Construction and application method of global descriptor for detecting tunnel segment leakage

Abstract

The invention discloses a construction and application method of a global descriptor for detecting tunnel segment leakage, and relates to the technical field of subway shield tunnel leakage segment and non-leakage segment detection, and the method comprises the following steps: S1, obtaining subway tunnel point cloud data by using a mobile laser scanning system; s2, positioning the circular seam according to the spatial position relationship between the bolt hole and the circular seam, so as to extract a single segment; s3, generating a global strength descriptor for a single segment; and S4, fusing the generated global intensity descriptor and the random forest classifier to detect leakage segments and non-leakage segments. The method has the beneficial effects that the overall strength descriptor is constructed, the strength distribution of the single segment is effectively represented, the leakage segment and the non-leakage segment are accurately positioned by utilizing the strength distribution difference of the leakage segment and the non-leakage segment, and important technical support and guarantee are provided for safe operation and maintenance of the subway tunnel.

Description

Technical Field

[0001] The invention relates to the technical field of subway shield tunnel leakage and non-leakage segment detection, and in particular to a construction and application method of a global descriptor for detecting tunnel segment leakage. Background Art

[0002] Tunnels are an important part of urban subway transportation, and their disease detection and maintenance are crucial to ensuring subway safety. During the long-term operation of tunnels, leakage often occurs in tunnels due to construction errors, changes in geological conditions, environmental impacts and other factors. This not only poses a threat to the durability of the tunnel structure, but may also pose a major hidden danger to the safety of subway operations. With the development of technology, lidar can provide high-precision three-dimensional point cloud data and has been widely used in tunnel leakage detection. However, existing leakage detection methods rely on single-dimensional reflection intensity characteristics, which makes it difficult to accurately locate the segment where the leakage is located. Summary of the invention

[0003] The purpose of the present invention is to overcome the problems in the background technology and provide a method for constructing and applying a global descriptor for detecting tunnel segment leakage. The core of the method is to characterize the strength distribution of a single segment and detect leaking segments and non-leaking segments by utilizing the difference in strength distribution of different segments.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically: a method for constructing and applying a global descriptor for detecting tunnel segment leakage, comprising the following steps:

[0005] S1, the mobile laser scanning system scans the subway tunnel and obtains the subway tunnel point cloud data;

[0006] S2, locate the annular seam according to the spatial position relationship between the bolt hole and the annular seam, and then extract a single segment;

[0007] S3, generating a global intensity descriptor for a single segment;

[0008] S4, Fusion of global intensity descriptor and random forest classifier to detect leaking and non-leaking segments.

[0009] Further, in step S2, the method of locating the annular seam according to the spatial position relationship between the bolt hole and the annular seam and then extracting a single segment includes the following steps:

[0010] S2.1, the collected tunnel point cloud is P(x i ,y i ,z i ) i＝1,2,…,n , project the point cloud to the XOZ plane and obtain a two-dimensional point set (x i ,z i ) i＝1,2,…n; Perform elliptical fitting on the two-dimensional point set to obtain the center coordinates (x 0 , z 0 ); Calculate the distance from each point in the two-dimensional point set to the center of the circle, set a distance threshold δ, if the distance from a certain point to the center of the circle is greater than the distance threshold, then determine that point as a bolt hole point and extract it;

[0011] S2.2, Use the DBSCAN algorithm to cluster the mileage of the bolt holes;

[0012] S2.3, After clustering the bolt holes using DBSCAN, each segment will form two clusters, which respectively represent two groups of bolt holes with similar mileage values. Subsequently, take the average value of the y coordinate values of the corresponding clusters in adjacent segments to determine the position loc of the circumferential joint v ;

[0013] S2.4, According to the circumferential joint position loc v , extract a single segment according to the following formula:

[0014]

[0015] where v is the index of the circumferential joint.

[0016] Furthermore, in step S2.2, the clustering of the mileage of the bolt holes using the DBSCAN algorithm includes the following steps:

[0017] S2.21, Set two parameters, the neighborhood radius ε and the density threshold MinPts. Randomly select an unvisited point, calculate the number of its neighborhood points within the given radius Eps. If the number of neighborhood points exceeds the preset density threshold minPts, then mark this point as a core point. Otherwise, temporarily mark it as a border point. At the same time, regard the points outside the neighborhood as noise points;

[0018] S2.22, For the core points, traverse all the points in their neighborhoods and group these points into the same cluster. If there are other core points in the cluster, then include these core points and their neighborhood points into the current cluster;

[0019] S2.23, Repeat steps S2.21 and S2.22 until all points have been visited.

[0020] Furthermore, in step S2.2, the mileage of the bolt is the y coordinate value.

[0021] Furthermore, in step S3, the generation of the global intensity descriptor for a single segment includes the following steps:

[0022] S3.1, Normalize the reflection intensity of a single segment according to the following formula:

[0023]

[0024] Among them,

[0025] I i is the reflection intensity value of each point,

[0026] I i ' is the normalized reflection intensity value of each point,

[0027] N is the total number of points of a single segment;

[0028] S3.2, the range of the normalized reflection intensity value is [0, 1], and the intensity interval range is [b k , b k+1 ), k = 1, 2,... K. According to the intensity interval width b k+1 -b k , [0, 1] is divided into K intensity intervals;

[0029] S3.3, calculate the number of points n K falling into each intensity interval through the indicator function, as shown in the following formula:

[0030]

[0031] Among them,

[0032] δ(I i ' ∈ [b k , b k+1 )) represents the indicator function. When a certain point falls into the intensity interval [b k , b k+1 )), the value of this function is 1; otherwise, the value of this function is 0;

[0033] S3.4, calculate the intensity probability density value p K of each intensity interval, and combine each intensity probability density value into a matrix [p 1 , p 2 ..., p K to obtain the K-dimensional global intensity descriptor of a single segment. The intensity probability density is calculated as shown in the following formula:

[0034]

[0035] Among them,

[0036] ΔI represents the intensity interval length, ΔI = b k+1 -b k .

[0037] Furthermore, in step S4, the detection of leakage segments and non-leakage segments by fusing the global intensity descriptor and the random forest classifier includes the following steps:

[0038] S4.1, Set the random forest parameters: the number of decision trees n_estimators, the criterion for measuring the quality of feature selection criterion, and the maximum depth of the decision tree max_depth;

[0039] S4.2, Construct n_estimators decision trees, divide the data into a training set and a test set, generate M sub-datasets by sampling with replacement from the training set, and each sub-dataset is used for training a single decision tree. The number of samples in the sub-dataset is M 1 and contains duplicate samples;

[0040] S4.3, Use the global descriptor as the input feature and select k sub-features from the input features;

[0041] S4.4, For each decision tree, select the optimal splitting feature and splitting node by maximizing criterion, and recursively split the nodes of the decision tree. If the maximum depth of the tree max_depth is reached, stop splitting;

[0042] S4.5, Use the trained multiple decision trees to classify the test set, and output the prediction result according to the majority voting mechanism, as shown in the following formula:

[0043]

[0044] where,

[0045] argmax is the maximum value function,

[0046] is the final classification result of the segment,

[0047] T m (x) is the classification result of the m-th tree,

[0048] c is the segment label, the label of the leaking segment is 0, and the label of the non-leaking segment is 1.

[0049] Beneficial effects: Compared with the prior art, the significant advantages of the present invention are:

[0050] 1. The present invention breaks through the dependence of traditional tunnel leakage detection methods on single-dimensional intensity features. By constructing a global intensity descriptor, the dimension of the intensity feature is effectively expanded, and the accurate characterization of the segment strength distribution is realized.

[0051] 2. The present invention uses the strength distribution of a single segment as the input data of the classifier, greatly reducing the computational complexity and significantly improving the detection efficiency.

[0052] 3. The present invention provides key technical support for the efficient detection and accurate positioning of leaking segments in subway tunnels, laying an important foundation for ensuring the safe operation of subway tunnels.

[0053] 4. The present invention effectively characterizes the strength distribution of a single segment by constructing a global strength descriptor, and accurately locates leaking segments and non-leaking segments by utilizing the strength distribution differences between them, providing important technical support and guarantee for the safe operation and maintenance of subway tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic flow diagram of the present invention.

[0055] Figure 2 It is a scatter plot of the point cloud of the subway tunnel obtained in the implementation of the present invention.

[0056] Figure 3 It is a scatter plot of the point cloud of the tunnel extracted for a single segment in the implementation of the present invention.

[0057] Figure 4 It is a schematic diagram of the generation of the global strength descriptor in the implementation of the present invention.

[0058] Figure 5 It is a schematic diagram of the fusion of the global strength descriptor and the random forest classifier in the implementation of the present invention to detect leaking segments and non-leaking segments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0060] The present invention will be further described below with reference to the accompanying drawings.

[0061] As Figure 1 shown, this embodiment provides a method for constructing and applying a global descriptor for detecting leakage of tunnel segments, including the following steps:

[0062] Step 1: A mobile laser scanning system scans the subway tunnel to obtain the point cloud P(x i , y i , z i ) i＝1,2,…,n , as Figure 2 shown.

[0063] Step 2: Locate the circumferential joints according to the spatial position relationship between the bolt holes and the circumferential joints, and then extract individual segments, as Figure 3 shown. Specifically as follows:

[0064] (1) The collected tunnel point cloud is P(x i , y i , z i )i＝1,2,…,n , project the point cloud onto the XOZ plane to obtain a two-dimensional point set (x i , z i ); i＝1,2,…n Perform elliptical fitting on the two-dimensional point set to obtain the center coordinates (x 0 , z 0 ); Calculate the distance from each point in the two-dimensional point set to the center of the circle, set the distance threshold δ to 2.752 m, if the distance from a certain point to the center of the circle is greater than the distance threshold, then determine and extract this point as a bolt hole point;

[0065] (2) Use the DBSCAN algorithm to cluster the mileage of the bolt holes. Among them, the mileage of the bolt is the y coordinate value. The algorithm steps are as follows:

[0066] ① Set the neighborhood radius ε to 0.15 m and the density threshold MinPts to 10. Randomly select an unvisited point and calculate the number of its neighborhood points within the given radius Eps. If the number of neighborhood points exceeds the preset density threshold minPts, then mark this point as a core point; otherwise, temporarily mark it as a border point. At the same time, regard the points outside the neighborhood as noise points;

[0067] ② For the core points, traverse all the points in their neighborhoods and group these points into the same cluster. If there are other core points in the cluster, then include these core points and their neighborhood points in the current cluster;

[0068] ③ Repeat steps ① and ② until all points are visited;

[0069] (3) After clustering the bolt holes using DBSCAN, each segment will form two clusters, which respectively represent two groups of bolt holes with similar mileage values. Subsequently, take the average value of the y coordinate values of the corresponding clusters in adjacent segments to determine the position loc of the circumferential joint v ;

[0070] (4) According to the circumferential joint position loc v , extract a single segment according to the following formula:

[0071]

[0072] where v is the index of the circumferential joint.

[0073] Step three: As Figure 4 shown, generate a global intensity descriptor for a single segment, specifically as follows:

[0074] (1) Normalize the reflection intensity of a single segment according to the following formula:

[0075]

[0076] where I iis the reflection intensity value for each point, I i ' is the normalized reflection intensity value for each point, and N is the total number of points for a single segment;

[0077] (2) The range of the normalized reflection intensity value is [0, 1], and the intensity interval range is [b k , b k+1 ), k = 1, 2,... K. Set the intensity interval width to 0.05, and divide [0, 1] into 20 intensity intervals;

[0078] (3) Calculate the number of points n K falling into each intensity interval through the indicator function, as shown in the following formula:

[0079]

[0080] where δ(I i ' ∈ [b k , b k+1 )) represents the indicator function. When a certain point falls into the intensity interval [b k , b k+1 )), the value of this function is 1; otherwise, the value of this function is 0;

[0081] (4) Calculate the intensity probability density value p K for each intensity interval, and combine each intensity probability density value into a matrix [p 1 , p 2 ..., p K to obtain a 20-dimensional global intensity descriptor for a single segment. The intensity probability density is calculated as shown in the following formula:

[0082]

[0083] where ΔI represents the intensity interval length, and ΔI = b k+1 - b k .

[0084] Step Four: Integrate the global intensity descriptor and the random forest classifier to detect leaking segments and non-leaking segments, specifically as follows:

[0085] (1) Set the random forest parameters: the number of decision trees n_estimators is 100, the criterion for measuring the quality of feature selection is entropy, and the maximum depth max_depth of the decision tree is 6;

[0086] (2) Construct 100 decision trees, divide the data into a training set and a test set, generate 100 sub-datasets by sampling with replacement from the training set, and each sub-dataset is used for training a single decision tree. The number of samples M 1 in the sub-dataset is the total number of points and includes duplicate samples;

[0087] (3) Use the global descriptor as the input feature and randomly select 5 features from the input features;

[0088] (4) For each decision tree, select the optimal splitting feature and splitting node by maximizing the criterion, and recursively split the decision tree. If the maximum depth max_depth of the tree is reached, stop growing;

[0089] (5) As Figure 5 shown, use the grown multiple decision trees to classify the test set, and output the prediction result according to the majority voting mechanism, as shown in the following formula:

[0090]

[0091] where argmax is the maximum value function, is the final classification result of the segment, T m (x) is the classification result of the m-th tree, c is the segment label, the label of the leaking segment is 0, and the label of the non-leaking segment is 1.

[0092] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for constructing and applying a global descriptor for detecting tunnel segment leakage, characterized in that: The following steps are involved: S1, the mobile laser scanning system scans the subway tunnel and obtains the subway tunnel point cloud data; S2, locate the annular seam according to the spatial position relationship between the bolt hole and the annular seam, and then extract a single segment; S3, generating a global intensity descriptor for a single segment; S4, Fusion of global intensity descriptor and random forest classifier to detect leaking and non-leaking segments.

2. The method for constructing and applying a global descriptor for detecting tunnel segment leakage according to claim 1 is characterized in that: In step S2, the method of locating the annular seam according to the spatial position relationship between the bolt hole and the annular seam and then extracting a single segment includes the following steps: S2.1, the collected tunnel point cloud is P(x i ,y i ,z i ) i＝1,2,…,n , project the point cloud to the XOZ plane and obtain a two-dimensional point set (x i ,z i ) i＝1,2,…n ; Perform ellipse fitting on the two-dimensional point set to obtain the coordinates of the center of the circle (x0, z0); calculate the distance from each point in the two-dimensional point set to the center of the circle, set the distance threshold δ, and if the distance from a point to the center of the circle is greater than the distance threshold, the point is determined to be a bolt hole point and extracted; S2.2, using DBSCAN algorithm to cluster the mileage of bolt holes; S2.3, after clustering the bolt holes using DBSCAN, each segment will form two clusters, representing two groups of bolt holes with similar mileage values. Then, the y coordinate values ​​of the corresponding clusters in the adjacent segments are averaged to determine the location of the annular seam. v ; S2.4, according to the annular gap position loc v , extract a single segment as follows: Where v is the index of the annular gap.

3. The method for constructing and applying a global descriptor for detecting tunnel segment leakage according to claim 2 is characterized in that: In step S2.2, the mileage clustering of bolt holes using the DBSCAN algorithm includes the following steps: S2.21, set the two parameters of neighborhood radius ε and density threshold MinPts, randomly select an unvisited point, calculate the number of its neighborhood points within the given radius Eps, if the number of neighborhood points exceeds the preset density threshold minPts, then mark the point as a core point, otherwise, temporarily mark it as an edge point, and at the same time, treat the points outside the neighborhood as noise points; S2.22, for a core point, traverse all points in its neighborhood and classify these points into the same cluster. If the cluster contains other core points, these core points and their neighborhood points are included in the current cluster; S2.23, repeat steps S2.21 and S2.22 until all points are visited.

4. The method for constructing and applying a global descriptor for detecting tunnel segment leakage according to claim 2, characterized in that: In step S2.2, the mileage of the bolt is the y coordinate value.

5. The method for constructing and applying a global descriptor for detecting tunnel segment leakage according to claim 1, characterized in that: In step S3, generating a global intensity descriptor for a single segment includes the following steps: S3.1, normalize the reflection intensity of a single segment according to the following formula: in, I i is the reflection intensity value of each point, I i 'For each point the normalized reflection intensity value, N is the total number of points in a single segment; S3.2, the normalized reflection intensity value range is [0,1], and the intensity interval range is [b k ,b k+1 ), k=1,2,...K, according to the intensity interval width b k+1 -b k , divide [0,1] into K intensity intervals; S3.3, calculate the number of points n falling into each intensity interval by the indicator function K , as shown below: in, δ(I i '∈[b k ,b k+1 )) represents the indicator function. When a point falls into the intensity interval [b k ,b k+1 ), the function takes the value of 1, otherwise, the function takes the value of 0; S3.4, calculate the intensity probability density value p for each intensity interval K , each intensity probability density value is combined into a matrix [p1,p2...,p K ], and the K-dimensional global intensity descriptor of a single segment is obtained. The intensity probability density is calculated as follows: in, ΔI represents the length of the intensity interval, ΔI = b k+1 -b k .

6. The method for constructing and applying a global descriptor for detecting tunnel segment leakage according to claim 1, characterized in that: In step S4, the fusion of the global intensity descriptor and the random forest classifier to detect leaking segments and non-leaking segments includes the following steps: S4.1, set random forest parameters: number of decision trees n_estimators, standard criterion for measuring feature selection quality, maximum depth of decision tree max_depth; S4.2, build n_estimators decision trees, divide the data into training set and test set, generate M sub-datasets from the training set with replacement sampling, each sub-dataset is used for single decision tree training, the number of sub-dataset samples is M1 and contains repeated samples; S4.3, taking the global descriptor as input feature, and selecting k sub-features from the input feature; S4.4, for each decision tree, by maximizing the criterion, selecting the optimal splitting feature and splitting node, recursively splitting the nodes of the decision tree, and stopping the splitting if the maximum depth of the tree max_depth is reached; S4.5, use the trained multiple decision trees to classify the test set and output the prediction results according to the majority voting mechanism, as shown in the following formula: in, argmax is the maximum value function, is the final classification result of the segment, T m (x) is the classification result of the mth tree, c is the segment label, the leaking segment label is 0, and the non-leaking segment label is 1.

Images