optimal plane as the plane feature; Step 8: Based on the optimal plane, calculate the covariance matrix of each point in the point cloud data. The calculation formula is: ; where C represents the covariance matrix; k represents the number of neighborhood points; represents the centroid of the neighborhood points; T represents the transpose; represents the i-th neighborhood point; For any point, solve its covariance matrix to obtain three eigenvalues , , , and < < ; Based on the three eigenvalues, calculate the curvature of the corresponding point. The calculation formula is: ; where K represents the curvature; the plane feature and the curvature constitute the point cloud feature.

[0022] Furthermore, analyzing the local curvature change based on the point cloud feature includes: Regions with high curvature correspond to depressions or pits on the surface of the rod; Identify the rust area, depth, and distribution pattern on the surface of the rod through the distribution of curvature.

[0023] In this embodiment, the grid structure is exposed to the environment for a long time, and there may be rust, dirt, mechanical damage, etc. on the surface, resulting in a large number of outliers and noise points in the collected three-dimensional point cloud data. Therefore, it is necessary to perform noise reduction and plane feature extraction on the point cloud data through the random sample consensus plane fitting algorithm, so as to identify clean surface features and provide a reliable data basis for subsequent rust detection.

[0024] Compare the difference between the preprocessed thickness data and the normal thickness data to obtain the third rust detection result including the degree of internal rust of the rod.

[0025] S3: Adopt a multi-modal data fusion strategy based on Bayesian inference to fuse the conditional probability distributions of the channel features, the point cloud features, and the thickness data, obtain the rust probabilities of each detection area, determine the rust areas based on the rust probabilities, and comprehensively obtain a rust comprehensive evaluation report for the rust areas based on the first rust detection result, the second rust detection result, and the third rust detection result of the rust areas.

[0026] In this embodiment, the fusion formula for the conditional probability distribution is: ; ; ; ; Among them, represents the rust probability of the corresponding detection area; F represents any one of the channel features, the point cloud features, and the thickness data, and H represents a hypothesis variable; represents the weight of the channel feature; represents the weight of the point cloud feature; represents the weight of the thickness data; represents the conditional probability distribution of the channel feature; represents the conditional probability distribution of the point cloud feature; represents the conditional probability distribution of the thickness data; represents summation; represents that the data follows a normal distribution and is used to describe the probability density of the data; represents the mean difference of the channel feature and is used to measure the central value of the channel feature in this area; represents the mean difference of the point cloud feature, which is the central value of the point cloud feature in this area; represents the mean difference of the thickness data, which reflects the average thickness of the rod in this area; represents the variance of the channel feature and is used to describe the fluctuation degree of the high-definition image data; represents the variance of the point cloud feature and is used to reflect the discreteness of the point cloud data; represents the variance of the thickness data and is used to represent the fluctuation of the thickness measurement value.

[0027] Furthermore, the determination of the rust area based on the rust probability includes: When the rust probability of the detection area is greater than the set threshold, then mark the corresponding detection area as a rust area; otherwise, do not mark.

[0028] In this embodiment, the rust comprehensive evaluation report further includes a distribution map of the rust probabilities of each detection area and the marked rust areas.

[0029] Bayesian inference can establish the correlation between different modalities of data by combining the conditional probability distributions of different modalities of data, thereby fusing the detection results of different sensors into a comprehensive rust evaluation model. Through the multi-modal data fusion strategy of Bayesian inference, the method provided in this embodiment can more comprehensively evaluate the severity and distribution of rust, and output a comprehensive rust evaluation report.

[0030] S4: Use the LSTM model to perform time series analysis on the channel features, point cloud features, and thickness data, predict the future rust development trend, and then conduct rust risk assessment to obtain the future rust risk prediction; combine the comprehensive rust evaluation report and the future rust risk prediction to obtain the final rust detection report.

[0031] Specifically, this step includes: Fuse the channel features, point cloud features, and thickness data at each time point, and represent them as a time series vector, denoted as: ; ; Among them, L represents the time series vector; Represents the features fused from the channel features, point cloud features, and thickness data at the t-th time point; Represents the number of historical time points; Represents the channel features at the t-th time point; Represents the point cloud features at the t-th time point; Represents the thickness data at the t-th time point; Normalize the values in the time series vector to the interval [0, 1]; Input the normalized time series vector into the LSTM model, and output the future rust development trend, where the future rust development trend includes the growth trend of rust area, the distribution evolution of high curvature regions, and the intensification of rust depth.

[0032] In this embodiment, the LSTM model includes an input gate, a forget gate, and an output gate; The operating formula of the forget gate (which determines how much historical information to retain) is: ; Among them, Represents the hidden state at the (t - 1)-th time point; Represents the output value of the forget gate at the t-th time point, with a value range of [0, 1], and if , it means that most of the historical information is retained, and if , it means that most of the historical information is forgotten; Represents the sigmoid activation function; The weight matrix representing the forget gate, which is used to adjust the influence of the hidden state and the current input on the output of the forget gate; Represents the bias term of the forget gate; The operating formula for the input gate (determining how much new information to add) is: ; ; where, Represents the hyperbolic tangent function; Represents the output value of the input gate at the t-th time point, with a value range of [0, 1], and if , it means that the information at the current time step will be added to the memory, and if , it means that no new information will be introduced; Represents the weight matrix of the input gate, which is used to adjust the influence of the hidden state and the input on the new information; Represents the bias term of the input gate; Represents the candidate cell state, that is, the new memory content at this time step; Represents the weight matrix of the candidate cell state, which is used to calculate the weight of the new information; Represents the bias term of the candidate cell state; The state update formula is: ; where, Represents the cell state at the t-th time point, Represents the cell state at the (t - 1)-th time point; The operating formula for the output gate (generating the hidden state) is: ; ; where, Represents the hidden state at the t-th time point; Represents the output value of the output gate at the t-th time point, which is used to control the final output of the LSTM at the current time step; Represents the weight matrix of the output gate, which is used to adjust the influence of the hidden state and the input on the output gate; Represents the bias term of the output gate.

[0033] The mean squared error is used as the loss function to measure the error between the predicted value and the true value. The calculation formula of the loss function is: ; where, N represents the number of time points; Denote the predicted value at the $i$-th time point; Denote the true value at the $i$-th time point; Divide the historical corrosion data into a training set (e.g., the first 80% of the time points) and a validation set (the last 20% of the time points). The data format is: Input: A time series of length $T$ ; Output: Features predicting the next $N$ time points ; Use the Adam optimizer to adjust the model weights, with the initial learning rate set to 10 -3 .

[0034] Calculate the error between the predicted value and the true value using the validation set to evaluate the prediction accuracy of the model.

[0035] Visualize the future corrosion development trend as a trend chart. When the corrosion area trend in the trend chart increases rapidly, it is considered that there is a future corrosion risk in the corresponding area; otherwise, it is considered that there is no future corrosion risk; Combine the comprehensive corrosion assessment report and the future corrosion risk prediction to obtain the final corrosion detection report.

[0036] The method for detecting corrosion targets of the grid structure provided in this embodiment obtains high-definition image data, point cloud data on the surface of the grid members to be measured, and thickness data inside the grid members; extracts channel features from the high-definition image data for corrosion analysis to obtain the first corrosion detection result; extracts point cloud features based on the preprocessed point cloud data to identify the second corrosion detection result; compares the difference between the preprocessed thickness data and the normal thickness data to obtain the third corrosion detection result; fuses the conditional probability distributions of channel features, point cloud features, and thickness data based on the multi-modal data fusion strategy of Bayesian inference to obtain the corrosion probability of each detection area, and then determines the corrosion area. Combine the three corrosion detection results of the corrosion area to obtain a comprehensive corrosion assessment report of the corrosion area; combine the comprehensive corrosion assessment report and the future corrosion risk prediction analyzed based on the LSTM model to obtain the final corrosion detection report. This method solves the limitations of single data source detection, solves the problem of lack of efficient feature extraction and classification methods, solves the problem of inability to comprehensively evaluate the impact of different types of corrosion, also solves the problem of lack of corrosion development trend prediction ability, and at the same time solves the problems of accuracy and reliability, providing an efficient and accurate feasible solution for the detection of corrosion targets of the grid structure.

[0037] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0038] The embodiments described above merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the scope of the patent application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A method for detecting the corrosion target of a grid structure, characterized in that Including: S1: Obtain the high-definition image data, point cloud data of the surface of the grid members to be measured, and the thickness data inside the grid members, and preprocess the point cloud data and the thickness data respectively; S2: Use a convolutional neural network model based on a deep residual network to extract texture features, color features, and edge features from the high-definition image data, and perform rust analysis based on these three channel features to obtain a first rust detection result including the rust position, shape, and area; Use a plane fitting algorithm of random sample consensus to process the preprocessed point cloud data, extract the point cloud features of the member surface, analyze the local curvature change based on the point cloud features, and identify a second rust detection result including surface depressions and holes of the member; Compare the difference between the preprocessed thickness data and the normal thickness data to obtain a third rust detection result including the degree of internal rust of the member; S3: Use a multi-modal data fusion strategy based on Bayesian inference to fuse the conditional probability distributions of the channel features, the point cloud features, and the thickness data to obtain the rust probability of each detection area, determine the rust area based on the rust probability, and comprehensively obtain a rust comprehensive evaluation report for the rust area based on the first rust detection result, the second rust detection result, and the third rust detection result of the rust area; S4: Use an LSTM model to perform time series analysis on the channel features, point cloud features, and thickness data, predict the future rust development trend, and then perform rust risk assessment to obtain a future rust risk prediction; Combine the rust comprehensive evaluation report and the future rust risk prediction to obtain a final rust detection report.

2. The method for detecting the corrosion target of the grid structure according to claim 1, characterized in that The preprocessing of the point cloud data and the thickness data respectively includes: Perform noise reduction processing on the point cloud data; Perform filtering and signal processing on the thickness data.

3. The method for detecting the corrosion target of the grid structure according to claim 1, characterized in that, The process of extracting texture features, color features, and edge features from high-definition image data includes: Step 1: Use a Gaussian filter to perform denoising processing on the high-definition image data; Normalize the RGB values of each pixel in the denoised high-definition image data; Perform data augmentation on the high-definition image data; Step 2: Use ResNet as the backbone network to process the high-definition image data processed in Step 1; ResNet is formed by stacking multiple residual blocks, and each residual block includes two or three convolutional operations; Define a small-sized convolutional kernel in ResNet and slide the convolutional kernel on the high-definition image data, Capture the texture features in the high-definition image data through the local receptive field operation of the convolutional kernel; Extract the color features of the RGB channels in the high-definition image data through the multi-channel operation of the convolutional kernel; Embed the Sobel operator into the convolutional kernel to enhance the boundary information of the rust area in the high-definition image data to obtain edge features.

4. The method for detecting the corrosion target of the grid structure according to claim 1, characterized in that, The process of performing rust analysis based on texture features, color features, and edge features includes: Step 1: For the feature map output by ResNet, denoted as: , where X represents the feature map, H represents the height of the feature map, W represents the width of the feature map, and 3 is the number of channels; the feature map includes texture features, color features, and edge features; Step 2: Perform global averaging on the 3 channels to obtain a global description vector of the channels, and the calculation formula is: ； Among them, represents the global description vector of the c-th channel; represents the value of the c-th channel at the (i, j) position in the feature map; Step 3: Input the global description vector into a fully connected layer to obtain channel weights, and the calculation formula is: ； Among them, represents the channel weight of the c-th channel; represents the weight matrix of the first fully connected layer; represents the weight matrix of the second fully connected layer; represents the ReLU activation function; represents the Sigmoid activation function; Step 4: Apply the channel weights to the corresponding channels to obtain the channel-weighted feature map; Step 5: Input the channel-weighted feature map into the classification network to obtain the first rust detection result including the rust location, shape, and area.

5. The method for detecting the corrosion target of the grid structure according to claim 1, wherein The process of extracting the point cloud features on the surface of the member includes: Step 1: Set parameters, including the maximum number of iterations, distance threshold, and inlier ratio threshold; Step 2: Randomly select 3 points from the preprocessed point cloud data in each iteration , and determine a fitting plane based on the 3 points. The equation of the fitting plane is as follows: ； ； ； where a, b, and c are the three components of the normal vector n of the fitted plane; d represents the offset; x, y, and z are the abscissa, ordinate, and vertical coordinate of any point in the point cloud data, respectively; Step 3: Calculate the distance from any point in the point cloud data to the fitted plane, and the calculation formula is: ； wherein, represents the distance from the i-th point to the fitting plane; , , respectively represent the abscissa, ordinate, and vertical coordinate of the i-th point; Step 4: Compare the distances from all points in the point cloud data to the fitted plane with the distance threshold, take the points corresponding to the distances less than the distance threshold as inliers, and count the number of inliers; Step 5: If the number of inliers in the current iteration is greater than the number of inliers in the previous iteration, update the three components a, b, and c and d in the fitted plane equation; Step 6: Repeat Steps 2 - 5 until the maximum number of iterations is reached or the inlier ratio exceeds the inlier ratio threshold to obtain the optimal plane (a, b, c, d) and the set of inliers belonging to the optimal plane; Step 7: Project the point cloud data onto the optimal plane, and use the normal vector of the optimal plane as the plane feature; Step 8: Based on the optimal plane, calculate the covariance matrix of each point in the point cloud data, and the calculation formula is: ； Among them, C represents the covariance matrix; k represents the number of neighborhood points; represents the centroid of the neighborhood points; T represents the transpose; represents the i-th neighborhood point; For any point, solve its covariance matrix to obtain three eigenvalues respectively , , , and < < ; Calculate the curvature of the corresponding point based on the three eigenvalues, and the calculation formula is: ； where K represents the curvature; the plane feature and the curvature constitute the point cloud feature.

6. The method for detecting the corrosion target of the grid structure according to claim 5, characterized in that, Analyzing the local curvature change based on the point cloud feature includes: Regions with high curvature correspond to depressions or pits on the surface of the member; Identify the rust area, depth, and distribution pattern on the surface of the member through the distribution of curvature.

7. The method for detecting the rust target of the grid structure according to claim 1, wherein The fusion formula of the conditional probability distribution is: ； ； ； ； Among them, represents the rust probability of the corresponding detection area; F represents any one of the channel feature, the point cloud feature, and the thickness data, and H represents a hypothesis variable; represents the weight of the channel feature; represents the weight of the point cloud feature; represents the weight of the thickness data; represents the conditional probability distribution of the channel feature; represents the conditional probability distribution of the point cloud feature; represents the conditional probability distribution of the thickness data; represents summation; indicates that the data follows a normal distribution; represents the mean difference of the channel feature; represents the mean difference of the point cloud feature; represents the mean difference of the thickness data; represents the variance of the channel feature; represents the variance of the point cloud feature; represents the variance of the thickness data.

8. The method for detecting the corrosion target of the grid structure according to claim 1, characterized in that, The determining the rust area based on the rust probability includes: When the rust probability of the detection area is greater than the set threshold, mark the corresponding detection area as the rust area; Otherwise, do not mark.

9. The method for detecting the corrosion target of the grid structure according to claim 1, characterized in that, The rust comprehensive evaluation report also includes the distribution map of the rust probability of each detection area and the marked rust areas.

10. The method for detecting the corrosion target of the grid structure according to claim 1, wherein, S4 includes: Fuse the channel feature, point cloud feature, and thickness data at each time point and represent them as a time series vector, denoted as: ； ； Among them, L represents the timing vector; represents the feature fused by the channel feature, point cloud feature, and thickness data at the t-th time point; represents the number of historical time points; represents the channel feature at the t-th time point; represents the point cloud feature at the t-th time point; represents the thickness data at the t-th time point; Normalize the values in the time series vector to the interval [0, 1]; Input the normalized time series vector into the LSTM model to output the future rust development trend, and the future rust development trend includes the rust area growth trend, the distribution evolution of high curvature regions, and the aggravation of rust depth; Visualize the future rust development trend as a trend graph. When the rust area trend in the trend graph increases rapidly, it is considered that there is a future rust risk in the corresponding area, otherwise it is considered that there is no future rust risk; Combine the rust comprehensive evaluation report and the future rust risk prediction to obtain the final rust detection report.

Images