Subway shield tunnel leakage detection method based on Z-order curve point cloud serialization
Through the combination of Z-sequence point cloud serialization and deep learning network, high-precision detection of subway shield tunnel leakage is achieved, solving the problems of insufficient detection accuracy and low efficiency in the existing technology, and improving the accuracy and robustness of the detection.
Patent Information
- Application Number
- CN202510349828.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-24
- Publication Date
- 2025-08-12
AI Technical Summary
In the prior art, the subway shield tunnel leakage detection method is insufficient, the robustness is not strong, and the three-dimensional structural information is insufficient, resulting in low detection efficiency and low accuracy.
Using a method based on Z-sequence point cloud serialization, three-dimensional point cloud data is obtained through mobile lidar, denoising and manual annotation is performed, and the point cloud data is serialized and encoded with Z-sequence point cloud data, training the model and leak detection, and finally the results are reverse serialized to three-dimensional space.
It improves the accuracy and efficiency of leakage detection, maintains the spatial locality of point clouds, reduces the computational complexity, shortens data processing and training time, and improves the accuracy and reliability of detection.
Smart Images
Figure CN120471823A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of tunnel leakage detection and structural health monitoring, and in particular to a subway shield tunnel leakage detection method based on Z-sequence curve point cloud serialization. Background Art
[0002] During the operation of subway shield tunnels, leakage problems can have a serious impact on tunnel structure and operational safety. Traditional leakage detection methods rely on manual inspections and simple sensor data analysis, which are inefficient and inaccurate. With the development of lidar technology, the use of point cloud data for tunnel leakage detection has become an effective method. However, the high dimensionality and large-scale nature of point cloud data place higher demands on data processing and analysis. Z-order curves, as an efficient spatial serialization method, can maintain the spatial locality of point cloud data, helping deep learning models better learn leakage characteristics. Summary of the Invention
[0003] Purpose of the invention: The purpose of the present invention is to overcome the problems of insufficient accuracy, weak robustness and insufficient utilization of three-dimensional structural information in the existing subway shield tunnel leakage detection methods, and to provide a subway shield tunnel leakage detection method based on Z-order curve point cloud serialization to achieve accurate detection of leakage points.
[0004] Technical solution: The present invention provides a method for detecting leakage in a subway shield tunnel based on Z-sequence curve point cloud serialization, comprising the following steps:
[0005] (1) Obtaining original 3D point cloud data inside the subway shield tunnel based on mobile LiDAR;
[0006] (2) De-noising the original 3D point cloud data;
[0007] (3) Manually annotate the denoised tunnel point cloud data with leakage instances and construct a tunnel leakage point cloud dataset;
[0008] (4) Divide the tunnel leakage point cloud dataset into a training set and a validation set, and perform data enhancement on the training set;
[0009] (5) Based on the tunnel leakage detection data set obtained in step (4), the point cloud data is serialized and encoded using a Z-sequence curve to map the three-dimensional point cloud to a one-dimensional sequence;
[0010] (6) Input the serialized point cloud data and the intensity value corresponding to each point into the deep learning network model, perform iterative model training, and obtain the optimal model;
[0011] (7) Based on the optimal model, the point cloud data of the tunnel to be inspected is predicted to obtain serialized leakage detection results; based on the inverse serialization method, the detection results are mapped back to the real three-dimensional space coordinate system to determine the three-dimensional spatial distribution of the tunnel leakage area.
[0012] Furthermore, the implementation process of step (2) is as follows:
[0013] Point cloud denoising is a denoising algorithm based on spatial distribution to reduce the noise points in point cloud data. First, search for each point p in the point cloud. i The k nearest neighboring points of the local neighborhood point set N(p i ); calculate the neighborhood point set N(p i ) in the Euclidean distance of k neighboring points:
[0014] d ij =||p i -p j ||
[0015] The mean μ and variance σ of the Euclidean distance are calculated, and a distance range is defined: [μ-σ, μ+σ]. Outliers outside this range are filtered and removed to obtain the denoised point cloud data.
[0016] Furthermore, the manual labeling in step (3) is to mark the points with tunnel leakage as leakage class, i.e. Label = 1, and the points without leakage as background class, i.e. Label = 0, to construct the training data set D train and validation dataset D val .
[0017] Furthermore, the data enhancement in step (4) includes random rotation, translation, scaling and intensity perturbation.
[0018] Furthermore, the implementation process of step (5) is as follows:
[0019] For each point p = (x, y, z), convert it to a fixed-length binary representation, assuming that each coordinate is represented by an n-bit binary number:
[0020] x=x n-1 x n-2 ...x1x0
[0021] y=y n-1 y n-2 ...y1y0
[0022] z=z n-1 z n-2 ...z1z0
[0023] Among them, x i ,y i ,zi The i-th binary bit, i=0 is the lowest bit;
[0024] Arrange the binary bits of each dimension alternately in bit order to form a new binary sequence; for the binarized x i ,y i ,z i , the interleaving order is x, y, z dimensions alternating in turn; for each bit position k:
[0025] Z 3k =x k
[0026] Z 3k+1 =y k
[0027] Z 3k+2 =z k
[0028] Finally, the interleaved bit sequence is:
[0029]
[0030] Among them, ∪ means that each group (x i ,y i ,z i ) are connected in sequence to form a new bit sequence; the interleaved bit sequence Z is merged into a single integer, which is the Z-order index of the point:
[0031]
[0032] Among them, 2 k is the bit weight, Z k is the kth bit in the interleaved bit sequence:
[0033]
[0034] Here, k starts from 0, k=0 is the lowest bit, and mod3 represents the modular operation, that is, the remainder after division by 3.
[0035] Furthermore, the deep learning network in step (6) includes an encoder and a decoder; the encoder extracts multi-level features from the input sequence, and the decoder performs step-by-step upsampling and feature fusion on the high-level semantic information; a batch of serialized inputs in the training set is Where L is the sequence length, d is the feature dimension of each point, including xyz coordinates and intensity feature I; the network output prediction probability is Y i It represents the probability that the i-th point is judged as a leakage class.
[0036] Furthermore, the iterative model training in step (6) includes:
[0037] Use the Lovasz loss function L Lovasz Optimize to solve the problem of sample imbalance in the dataset:
[0038]
[0039] in, is the predicted value The i-th prediction after sorting from high to low confidence, ΔJaccard is the Jaccard distance, which is calculated point by point by iterative calculation of the subset of the sorted prediction set;
[0040] During the training process, record the mean intersection over union (mIoU) and mean accuracy (mAcc) metrics:
[0041]
[0042] Among them TP i , FP i , FN i , TN i They represent the true positive, false positive, false negative, and true negative of the Nth category respectively; the model parameters with the highest mIoU and mAcc indicators on the validation set are selected as the best model.
[0043] Furthermore, the implementation process of step (7) is as follows:
[0044] First, convert the Z-order index Z of the serialized detection result into a fixed-length binary bit sequence; each coordinate is represented by an n-bit binary number, and the binary representation of the entire Z-order index is 3n bits:
[0045] Z=Z 3(n-1) Z 3(n-1)+1 Z 3(n-1)+2 ...Z3Z2Z1Z0
[0046] Among them, Z i Represents the i-th binary bit of the Z-order index Z, where i=0 is the lowest bit;
[0047] According to the bit interleaving rule, the Z-order bit sequence Z is separated into three binary bit sequences of coordinates x, y, z; for each bit position k, k ranges from 0 to (n-1):
[0048]
[0049] The deinterleaved binary bit sequence x, y, z is converted back to decimal coordinates to obtain the final leakage detection result.
[0050] Beneficial effects: Compared with the existing technology, the beneficial effects of the present invention are: the present invention combines deep learning with Z-order curve point cloud serialization technology to achieve high-precision detection of subway shield tunnel leakage; the three-dimensional point cloud data is converted into a one-dimensional sequence through the Z-order curve, which effectively maintains the spatial locality of the data and ensures the relative position relationship of adjacent points in the sequence, thereby maximizing the retention of the spatial structure information of the point cloud; this serialization method significantly reduces the dimension of the data, reduces the computational complexity of the deep learning model, and improves the training and inference efficiency of the model; in addition, the recursive fractal characteristics of the Z-order curve make the data processing process more efficient and shorten the time cost of data preprocessing and model training; by optimizing the data processing process and utilizing the deep learning model, not only the accuracy and reliability of leakage detection are improved, but also the overall detection efficiency is greatly improved, which is suitable for leakage monitoring and maintenance of actual subway tunnels, and has significant technical advantages and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 is a flow chart of the present invention;
[0052] Figure 2 This is a schematic diagram of the subway shield tunnel leakage dataset collected by the present invention;
[0053] Figure 3 Schematic diagram of the serialized connection method of the Z-sequence curve of the three-dimensional point cloud of the present invention;
[0054] Figure 4 This is a schematic diagram of the Z-sequence curve serialization of the tunnel leakage point cloud of the present invention;
[0055] Figure 5 This is the leakage detection result after serialization of the Z-order curve point cloud in the present invention. DETAILED DESCRIPTION
[0056] The present invention will be described in further detail below with reference to the accompanying drawings.
[0057] like Figure 1 As shown, the present invention provides a method for detecting leakage in a subway shield tunnel based on Z-sequence curve point cloud serialization, comprising the following steps:
[0058] Step 1: Use a mobile lidar scanning system to obtain the point cloud data P(x, y, z) of the subway shield tunnel.
[0059] Step 2: For the point cloud data P(x,y,z) of the subway shield tunnel, search for each point p in P(x,y,z) i The k nearest neighboring points of the local neighborhood point set N(p i ).
[0060] Calculate the neighborhood point set N(p i) in the Euclidean distance of k neighboring points:
[0061] d ij =||p i -p j ||
[0062] Calculate the mean μ and variance σ of these distances, define a distance range: [μ-σ,μ+σ], filter and remove abnormal points outside this range, and obtain denoised point cloud data.
[0063] For the pre-processed point cloud data, artificial leakage instance labeling is performed, and the points with tunnel leakage are marked as leakage class (Label = 1), and the points without leakage are marked as background class (Label = 0). The training data set D is constructed. train and validation dataset D val .
[0064] Data enhancement is performed on the point cloud data in the leakage detection dataset, including random rotation, translation, scaling and intensity perturbation. Finally, the tunnel leakage dataset used for network training is obtained. Some tunnel leakage datasets are as follows: Figure 2 shown.
[0065] Step 3: Use Z-order curve to serialize point cloud. The serialization connection method of Z-order in three-dimensional space is as follows: Figure 3 The specific implementation is as follows: for each point p = (x, y, z), convert it into a fixed-length binary representation. Assume that each coordinate is represented by an n-bit binary number:
[0066] x=x n-1 x n-2 ...x1x0
[0067] y=y n-1 y n-2 ...y1y0
[0068] z=z n-1 z n-2 ...z1z0
[0069] where x i ,y i ,z i is the i-th binary bit (i=0 is the lowest bit).
[0070] The binary bits of each dimension are arranged alternately in bit-level order to form a new binary sequence. i ,y i ,z i The interleaving order is x, y, and z dimensions alternating in sequence.
[0071] For each bit position k (from 0 to n-1):
[0072] Z 3k =x k
[0073] Z 3k+1 =y k
[0074] Z 3k+2 =z k
[0075] Finally, the interleaved bit sequence is:
[0076]
[0077] Among them, ∪ means that each group (x i ,y i ,z i ) are connected in sequence to form a new bit sequence.
[0078] Combine the interleaved bit sequence Z into a single integer, which is the Z-order index of the point:
[0079]
[0080] Among them, 2 k is the bit weight, Z k is the kth bit in the interleaved bit sequence (starting from 0, k = 0, the least significant bit):
[0081]
[0082] Among them, mod3 represents the modulus operation (remainder operation), that is, the remainder after division by 3. The tunnel leakage after point cloud serialization using Z-order curve is as follows Figure 4 shown.
[0083] Step 4: Input the serialized point cloud data and its corresponding point cloud intensity values into the PointTransformer deep learning network for model training. The PointTransformer network consists of four encoder and four decoder modules. The encoder extracts multi-level features from the input sequence, and the decoder performs step-by-step upsampling and feature fusion of high-level semantic information.
[0084] The training steps include: a batch of serialized inputs in the training set is Where L is the sequence length, d is the feature dimension of each point, including xyz coordinates and intensity feature I. The network output prediction probability is where Y i It represents the probability that the i-th point is judged as a leakage class.
[0085] Point Transformer deep learning network uses Lovasz loss function L Lovasz Optimize to solve the problem of sample imbalance in the dataset:
[0086]
[0087] in, is the predicted value The i-th prediction is sorted from high to low confidence, and ΔJaccard is the Jaccard distance, which is calculated point by point by iterative calculation of the subset of the sorted prediction set.
[0088] During the training process, record the mIoU (mean intersection over union) and mAcc (mean accuracy) metrics:
[0089]
[0090] Among them, TP i , FP i , FN i ,TN i They represent the true positive, false positive, false negative and true negative of the Nth category respectively.
[0091] The model parameters with the highest mIoU and mAcc metrics on the validation set are selected as the final optimal model. Step 5: Use the trained optimal model to perform leakage detection on the tunnel point cloud data to be inspected. The serialized results obtained from the detection are pseudo-serialized to obtain the true 3D coordinates of the leakage points.
[0092] First, convert the Z-order index Z of the serialized detection result into a fixed-length binary bit sequence. Each coordinate is represented by an n-bit binary number, so the binary representation of the entire Z-order index is 3n bits:
[0093] Z=Z 3(n-1) Z 3(n-1)+1 Z 3(n-1)+2 ...Z3Z2Z1Z0
[0094] where Z i Represents the i-th binary bit of the Z-order index Z (i=0 is the least significant bit).
[0095] According to the rules of bit interleaving, the Z-order bit sequence Z is separated into three binary bit sequences of coordinates x, y, z. For each bit position k (from 0 to n-1):
[0096]
[0097] Finally, the deinterleaved binary bit sequence x, y, z is converted back to decimal coordinates to obtain the final leakage detection result. The detection results are displayed in a visual form. After using the Z-order point cloud serialization, the PointTransformer network is trained and the leakage detection results obtained by predicting the test data are as follows Figure 5 As shown in Figure 2, the results show that the network serialized using Z-order point cloud can accurately identify tunnel leakage.
[0098] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A subway shield tunnel leakage detection method based on Z-sequence curve point cloud serialization, characterized in that: The following steps are involved: (1) Obtaining original 3D point cloud data inside the subway shield tunnel based on mobile LiDAR; (2) De-noising the original 3D point cloud data; (3) Manually annotate the denoised tunnel point cloud data with leakage instances and construct a tunnel leakage point cloud dataset; (4) Divide the tunnel leakage point cloud dataset into a training set and a validation set, and perform data enhancement on the training set; (5) Based on the tunnel leakage detection data set obtained in step (4), the point cloud data is serialized and encoded using a Z-sequence curve to map the three-dimensional point cloud to a one-dimensional sequence; (6) Input the serialized point cloud data and the intensity value corresponding to each point into the deep learning network model, perform iterative model training, and obtain the optimal model; (7) Predict the point cloud data of the tunnel to be inspected based on the optimal model to obtain serialized leakage detection results; Based on the inverse serialization method, the detection results are mapped back to the real three-dimensional space coordinate system to determine the three-dimensional spatial distribution of the tunnel leakage area.
2. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The implementation process of step (2) is as follows: Point cloud denoising is a denoising algorithm based on spatial distribution to reduce the noise points in point cloud data. First, search for each point p in the point cloud. i The k nearest neighboring points of the local neighborhood point set N(p i ); calculate the neighborhood point set N(p i ) in the Euclidean distance of k neighboring points: d ij =||p i -p j || The mean μ and variance ρ of the Euclidean distance are calculated, and a distance range is defined: [μ-σ, μ+σ]. Outliers outside this range are filtered and removed to obtain the denoised point cloud data.
3. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The manual labeling in step (3) is to mark the points with tunnel leakage as leakage class, i.e. Label = 1, and the points without leakage as background class, i.e. Label = 0, to construct the training data set D train and validation dataset D val .
4. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The data enhancement in step (4) includes random rotation, translation, scaling and intensity perturbation.
5. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The implementation process of step (5) is as follows: For each point p = (x, y, z), convert it to a fixed-length binary representation, assuming that each coordinate is represented by an n-bit binary number: x=x n-1 x n-2 ...x1x0 y=y n-1 and n-2 ...y1y0 from=from n-1 from n-2 ...z1z0 Among them, x i ,y i ,z i The i-th binary bit, i=0 is the lowest bit; the binary bits of each dimension are arranged alternately in the order of bit level to form a new binary sequence; for the binarized x i ,y i ,z i , the interleaving order is x, y, z dimensions alternating in turn; for each bit position k: Z 3k =x k Z 3k+1 =y k WITH 3k+2 =z k Finally, the interleaved bit sequence is: Among them, ∪ means that each group (x i ,y i ,z i ) are connected in sequence to form a new bit sequence; the interleaved bit sequence Z is merged into a single integer, which is the Z-order index of the point: Among them, 2 k is the bit weight, Z k is the kth bit in the interleaved bit sequence: Here, k starts from 0, k=0 is the lowest bit, and mod3 represents the modular operation, that is, the remainder after division by 3.
6. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The deep learning network in step (6) includes an encoder and a decoder; the encoder extracts multi-level features from the input sequence, and the decoder performs step-by-step upsampling and feature fusion on high-level semantic information; The serialized input of a batch in the training set is Where L is the sequence length, d is the feature dimension of each point, including xyz coordinates and intensity feature I; the network output prediction probability is Y i It represents the probability that the i-th point is judged as a leakage class.
7. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The iterative model training in step (6) includes: Use the Lovasz loss function L Lovasz Optimize to solve the problem of sample imbalance in the dataset: in, is the predicted value The i-th prediction after sorting from high to low confidence, ΔJaccard is the Jaccard distance, which is calculated point by point by iterative calculation of the subset of the sorted prediction set; During the training process, record the mean intersection over union (mIoU) and mean accuracy (mAcc) metrics: Among them TP i , FP i , FN i , TN i They represent the true positive, false positive, false negative, and true negative of the Nth category respectively; the model parameters with the highest mIoU and mAcc indicators on the validation set are selected as the best model.
8. The subway shield tunnel leakage detection method based on Z-order curve point cloud serialization according to claim 1 is characterized in that: The implementation process of step (7) is as follows: First, convert the Z-order index Z of the serialized detection result into a fixed-length binary bit sequence; each coordinate is represented by an n-bit binary number, and the binary representation of the entire Z-order index is 3n bits: Z=Z 3(n-1) WITH 3(n-1)+1 WITH 3(n-1)+2 ...Z3Z2Z1Z0 Among them, Z i Represents the i-th binary bit of the Z-order index Z, where i=0 is the lowest bit; According to the bit interleaving rule, the Z-order bit sequence Z is separated into three binary bit sequences of coordinates x, y, z; for each bit position k, k ranges from 0 to (n-1): The deinterleaved binary bit sequence x, y, z is converted back to decimal coordinates to obtain the final leakage detection result.