Railway turnout identification method, device, equipment and medium

Abstract

The invention provides a railway turnout identification method and device, equipment and a medium, and the method comprises the steps: S1, constructing a classification network, obtaining training data, carrying out the customization training of the classification network, and obtaining a trained railway turnout classification model; s2, obtaining turnout rail shape three-dimensional visual point cloud data in real time, and processing the turnout rail shape three-dimensional visual point cloud data through the railway turnout classification model to obtain turnout rail shape classification data; s3, adding the turnout rail shape classification data to a preset sliding window queue, and enabling the preset sliding window queue to keep a preset length; and S4, performing statistical analysis on the sliding window queue to obtain a railway turnout classification result. According to the technical scheme provided by the invention, the accuracy of visual turnout identification in the railway industry can be improved by comprehensively considering the turnout identification algorithm of the left and right rail states in the time window.

Description

Technical Field

[0001] This document relates to the technical field of railway turnout identification, and particularly to a railway turnout identification method, device, equipment and medium. Background Technique

[0002] With the increase in the speed of high-speed railways and the density of EMU train operations, the line equipment is operating at full load, and "planned maintenance" is difficult to meet the demand, so "condition-based maintenance" has emerged. The detection of track cross-section defects involves a detailed inspection of the condition of railway tracks, aiming to ensure the normal operation and excellent condition of the tracks. Whether the tracks are in good condition is directly related to the safety, comfort and stability of train operation. If the turnout information can be accurately distinguished from the detection data provided by the track cross-section defect detection vehicle and the data of ordinary steel rails, targeted quality assessments can be carried out on the turnouts and ordinary steel rails, and accurate evaluations can be made according to their respective different assessment criteria. At present, there is little research on turnout identification technology. Existing research focuses on the fields of damage detection, condition monitoring and fault diagnosis. Some research uses thermal imagers and infrared image processing technology and the "triangulation method" to identify the turnout direction, and there is also research on the online turnout identification method based on structured light technology.

[0003] The problems faced by the current research on turnouts include: First, structured light and infrared light have significant deficiencies compared to laser. The main defects of structured light are that it is easily affected by ambient light, has limited depth resolution and lacks sensitivity to transparent objects; while the limitations of infrared light are reflected in its dependence on thermal radiation, limited penetration and susceptibility to temperature and humidity changes. Second, the accuracy of the existing neural network classification model is not high enough. Finally, the existence of normal turnouts in the left and right tracks and turnouts has not been fully considered. These problems will undoubtedly lead to a decrease in the turnout identification accuracy.

[0004] Therefore, how to comprehensively consider the turnout identification algorithm with time window and the states of left and right tracks has become a very challenging problem. Summary of the Invention

[0005] The present invention provides a railway turnout identification method, device, equipment and medium, aiming to solve the above problems.

[0006] According to an embodiment of the present invention, a railway turnout identification method is provided, including:

[0007] S1. Construct a classification network, obtain training data, customize the training of the classification network, and obtain a trained railway turnout classification model;

[0008] S2. Real-time obtain the three-dimensional visual point cloud data of the turnout rail shape, and process the three-dimensional visual point cloud data of the turnout rail shape through the railway turnout classification model to obtain turnout rail shape classification data;

[0009] S3. Add the turnout rail shape classification data to a preset sliding window queue to keep the preset sliding window queue at a preset length;

[0010] S4. Obtain the railway turnout classification result by statistically analyzing the sliding window queue.

[0011] According to an embodiment of the present invention, there is provided a railway turnout recognition device, including:

[0012] A classification model construction module, configured to construct a classification network, obtain training data, customize the training of the classification network, and obtain a trained railway turnout classification model;

[0013] A classification data acquisition module, configured to acquire three-dimensional visual point cloud data of the turnout rail shape in real time, and process the three-dimensional visual point cloud data of the turnout rail shape through the railway turnout classification model to obtain turnout rail shape classification data;

[0014] A sliding window module, configured to add the turnout rail shape classification data to a preset sliding window queue to keep the preset sliding window queue at a preset length;

[0015] A result acquisition module, which obtains the railway turnout classification result by statistically analyzing the sliding window queue.

[0016] According to an embodiment of the present invention, there is provided an electronic device, including:

[0017] A processor; and,

[0018] A memory arranged to store computer-executable instructions, and when the computer-executable instructions are executed, the processor is caused to execute the steps of the above-mentioned railway turnout recognition method.

[0019] According to an embodiment of the present invention, there is provided a storage medium for storing computer-executable instructions, and when the computer-executable instructions are executed, the steps of the above-mentioned railway turnout recognition method are implemented.

[0020] By adopting the embodiments of the present invention, it is possible to accurately classify the normal profiles and suspected turnout profiles of the single-track three-dimensional vision point cloud profiles. With the help of the constructed classification network and the exported onnx model file, real-time classification and judgment of the collected three-dimensional vision point cloud data of the track profiles are realized, ensuring that the recognition of the turnout rail shape can be carried out under the requirements of real-time performance. The sliding window queue and statistical analysis method are introduced to reveal the rules through big data statistics. When a certain proportion of turnout profiles appear in the sliding window, it can be accurately determined that the sliding window is a turnout, thereby improving the stability and reliability of the judgment. By the method of backward deduction of a window length, the lag of the sliding window judgment is effectively addressed, and then the starting point of the turnout can be more accurately determined, improving the accuracy of the judgment. Generally speaking, the present invention achieves positive effects of high accuracy, real-time performance and reliability in the classification of turnout rail shapes, providing an effective technical means for track detection and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in one or more embodiments of this specification 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 drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a flowchart of the railway turnout recognition method according to the embodiments of the present invention;

[0023] Figure 2 It is a schematic diagram of the railway turnout recognition device according to the embodiments of the present invention;

[0024] Figure 3 It is a schematic diagram of the scenario of the railway turnout recognition method according to the embodiments of the present invention;

[0025] Figure 4 It is a specific implementation flowchart of the railway turnout recognition method according to the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will clearly and completely describe the technical solutions in one or more embodiments of this specification in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only some embodiments of this specification, rather than all embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this document.

[0027] Method Embodiment

[0028] According to an embodiment of the present invention, a method for identifying railway turnouts is provided. Figure 1 The following is a flowchart of the method for identifying railway turnouts according to an embodiment of the present invention. According to Figure 1 As shown, the method for identifying railway turnouts according to an embodiment of the present invention includes:

[0029] S1. Construct a classification network, obtain training data, customize the training of the classification network, and obtain a trained railway turnout classification model;

[0030] In an embodiment of the present invention, with the help of Roboflow intelligent service, a network for classification is constructed, which can distinguish the three-dimensional visual point cloud profile of a single track and identify the normal profile and the suspected turnout profile. Through the customized training and inference process, a classification model file with an accuracy rate exceeding 99% is successfully obtained.

[0031] With the help of the Roboflow intelligent service tool, export the required ONNX model file for subsequent inference classification work.

[0032] First, convert the three-dimensional visual point cloud data from the track profile into 200*200 image data, and then obtain the classification model according to the following steps:

[0033] ① Create a Roboflow account and Roboflow service;

[0034] ② Create a new project: Log in to the Roboflow portal (Roboflow);

[0035] ③ Create a new project and select "Image Classification" as the project type;

[0036] ④ Add images: In the new project, upload the dataset containing the images you want to train. Ensure that the dataset has clear categories;

[0037] ⑤ Label the images: After the images are uploaded, label the images, that is, assign the correct category to each image;

[0038] ⑥ Train the model: Select the appropriate training option (quick training or advanced training) and start training the model;

[0039] ⑦ Test the model: After training is completed, use the test images to evaluate the performance of the model;

[0040] ⑧ Publish the model: If you are satisfied with the model performance, publish the model to obtain the API key and endpoint.

[0041] Obtain the ONNX file according to the following steps:

[0042] ① Log in to the Roboflow portal;

[0043] ② Select project: Select the project that contains the trained model;

[0044] ③ Select model: In the project, select the trained model you want to export;

[0045] ④ Export as ONNX: In the upper right corner of the model page, you should be able to find the "Export" option, select Export, and choose the ONNX format;

[0046] ⑤ Set export options: Provide the name of the exported model and select the version of the exported model. You may also need to select the type of the exported model (e.g., export the transfer learning model) and other options, depending on the updates of Roboflow.

[0047] ⑥ Start export: Click Start Export, and Roboflow will generate an ONNX file containing the model parameters.

[0048] ⑦ Download the model: After the export is completed, you should be able to download the ONNX file and save it locally.

[0049] S2. Obtain the 3D visual point cloud data of the switch rail shape in real time, and process the 3D visual point cloud data of the switch rail shape through the railway switch classification model to obtain the switch rail shape classification data;

[0050] Real-time processing of 3D visual point cloud data: Design to call the ONNX model for inference and classification, real-time process the collected 3D visual point cloud data of the track profile, classify and judge the switch rail shape, perform rail shape classification on the 3D visual point cloud data on both the left and right sides, and if there is a switch on either side, determine that this profile is a switch profile.

[0051] In a specific implementation, S2 specifically includes:

[0052] ① 3D visual point cloud acquisition: Select or build an appropriate 3D visual point cloud acquisition device, such as a lidar. Real-time collect the 3D visual point cloud data of the track profile;

[0053] ② Data preprocessing: Preprocess the collected 3D visual point cloud data to meet the input requirements of the model. This may include operations such as 3D visual point cloud downsampling, denoising, and coordinate transformation;

[0054] ③ Model inference: Use ONNX Runtime or other frameworks that support the ONNX format to load and run the Roboflow model you exported. Input the preprocessed 3D visual point cloud data into the model to obtain the classification result;

[0055] ④ Classification result processing: Based on the classification results output by the model, determine the type of turnout rail shape. Perform possible post-processing, such as filtering out results with low confidence or smoothing the output using methods like sliding windows.

[0056] ⑤ Result display: Display the classification results on the user interface, or record and notify in other forms.

[0057] The 3D visual point cloud data and classification results can be superimposed on the track diagram for visual analysis.

[0058] S3. Add the turnout rail shape classification data to a preset sliding window queue to keep the preset sliding window queue at a preset length.

[0059] Establish a sliding window queue, and the rail shape classification results of each profile are added to the queue in real time. When the queue length reaches the statistically obtained length, maintain the queue length.

[0060] The sliding window queue is used to store the rail shape classification results of each track profile in real time. When a new rail shape classification result arrives, it is added to the end of the queue, and the queue length is ensured to be within the statistically obtained length. If the queue is full, adding a new result will cause the result at the head of the queue to be removed to keep the queue length unchanged.

[0061] S4. Obtain the railway turnout classification results through statistical analysis of the sliding window queue.

[0062] Perform statistical analysis on each queue, and obtain rules through big data-based statistical analysis. For example, when a certain proportion of turnout profiles are in the sliding window, the sliding window is identified as a turnout.

[0063] Figure 3 This is a schematic diagram of the scenario of the railway turnout recognition method according to the embodiment of the present invention. The steps to realize statistical analysis to determine whether the sliding window contains a certain proportion of turnout profiles are as follows:

[0064] ① Define a statistical analysis function: Create a function that accepts a list of rail shape classification results in the sliding window and calculates the proportion of turnout profiles among them.

[0065] ② Set a threshold: Set a threshold to determine when to consider the sliding window as a turnout. This threshold can be adjusted according to actual needs.

[0066] ③ Invoke statistical analysis: When adding a new rail shape classification result to the queue, invoke the statistical analysis function, pass in the results of the current sliding window, and obtain the proportion of turnout profiles.

[0067] ④ Judgment and processing: If the turnout profile ratio obtained by statistics exceeds the set threshold, it is determined that the sliding window is a turnout and corresponding processing is performed.

[0068] Due to the lag in the sliding window judgment, at the moment when the sliding window is judged as a turnout, a window length is reversed to determine the starting point of the turnout. In this way, the position of the turnout can be more accurately located.

[0069] Figure 4 This is the specific implementation flowchart of the railway turnout recognition method according to the embodiment of the present invention. According to Figure 4 As shown, a specific implementation step of the railway turnout recognition method according to the embodiment of the present invention includes:

[0070] Step 1: Establish a classification network: Use the Roboflow intelligent service to establish a classification network to classify the normal profile and suspected turnout profile of the single-track three-dimensional visual point cloud profile. Through personalized training and inference, a classification model file with an accuracy of more than 99% is obtained.

[0071] Step 2: Export the model file: Use the Roboflow intelligent service tool to export the required onnx model file for subsequent inference classification.

[0072] Step 3: Real-time process the three-dimensional visual point cloud data: Design to call the onnx model for inference classification, and real-time process the collected three-dimensional visual point cloud data of the track profile to classify and judge the turnout rail shape.

[0073] Step 4: Classification of left and right rail shapes: Classify the three-dimensional visual point cloud data of the left and right rails. As long as either the left or right side is a turnout, the profile is regarded as a turnout profile.

[0074] Step 5: Sliding window queue: Establish a sliding window queue, and the rail shape classification results of each profile are added to the queue in real time. When the queue length reaches the statistically obtained length, the queue length is maintained.

[0075] Step 6: Statistical analysis: Conduct statistical analysis on each queue, and obtain rules through big data-based statistical analysis. For example, when there is a certain proportion of turnout profiles in the sliding window, it is determined that the sliding window is a turnout.

[0076] Step 7: Lag processing: Due to the lag in the sliding window judgment, at the moment when the sliding window is judged as a turnout, a window length is reversed to determine the starting point of the turnout.

[0077] By adopting the embodiment of the present invention, the following beneficial effects are achieved:

[0078] It can accurately classify the normal section and suspected turnout section of the single-track three-dimensional visual point cloud profile. With the constructed classification network and the exported onnx model file, it realizes the real-time classification and judgment of the collected three-dimensional visual point cloud data of the track profile, ensuring that the recognition of the turnout rail shape can be carried out under the real-time requirements. The sliding window queue and statistical analysis method are introduced to reveal the law through big data statistics. When a certain proportion of turnout sections appear in the sliding window, the sliding window can be accurately determined as a turnout, thereby improving the stability and reliability of the judgment. By the method of backward deduction of a window length, the lag of the sliding window judgment is effectively dealt with, and then the starting point of the turnout can be determined more accurately, improving the accuracy of the judgment. Generally speaking, the present invention achieves the positive effects of high accuracy, real-time performance and reliability in the classification of turnout rail shapes, providing an effective technical means for track detection and maintenance.

[0079] Device Embodiment 1

[0080] According to an embodiment of the present invention, there is provided a railway turnout recognition device. Figure 2 It is a schematic diagram of the railway turnout recognition device according to the embodiment of the present invention. According to Figure 2 As shown, the railway turnout recognition device according to the embodiment of the present invention includes:

[0081] A classification model construction module 20, configured to construct a classification network, obtain training data, customize the training of the classification network, and obtain a trained railway turnout classification model;

[0082] A classification data acquisition module 22, configured to acquire three-dimensional visual point cloud data of the turnout rail shape in real time, and process the three-dimensional visual point cloud data of the turnout rail shape through the railway turnout classification model to obtain turnout rail shape classification data;

[0083] A sliding window module 24, configured to add the turnout rail shape classification data to a preset sliding window queue, so that the preset sliding window queue maintains a preset length;

[0084] A result acquisition module 26, configured to obtain a railway turnout classification result by performing statistical analysis on the sliding window queue.

[0085] Device Embodiment 2

[0086] According to an embodiment of the present invention, there is provided an electronic device, including:

[0087] A processor; and,

[0088] A memory arranged to store computer-executable instructions, which when executed cause the processor to execute the steps of the above-mentioned railway turnout recognition method.

[0089] Device Embodiment 3

[0090] According to an embodiment of the present invention, there is provided a storage medium for storing computer-executable instructions, and the computer-executable instructions, when executed, implement the steps of the above-mentioned railway switch identification method.

[0091] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A railway turnout identification method, characterized in that: The following steps are involved: S1. Build a classification network, obtain training data, perform customized training on the classification network, and obtain a trained railway turnout classification model; S2, acquiring three-dimensional visual point cloud data of the turnout track shape in real time, and processing the three-dimensional visual point cloud data of the turnout track shape through the railway turnout classification model to obtain turnout track shape classification data; S3, adding the turnout track shape classification data to a preset sliding window queue, so that the preset sliding window queue maintains a preset length; S4. Obtain railway turnout classification results by performing statistical analysis on the sliding window queue.

2. The railway turnout identification method according to claim 1, characterized in that: The step of obtaining the training data comprises: Data acquisition equipment preparation, including: selecting 3D visual point cloud acquisition equipment, calibrating and debugging the acquisition equipment; Determine the collection scenario, including: making the collection scenario cover straight track sections, curved track sections and track areas with different slopes, taking into account different weather conditions and railway operation periods; Real-time collection of track profile 3D visual point cloud data, specifically including: continuous collection along the railway track at a predetermined sampling interval to ensure the continuity and representativeness of the collected data, and numbering and recording each track profile point cloud data collected; Data preprocessing, specifically including: downsampling, denoising and coordinate transformation of the collected 3D visual point cloud data; Data labeling, specifically including: converting the pre-processed 3D visual point cloud data into 200*200 image data, judging whether it is a normal section or a suspected turnout section based on the actual morphological characteristics of the track section, and assigning it a corresponding category label; Data set division specifically includes: dividing the labeled data set into training set, validation set and test set according to a certain ratio.

3. The method according to claim 1, characterized in that The sliding window queue adds the switch track classification data to the end of the queue each time new switch track classification data arrives, and ensures that the queue length remains at a preset length. If the sliding window queue is full, the switch track classification data at the head of the sliding window queue is removed and the new switch track classification data is added to the end of the queue.

4. The method according to claim 1, characterized in that The S4 specifically includes: Create a statistical analysis function that accepts a list of track shape classification results in a sliding window queue and calculates the proportion of turnout profiles in it; If the ratio of the turnout section is greater than a preset threshold, the sliding window is determined to be a turnout.

5. The method according to claim 1, characterized in that When there is a lag in the judgment of the sliding window queue, when the sliding window queue is judged as a turnout, a window length is pushed forward to determine the starting point of the turnout.

6. The railway turnout identification method according to claim 1, characterized in that: The railway turnout classification model is constructed using a deep learning framework. During the training process, a suitable loss function is selected to measure the difference between the model prediction results and the true labels. At the same time, a suitable optimization algorithm is used to update the model parameters to improve the accuracy and generalization ability of the model.

7. The railway turnout identification method according to claim 1, characterized in that: After the three-dimensional visual point cloud data of the turnout track shape is processed by the railway turnout classification model to obtain the turnout track shape classification data, the classification data is evaluated for credibility, a credibility threshold is set, and the classification data with credibility lower than the threshold is filtered out, and the remaining turnout track shape classification data is added to a preset sliding window queue to ensure the reliability of subsequent statistical analysis results.

8. A railway turnout identification device, comprising: A classification model building module is used to build a classification network, obtain training data to perform customized training on the classification network, and obtain a trained railway turnout classification model; A classification data acquisition module is used to acquire three-dimensional visual point cloud data of the turnout track shape in real time, and to process the three-dimensional visual point cloud data of the turnout track shape through the railway turnout classification model to acquire the turnout track shape classification data; A sliding window module, used for adding the turnout track shape classification data to a preset sliding window queue so that the preset sliding window queue maintains a preset length; The result acquisition module obtains the railway turnout classification results by performing statistical analysis on the sliding window queue.

9. An electronic device, comprising: processor; as well as, A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps of the railway switch identification method according to any one of claims 1 to 7.

10. A storage medium for storing computer executable instructions, wherein the computer executable instructions, when executed, implement the steps of the railway turnout identification method according to any one of claims 1 to 7.

Images