Method and device for identifying section insulators of a catenary based on 3D imaging technology

Abstract

The application discloses a contact net section insulator recognition method and device based on a 3D imaging technology, and the method comprises the following steps: collecting point cloud data of a contact net section insulator; pre-processing the collected point cloud data; converting the pre-processed point cloud data into a three-dimensional matrix; inputting the three-dimensional matrix as a training set into a convolutional neural network for model training; testing the trained convolutional neural network model, and if the test result meets the requirements, outputting the model for section insulator recognition, otherwise, continuing the training. The application adopts a deep convolutional neural network to intelligently recognize the contact net section insulator, and improves the recognition efficiency and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of contact wire insulator identification technology, specifically relating to a method and device for identifying contact wire segment insulators based on 3D imaging technology. Background Technology

[0002] Sectional insulators are insulating devices used when electrical sections of the overhead contact system are segmented. When a fault occurs in a section of the contact system, the isolating switch at the sectional insulator is opened to disconnect power to that section of the contact system, while other sections are powered normally, improving the reliability and flexibility of the contact system operation. Due to voltage differences, installation defects, and pantograph wear, arcing may occur when a train passes over a sectional insulator, damaging both the insulator and the contact system. In severe cases, this can cause the pantograph of the electric locomotive to strike. Therefore, fault detection and identification of sectional insulators directly affect the operational safety of electric locomotives.

[0003] Currently, the main methods for detecting and identifying segmented insulators include multi-point detection and online monitoring. Multi-point detection involves detecting multiple points on the segmented insulator, such as the slide rail connection clamp and the contact wire connection clamp, to identify the location information of the segmented insulator and thus determine whether it is a segmented insulator. Online monitoring uses a camera to take continuous pictures and manually examines a fixed number of frames to determine whether it is a segmented insulator, which is less efficient. Summary of the Invention

[0004] To address the low efficiency of existing contact wire segment insulator identification technologies, this invention provides a contact wire segment insulator identification method based on 3D imaging technology. This invention employs a deep convolutional neural network for intelligent identification of contact wire segment insulators, improving both identification efficiency and accuracy.

[0005] This invention is achieved through the following technical solution:

[0006] A method for identifying contact wire segment insulators based on 3D imaging technology includes:

[0007] Collect point cloud data of the contact wire segment insulators;

[0008] Preprocess the collected point cloud data;

[0009] Convert the preprocessed point cloud data into a 3D matrix;

[0010] The three-dimensional matrix is ​​used as a training set and input into the convolutional neural network for model training.

[0011] The trained convolutional neural network model is tested. If the test results meet the requirements, the model is output for segmented insulator identification; otherwise, training continues.

[0012] Preferably, the preprocessing procedure of the present invention specifically includes:

[0013] The collected point cloud data is denoised.

[0014] Downsampling is performed on the denoised point cloud data to ensure consistency in the number of data points.

[0015] Preferably, the denoising process of the present invention specifically employs the KNN algorithm for clustering to remove noise points;

[0016] The downsampling process specifically involves: selecting a rectangular window of a preset size, dividing the rectangular window into smaller windows of the same size, taking one point from each window, and replacing the total number of points in the smaller window with the average value.

[0017] Preferably, the step of converting the preprocessed point cloud data into a three-dimensional matrix in this invention specifically includes:

[0018] The data is divided into three matrices: X×Y×1, Y×Z×1, and X×Z×1, based on the X, Y, and Z axes respectively.

[0019] By concatenating the three matrices according to their dimensions, a three-dimensional matrix of size W×H×3 is obtained; where W and H represent the width and height of the three-dimensional matrix, respectively.

[0020] Preferably, the model training process of the present invention specifically includes:

[0021] The three-dimensional matrix is ​​normalized to map the data uniformly to the [0,1] interval;

[0022] The dataset obtained after normalization is divided into a training set and a test set;

[0023] The ResNet-18 convolutional neural network model was trained using the training set.

[0024] Preferably, the testing process of the present invention specifically includes:

[0025] Input the test set into the trained model;

[0026] The confusion matrix is ​​obtained based on the model's classification output.

[0027] Precision and recall are calculated based on the confusion matrix.

[0028] Determine whether the model needs to continue training based on accuracy and recall.

[0029] Preferably, the accuracy calculation formula of the present invention is as follows:

[0030]

[0031] The recall rate is calculated using the following formula:

[0032]

[0033] Wherein, TP represents the number of correctly identified segmented insulators, FP represents the number of other components incorrectly identified as segmented insulators, FN represents the number of segmented insulators incorrectly identified as other components, and TN represents the number of correctly identified non-segmented insulators.

[0034] Secondly, the present invention proposes a contact network segment insulator identification device based on 3D imaging technology, including a data acquisition module, a data preprocessing module, a data conversion module, a model training module, and a model testing module;

[0035] The data acquisition module is used to acquire point cloud data of the contact wire segment insulator;

[0036] The data preprocessing module preprocesses the collected point cloud data;

[0037] The data conversion module converts the preprocessed point cloud data into a three-dimensional matrix.

[0038] The model training module uses a three-dimensional matrix as a training set and inputs it into a convolutional neural network for model training.

[0039] The model testing module is used to test the convolutional neural network model after training. If the test is successful, the model is output for segmented insulator identification; otherwise, training continues.

[0040] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described in the present invention.

[0041] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the present invention.

[0042] The present invention has the following advantages and beneficial effects:

[0043] 1. This invention collects three-dimensional point cloud data of contact network segment insulators, preprocesses and transforms the data to construct a training dataset for training a convolutional neural network model, and tests the model to obtain an insulator recognition model for automatic identification of contact network segment insulators. This invention has low computational cost and improves recognition efficiency and accuracy.

[0044] 2. This invention uses the ResNet-18 model, which has a fast convergence speed and short time, further improving the recognition efficiency. Attached Figure Description

[0045] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0046] Figure 1 This is a schematic diagram of the identification method according to an embodiment of the present invention.

[0047] Figure 2 The image shows a 3D image of a segmented insulator acquired in an embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating distance-based clustering within the scope of an embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram of downsampling in an embodiment of the present invention.

[0050] Figure 5 This is a schematic diagram of point cloud data preprocessing according to an embodiment of the present invention.

[0051] Figure 6 This is a diagram of the feature extractor according to an embodiment of the present invention.

[0052] Figure 7 This is a schematic diagram of the ResNet-18 network structure according to an embodiment of the present invention.

[0053] Figure 8 This is a schematic diagram of the block structure according to an embodiment of the present invention.

[0054] Figure 9 This is a structural diagram of a computer device according to an embodiment of the present invention.

[0055] Figure 10 This is a block diagram illustrating the principle of the identification device according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0057] Example 1

[0058] This embodiment proposes a method for identifying segmented insulators in overhead contact lines based on 3D imaging technology, such as... Figure 1 As shown, it includes:

[0059] Step 1: Collect point cloud data of the contact wire segment insulators.

[0060] In this embodiment, a 3D camera can be used to collect point cloud data of the contact wire segment insulators, such as... Figure 2 As shown.

[0061] Step 2: Preprocess the collected point cloud data.

[0062] In this embodiment, in order to ensure that the number of point cloud data of the segmented insulators acquired by the 3D camera is the same, the acquired point cloud data is subjected to noise reduction and downsampling processing.

[0063] In this embodiment, the noise reduction process for point cloud data is specifically performed as follows:

[0064] This embodiment uses a clustering algorithm to cluster the data, thereby removing noisy points. This embodiment improves upon the KNN algorithm to achieve clustering. The idea behind KNN is that if a point has k nearest neighbors in the feature space, then these points belong to the same category. This embodiment sets a K value to compare the distance between two points. Point cloud data acquired by 3D cameras often contains many points that do not actually exist, which can significantly affect the processing of point clouds. Therefore, it is necessary to denoise the data. Euclidean clustering is used to calculate the distance between two points. The calculation formula is:

[0065]

[0066] Where (X1, X2) and (Y1, Y2) represent the coordinates of the two points respectively.

[0067] Clustering is performed based on the K value to remove noise points, such as... Figure 3 As shown in the diagram. Circles in the diagram represent clusters. After clustering, each cluster is merged based on its centroid. The merging method is as follows:

[0068] d1 = abs(AB)

[0069] Where A and B represent the coordinates of the center points of the two clusters, respectively.

[0070] The downsampling process in this embodiment to obtain an equal number of points is specifically as follows:

[0071] This embodiment analyzes the segmented insulator data characteristics and the collected point cloud data, selecting a fixed-size rectangular window and downsampling the point cloud data of that area. Specifically, the rectangular window is divided into smaller windows of equal size, and one point is taken from each window. The average value is then used to represent the total points in each smaller window. Figure 4 As shown.

[0072] Step 3: Convert the preprocessed point cloud data into a 3D matrix.

[0073] In this embodiment, the coordinates of the data points are in the form of (X, Y, Z). Three point coordinates are obtained with the X, Y, and Z axes as dimensions. After downsampling, the point data in space is transformed into three point coordinates in a Cartesian coordinate system. Following this method, the following is obtained: Figure 5 The diagram shows a three-dimensional matrix. In the diagram, the X,Y matrix represents the coordinates of the points obtained after transforming the X,Y coordinates of the point cloud data; the X,Z matrix represents the coordinates of the points obtained after transforming the X,Z coordinates of the point cloud data; and the Y,Z matrix represents the coordinates of the points obtained after transforming the Y,Z coordinates of the point cloud data. Finally, the three matrices are concatenated to obtain a W×H×3 three-dimensional matrix (W and H represent the width and height of the matrix, respectively).

[0074] Step 4: Use the 3D matrix as a training set and input it into the convolutional neural network for model training.

[0075] In this embodiment, the obtained three-dimensional matrix of size W×H×3 is normalized to map the data uniformly to the [0,1] interval, thereby accelerating the convergence speed of the model in the later stages. The normalization formula is as follows:

[0076]

[0077] Where a, b represent the normalization interval [a, b], and in this embodiment, the data is normalized to between 0 and 1; Max, Min represent the maximum and minimum values ​​in the matrix, respectively; X1 represents the value of the point to be normalized.

[0078] The normalized dataset was divided into a training set (90%) and a test set (10%).

[0079] This embodiment uses an improved ResNet-18 convolutional neural network model for training, and the network structure is as follows: Figure 7 As shown, the ResNet-18 model introduces a residual block structure to prevent the gradient vanishing problem during neural network training. This embodiment uses the ResNet-18 model, adding a feature extractor before the input network. The feature extractor extracts features using an input size of 2409×3209. Figure 6 As shown, convolution and pooling operations are mainly used to compress and downsample the image, reducing excessive pixel loss during data compression. Simultaneously, the normalization operation in step 4 adds a batch normalization (BN) layer before the output of each layer in the ResNet-18 model, normalizing the data to the [0,1] interval, and adding the ReLU activation function, as shown... Figure 7As shown in the B / R layer, this prevents changes in data distribution caused by increased network depth and accelerates model convergence. Finally, the fully connected (FC) layers are replaced with 1×1 convolutional layers, so that the output image size is not limited by the input image size (fully connected layers require a fixed input size to reconstruct the convolutional image information) and the model parameters are reduced. The block structure is as follows: Figure 8 As shown.

[0080] In this embodiment, the above convolutional neural network model is trained using a training set. During the training process, the number of convolutional kernels is continuously adjusted, and the feature maps of each layer are upgraded and reduced in dimensionality to learn the low-level, middle-level, and high-level features of the data. Finally, the input data is restored through a 1×1 convolutional layer and classified using the softmax function.

[0081] Step 5: Test the trained convolutional neural network model. If the test results meet the requirements, the segmented insulator recognition model is obtained; otherwise, return to step 3.

[0082] In this embodiment, after the model training is completed, the trained model is tested using a test set, and the model performance is evaluated using accuracy and recall. The formulas for calculating accuracy and recall are as follows:

[0083]

[0084]

[0085] Wherein, TP represents the number of correctly identified segmented insulators, FP represents the number of other components incorrectly identified as segmented insulators, FN represents the number of segmented insulators incorrectly identified as other components, and TN represents the number of correctly identified non-segmented insulators; TP, FP, FN, and TN are calculated using a confusion matrix, as shown in Table 1:

[0086] Table 1. Confusion Matrix for Segmented Insulator Classification

[0087]

[0088] The need for further model training is determined based on accuracy and recall (the proportion of correctly identified positive segmented insulators out of all predicted segmented insulators).

[0089] Step 6: Use the segmented insulator identification model to identify the segmented insulators.

[0090] This embodiment also proposes a computer device for performing the methods described above in this embodiment.

[0091] Specifically, such as Figure 9As shown, a computer device includes a processor, internal memory, and a system bus; various device components, including the internal memory and processor, are connected to the system bus. The processor is hardware used to execute computer program instructions through basic arithmetic and logical operations within the computer system. Internal memory is a physical device used for temporary or permanent storage of computational programs or data (e.g., program state information). The system bus can be any of several types of bus architectures, including a memory bus or memory controller, a peripheral bus, and a local bus. The processor and internal memory can communicate via the system bus. Internal memory includes read-only memory (ROM) or flash memory (not shown in the figure), and random access memory (RAM), which typically refers to the main memory loaded with the operating system and computer programs.

[0092] Computer devices typically include an external storage device. The external storage device can be selected from a variety of computer-readable media, which are any usable media accessible by a computer device, including both removable and fixed media. Examples of computer-readable media include, but are not limited to, flash memory (microSD cards), CD-ROMs, digital versatile optical discs (DVDs) or other optical disc storage, magnetic tape cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other media that can be used to store desired information and is accessible by a computer device.

[0093] Computer devices can logically connect to one or more network terminals in a network environment. Network terminals can be personal computers, servers, routers, smartphones, tablets, or other public network nodes. Computer devices connect to network terminals through network interfaces (LAN interfaces). A Local Area Network (LAN) is a computer network interconnected within a limited area, such as a home, school, computer lab, or office building using network media. WiFi and twisted-pair Ethernet are the two most commonly used technologies for building LANs.

[0094] It should be noted that other computer systems, including more or fewer subsystems than computer equipment, are also applicable to the invention.

[0095] As described in detail above, the computer device applicable to this embodiment can perform specified operations of the contact wire segment insulator identification method based on 3D imaging technology. The computer device executes these operations through software instructions run by a processor on a computer-readable medium. These software instructions can be read into memory from a storage device or from another device via a local area network interface. The software instructions stored in memory cause the processor to execute the aforementioned group membership information processing method. Furthermore, the present invention can also be implemented through hardware circuitry or hardware circuitry combined with software instructions. Therefore, implementing this embodiment is not limited to any specific combination of hardware circuitry and software.

[0096] Example 2

[0097] This embodiment proposes a contact wire segment insulator identification device based on 3D imaging technology, such as... Figure 10 As shown, it includes a data acquisition module 10, a data preprocessing module 20, a data conversion module 30, a model training module 40, and a model testing module 50.

[0098] Among them, the data acquisition module 10 is used to acquire point cloud data of the contact wire segment insulator.

[0099] The data preprocessing module 20 preprocesses the collected point cloud data.

[0100] The data conversion module 30 converts the preprocessed point cloud data into a three-dimensional matrix.

[0101] The model training module 40 uses the three-dimensional matrix as a training set and inputs it into the convolutional neural network for model training.

[0102] The model testing module 50 is used to test the trained convolutional neural network model. If the test is successful, the model is output for segmented insulator identification; otherwise, training continues.

[0103] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying segmented insulators of overhead contact lines based on 3D imaging technology, characterized in that, include: Collect point cloud data of the contact wire segment insulators; Preprocess the collected point cloud data; Convert the preprocessed point cloud data into a 3D matrix; The three-dimensional matrix is ​​used as a training set and input into the convolutional neural network for model training. The trained convolutional neural network model is tested. If the test results meet the requirements, the model is output for segmented insulator identification; otherwise, training continues. The specific steps for converting the preprocessed point cloud data into a 3D matrix include: The data is divided along the X, Y, and Z axes into three matrices: X×Y×1, Y×Z×1, and X×Z×1. X,Y represent the coordinates of a point obtained after transforming the point cloud data using the X and Y coordinates; X,Z represent the coordinates of a point obtained after transforming the point cloud data using the X and Z coordinates; and Y,Z represent the coordinates of a point obtained after transforming the point cloud data using the Y and Z coordinates. The transformation refers to converting the point cloud data in space into the coordinates of three points in a Cartesian coordinate system. By concatenating the three matrices according to their dimensions, a three-dimensional matrix of size W×H×3 is obtained; where W and H represent the width and height of the three-dimensional matrix, respectively.

2. The contact wire segment insulator identification method based on 3D imaging technology according to claim 1, characterized in that, The preprocessing process specifically includes: The collected point cloud data is denoised. Downsampling is performed on the denoised point cloud data to ensure consistency in the number of data points.

3. The contact wire segment insulator identification method based on 3D imaging technology according to claim 2, characterized in that, The denoising process specifically employs the KNN algorithm for clustering to remove noise points; The downsampling process specifically involves: selecting a rectangular window of a preset size, dividing the rectangular window into smaller windows of the same size, taking one point from each window, and replacing the total number of points in the smaller window with the average value.

4. The contact wire segment insulator identification method based on 3D imaging technology according to claim 1, characterized in that, The model training process specifically includes: The three-dimensional matrix is ​​normalized to map the data uniformly to the [0,1] interval; The dataset obtained after normalization is divided into a training set and a test set; The ResNet-18 convolutional neural network model was trained using the training set.

5. The contact wire segment insulator identification method based on 3D imaging technology according to claim 4, characterized in that, The testing process specifically includes: Input the test set into the trained model; The confusion matrix is ​​obtained based on the model's classification output. Precision and recall are calculated based on the confusion matrix. Determine whether the model needs to continue training based on accuracy and recall.

6. The contact wire segment insulator identification method based on 3D imaging technology according to claim 5, characterized in that, The accuracy The calculation formula is: ； The recall rate The calculation formula is: ； Wherein, TP represents the number of correctly identified segmented insulators, FP represents the number of other components incorrectly identified as segmented insulators, FN represents the number of segmented insulators incorrectly identified as other components, and TN represents the number of correctly identified non-segmented insulators.

7. A contact wire segment insulator identification device based on 3D imaging technology, characterized in that, It includes a data acquisition module, a data preprocessing module, a data conversion module, a model training module, and a model testing module; The data acquisition module is used to acquire point cloud data of the contact wire segment insulator; The data preprocessing module preprocesses the collected point cloud data; The data conversion module converts the preprocessed point cloud data into a three-dimensional matrix. The model training module uses a three-dimensional matrix as a training set and inputs it into a convolutional neural network for model training. The model testing module is used to test the convolutional neural network model after training. If the test is qualified, the model is output for segmented insulator identification; otherwise, training continues. The specific steps for converting preprocessed point cloud data into a 3D matrix include: The data is divided along the X, Y, and Z axes into three matrices: X×Y×1, Y×Z×1, and X×Z×1. X,Y represent the coordinates of a point obtained after transforming the point cloud data using the X and Y coordinates; X,Z represent the coordinates of a point obtained after transforming the point cloud data using the X and Z coordinates; and Y,Z represent the coordinates of a point obtained after transforming the point cloud data using the Y and Z coordinates. The transformation refers to converting the point cloud data in space into the coordinates of three points in a Cartesian coordinate system. By concatenating the three matrices according to their dimensions, a three-dimensional matrix of size W×H×3 is obtained; where W and H represent the width and height of the three-dimensional matrix, respectively.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images