Railway roadbed disease detection method and device based on deep learning

Abstract

The invention relates to a railway roadbed disease detection method and device based on deep learning. The method comprises the following steps: acquiring a detection radar image through a vehicle-mounted roadbed detection radar system; preprocessing the detection radar image to obtain a preprocessed image; inputting the preprocessed image into a pre-trained railroad bed disease detection model for disease detection to obtain a railroad bed disease detection result; wherein the railway roadbed disease detection result comprises a classification result used for representing waveform types and a regression result used for representing vertical positions of abnormal waveforms, and the waveform types comprise normal waveforms and abnormal waveforms; compared with the prior art, the railway roadbed disease detection method has the advantage that the railway roadbed disease detection efficiency and accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of railway track detection technology, and in particular to a railway roadbed disease detection method and device based on deep learning. Background Art

[0002] Railway subgrade is a critical component of the railway system, essentially providing stable support for the tracks (e.g., ballasted track and slab track) to ensure safe train operation. However, with the increase in high-intensity and high-speed train loads, axle loads, and traffic volume, the subgrade has experienced unacceptably rapid degradation, leading to frequent maintenance activities. Railway subgrade diseases (also known as defects, diseases, or anomalies) are dangerous because they lead to rapid degradation of the track (e.g., irregular track geometry, uneven ballast settlement, and sleeper damage), affecting ride comfort and safe train operation. In some cases, subgrade problems may even pose a risk of train derailment.

[0003] Furthermore, Ground Penetrating Radar (GPR) is a non-destructive technology that has been used for non-destructive testing and assessment of railway subgrade conditions. However, most existing GPR data processing and interpretation methods rely on time-consuming manual interpretation, and there is limited research on the application of machine learning methods. Furthermore, GPR data are complex, high-dimensional, and redundant, especially with non-negligible noise. Traditional machine learning methods are ineffective when applied to GPR data processing and interpretation. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a railway roadbed disease detection method and device based on deep learning, which solves the technical problem of time-consuming detection in the prior art.

[0006] (II) Technical solution

[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, an embodiment of the present invention provides a railway roadbed disease detection method based on deep learning, comprising: collecting detection radar images through a vehicle-mounted roadbed detection radar system; preprocessing the detection radar images to obtain preprocessed images; inputting the preprocessed images into a pre-trained railway roadbed disease detection model to perform disease detection to obtain railway roadbed disease detection results; wherein the railway roadbed disease detection results include classification results for representing waveform types and regression results for representing upper and lower positions of abnormal waveforms, and the waveform types include normal waveforms and abnormal waveforms; post-processing the railway roadbed disease detection results to determine the abnormal position.

[0009] In a possible embodiment, the preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

[0010] In a possible embodiment, the railway roadbed disease detection model includes a CNN layer, and the CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence.

[0011] In a possible embodiment, the railway roadbed disease detection model further includes an RNN layer connected to the CNN layer, and the RNN layer includes a plurality of GRU layers connected in sequence and an output layer connected to the last GRU layer.

[0012] In the second aspect, an embodiment of the present invention provides a railway roadbed disease detection device based on deep learning, including: an acquisition module, used to acquire detection radar images through a vehicle-mounted roadbed detection radar system; a preprocessing module, used to preprocess the detection radar images to obtain preprocessed images; a disease detection module, used to input the preprocessed images into a pre-trained railway roadbed disease detection model to perform disease detection, and obtain railway roadbed disease detection results; wherein the railway roadbed disease detection results include classification results for representing waveform types and regression results for representing upper and lower positions of abnormal waveforms, and the waveform types include normal waveforms and abnormal waveforms; a post-processing module, used to post-process the railway roadbed disease detection results to determine the abnormal position.

[0013] In a possible embodiment, the preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

[0014] In a possible embodiment, the railway roadbed disease detection model includes a CNN layer, and the CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence.

[0015] In a possible embodiment, the railway roadbed disease detection model further includes an RNN layer connected to the CNN layer, and the RNN layer includes a plurality of GRU layers connected in sequence and an output layer connected to the last GRU layer.

[0016] In a third aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, and the computer program is executed by a processor to execute the method described in the first aspect or any optional implementation of the first aspect.

[0017] In a fourth aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the method described in the first aspect or any optional implementation method of the first aspect is performed.

[0018] In a fifth aspect, the present application provides a computer program product, which, when running on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0019] (III) Beneficial effects

[0020] The beneficial effects of the present invention are:

[0021] The embodiment of the present application proposes a railway roadbed disease detection method and device based on deep learning, which collects detection radar images through a vehicle-mounted roadbed detection radar system, and preprocesses the detection radar images to obtain preprocessed images, and inputs the preprocessed images into a pre-trained railway roadbed disease detection model to perform disease detection to obtain railway roadbed disease detection results, and post-processes the railway roadbed disease detection results to determine the abnormal location. Compared with the existing technology, it can improve the efficiency and accuracy of railway roadbed disease detection.

[0022] In order to make the above-mentioned objectives, features and advantages to be achieved by the embodiments of the present application more obvious and understandable, the following specifically cites preferred embodiments and describes them in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 A flowchart of a railway subgrade disease detection method based on deep learning provided in an embodiment of the present application is shown;

[0025] Figure 2 A schematic diagram of the model structure of a railway subgrade disease detection model provided in an embodiment of the present application is shown;

[0026] Figure 3 A structural block diagram of a railway subgrade disease detection device based on deep learning provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0028] In order to solve the technical problem of time-consuming detection in the prior art, the embodiment of the present application proposes a railway roadbed disease detection method and device based on deep learning, which collects detection radar images through a vehicle-mounted roadbed detection radar system, and preprocesses the detection radar images to obtain preprocessed images, and inputs the preprocessed images into a pre-trained railway roadbed disease detection model to perform disease detection to obtain railway roadbed disease detection results, and post-processes the railway roadbed disease detection results to determine the abnormal location. Compared with the prior art, it can improve the efficiency and accuracy of railway roadbed disease detection.

[0029] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0030] See also Figure 1 , Figure 1 FIG. 1 is a flow chart of a railway subgrade disease detection method based on deep learning provided in an embodiment of the present application. Figure 1 As shown, the railway subgrade disease detection method can be performed by a railway subgrade disease detection device, and the specific device of the railway subgrade disease detection device can be set according to actual needs, and the embodiments of the present application are not limited thereto. For example, the railway subgrade disease detection device can be a computer or a server. Specifically, the railway subgrade disease detection method includes:

[0031] Step S110, collecting detection radar images through the vehicle-mounted roadbed detection radar system.

[0032] Specifically, the electromagnetic wave energy reflected by the internal structural layers of the railway substructure (for example, it includes the ballast layer, the subballast layer and the roadbed) depends on the difference in dielectric constant between each layer of the medium. As the ground penetrating radar antenna moves along the track, it records the motion characteristics of the reflected signal, including the two-way travel duration, amplitude and phase.

[0033] The vehicle-mounted roadbed detection radar system works by synchronously transmitting and receiving electromagnetic waves. When the antenna moves along the track, it creates a series of scan lines (a-Scans). The scan lines of each measurement point are collected together according to the acquisition interval to generate a detection radar image of the railway substructure, especially the roadbed. The detection radar image intuitively represents the changes in the interface of the railway structure layer and the physical properties of the underground medium, which can be used to identify and extract the necessary information to understand the possible defects in the railway substructure.

[0034] Step S120, preprocessing the detection radar image to obtain a preprocessed image.

[0035] It should be understood that the specific method included in the preprocessing can be set according to actual needs, and the embodiments of the present application are not limited to this.

[0036] Optionally, the preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

[0037] For example, in ground penetrating radar measurements, broadband signals are often used to capture a large number of interference signals at the same time, resulting in more effective reflection wave characteristics. However, to extract roadbed defect information using artificial intelligence models, these interference signals must be eliminated to improve the signal-to-noise ratio of the data, and interference signals may come from three aspects: First, interference signals from the equipment system itself, including antennas, cables, and internal connectors, this interference is inevitable; the second is interference from other signal sources, such as radio and television transmissions, communication signals, etc.; the last is interference from the ground or underground / underground. This type of interference is caused by the direct coupling wave or the transmitting and receiving antennas that form the ground reflection wave. Due to the inhomogeneity of the underground medium and the occurrence of strong scattering, the bypass wave is superimposed on the formation of the interference wave. Therefore, it is very necessary to preprocess the detection radar image.

[0038] And, the specific process of preprocessing the detection radar image in this application includes:

[0039] Discrete sampling processing: Before filtering the original detection radar image, discrete sampling is required, and the sampling process needs to satisfy the sampling theorem in the frequency domain equation; otherwise, the data spectrum will be mixed, resulting in pseudo-frequency;

[0040] Zero line correction processing: Zero line setting is mainly performed by cutting the air layer to a fixed threshold value, which is set at the most stable point on the electromagnetic wave trajectory a scan. And, according to the type and center frequency of the antenna, setting an appropriate threshold position along the A scan reflects the accuracy of the result, which can be summarized as the initial arrival of the wave, the position of the first trough, the position of the first peak of the zero amplitude value between the first peak and the trough, the position of half the amplitude value between the first peak and the trough, and the position of the first peak;

[0041] Gain setting processing: Automatic gain control AGC can be used for processing. When the signal is strong, its gain is automatically reduced, and when the signal is weak, the gain is automatically increased. At the same time, it can ensure the uniformity of strong and weak signals and help track effective waves;

[0042] Bandpass filtering: The bandpass filter works by removing the streaking bands from the spectrum of the GPR data, and this FIR bandpass filter consists of two filters, a high-pass filter and a low-pass filter, which modifies the GPR signal by removing the low-frequency and high-frequency components of the spectrum;

[0043] Autoregressive moving average processing: Due to the high background noise of the ground penetrating radar signal, the autoregressive moving average spectrum estimation (ARMA) in modern spectrum estimation is used to analyze and identify the non-smoothed signal of the effective reflection signal. Signal features can be extracted under low signal-to-noise ratio, and the spectrum estimation accuracy is high.

[0044] In addition, after the above preprocessing process, it can suppress the inevitable random interference signals in the ground penetrating radar image and improve the signal-to-noise ratio. In addition, the changes in the double travel time (or depth) and mileage are reflected in the image features of the ground penetrating radar image, and these image features are the reflection wave amplitude, waveform, phase, spectrum and roadbed defects of the interface between the roadbed layers.

[0045] Step S130, input the preprocessed image into a pre-trained railway roadbed disease detection model to perform disease detection, and obtain a railway roadbed disease detection result. The railway roadbed disease detection result includes a classification result for indicating the waveform type and a regression result for indicating the upper and lower positions of the abnormal waveform, and the waveform type includes a normal waveform and an abnormal waveform.

[0046] It should be understood that the model type and specific model structure of the railway subgrade disease detection model can be set according to actual needs, and the embodiments of the present application are not limited to this.

[0047] Optionally, the railway roadbed disease detection model model can be a CRNN network model, and CRNN is a fusion network that combines CNN and RNN to process different GPR features. And, it is used as a highly nonlinear mapping method in a supervised learning model to establish a connection between a target sample and an input. RNN is a neural network with recursive connections that can model sequence data for sequence recognition and prediction. They can also store information using recursive iterative functions, capture contextual information well, and realize transient dependency learning. And, the CRNN network model uses CNN and RNN to process GPR image data, making it an effective tool for detecting roadbed defects.

[0048] Optionally, see Figure 2 , Figure 2 FIG. 1 is a schematic diagram showing a model structure of a railway subgrade disease detection model provided in an embodiment of the present application. Figure 2 As shown, the railway subgrade disease detection model includes a CNN layer and a RNN layer connected in sequence.

[0049] The CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence. Each of the 9 target layers includes a first residual layer and a second convolutional layer connected in sequence.

[0050] It should be understood that the layer structure and parameters of each layer in the CNN layer can be set according to actual needs, and the embodiments of the present application are not limited to this.

[0051] For example, the first convolution layer may be a one-dimensional convolution layer, through which the preprocessed image is convolved. Also, in order to improve the convergence speed of the network, a normalization layer may be added to the first convolution layer. In addition, the convolution kernel of the second convolution layer is 2×2, so that after 9 target layers, a 9F feature vector may be obtained, and F represents the number of filters in the second convolution layer, and the number of filters in each of the 9 second convolution layers is the same.

[0052] And, continue to see Figure 2 , the RNN layer consists of multiple GRU (Gated Recurrent Unit) layers connected in sequence and an output layer connected to the last GRU layer. In other words, the RNN layer can be viewed as a collection of interconnected networks, where the output of one network is fed to the next network in a chain design.

[0053] In addition, the specific number of layers of the GRU layer can be set according to actual needs, and the embodiments of the present application are not limited to this.

[0054] It should be understood that the layer structure and parameters of each layer in the RNN layer can be set according to actual needs, and the embodiments of the present application are not limited to this.

[0055] For example, RNN is a unidirectional RNN, that is, it can process current and previous channel features and produce accurate detection results;

[0056] For another example, the output layer may include a first output layer for outputting classification results and a second output layer for outputting regression results. The first output layer may use a softmax function to constrain the output; the second output layer may use a sigmoid function to constrain the output, thereby limiting the boundaries of the regression output by using an S-shaped function so that the upper and lower positions do not exceed the length of the waveform, thereby enhancing the robustness of the regression output.

[0057] It should be noted here that, since the preprocessed image can be a multi-channel image, its output result can be the classification result corresponding to each channel in the multiple channels and the output result corresponding to each channel.

[0058] Step S140, post-processing the railway subgrade disease detection results to determine the abnormal location.

[0059] It should be understood that the specific means for post-processing the railway subgrade disease detection results can be set according to actual needs, and the embodiments of the present application are not limited thereto.

[0060] Optionally, since the railway subgrade disease detection result can represent the classification result and output result corresponding to each channel, the railway subgrade disease detection result can be post-processed to display the relevant data of all channels in one image, and the abnormal position can be determined through the image. In the target area of ​​the image, the start and end of the waveform type can be used to constrain the left and right boundaries of the abnormality (i.e., the left and right boundaries of the target area), and the upper and lower boundaries of the target area can be determined by the average value of the abnormal type.

[0061] Therefore, with the help of the above-mentioned technical scheme, the embodiment of the present application collects detection radar images through the vehicle-mounted roadbed detection radar system, and preprocesses the detection radar images to obtain preprocessed images, and inputs the preprocessed images into a pre-trained railway roadbed disease detection model to perform disease detection to obtain railway roadbed disease detection results, and post-processes the railway roadbed disease detection results to determine the abnormal location. Compared with the existing technology, it can improve the efficiency and accuracy of railway roadbed disease detection.

[0062] It should be understood that the above-mentioned railway subgrade disease detection method based on deep learning is only exemplary, and those skilled in the art can make various modifications based on the above-mentioned method, and the modified scheme also falls within the protection scope of this application.

[0063] See also Figure 3 , Figure 3 The structural block diagram of a railway subgrade disease detection device 300 based on deep learning provided in an embodiment of the present application is shown. It should be understood that the railway subgrade disease detection device 300 is capable of executing each step in the above method embodiment. The specific functions of the railway subgrade disease detection device 300 can be found in the description above. To avoid repetition, the detailed description is appropriately omitted here. The railway subgrade disease detection device 300 includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the railway subgrade disease detection device 300. Specifically, the railway subgrade disease detection device 300 includes:

[0064] The acquisition module 310 is used to acquire detection radar images through the vehicle-mounted roadbed detection radar system;

[0065] A preprocessing module 320 is used to preprocess the detection radar image to obtain a preprocessed image;

[0066] The defect detection module 330 is used to input the pre-processed image into a pre-trained railway roadbed defect detection model to perform defect detection and obtain a railway roadbed defect detection result; wherein the railway roadbed defect detection result includes a classification result for indicating a waveform type and a regression result for indicating an upper and lower position of an abnormal waveform, and the waveform type includes a normal waveform and an abnormal waveform;

[0067] The post-processing module 340 is used to post-process the railway subgrade disease detection results and determine the abnormal location.

[0068] In a possible embodiment, the preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

[0069] In a possible embodiment, the railway roadbed disease detection model includes a CNN layer, and the CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence.

[0070] In a possible embodiment, the railway roadbed disease detection model further includes an RNN layer connected to the CNN layer, and the RNN layer includes a plurality of GRU layers connected in sequence and an output layer connected to the last GRU layer.

[0071] Since the device described in the above embodiment of the present invention is a device used to implement the method of the above embodiment of the present invention, based on the method described in the above embodiment of the present invention, a person skilled in the art can understand the specific structure and deformation of the device, so it is not described here. All devices used in the method of the above embodiment of the present invention belong to the scope of protection of the present invention.

[0072] The present application provides a storage medium having a computer program stored thereon. The computer program is executed by a processor to execute the method described in the embodiment.

[0073] The present application also provides a computer program product, which, when executed on a computer, enables the computer to execute the method described in the method embodiment.

[0074] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0076] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0077] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0078] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0079] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. A railway subgrade disease detection method based on deep learning, characterized in that: include: Collect detection radar images through the vehicle-mounted roadbed detection radar system; Preprocessing the detection radar image to obtain a preprocessed image; Input the preprocessed image into a pre-trained railway roadbed disease detection model to perform disease detection, and obtain a railway roadbed disease detection result; wherein the railway roadbed disease detection result includes a classification result for indicating a waveform type and a regression result for indicating an upper and lower position of an abnormal waveform, and the waveform type includes a normal waveform and an abnormal waveform; The railway subgrade disease detection results are post-processed to determine the abnormal location.

2. The railway subgrade disease detection method according to claim 1, characterized in that: The preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

3. The railway subgrade disease detection method according to claim 1, characterized in that: The railway subgrade disease detection model includes a CNN layer, and the CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence.

4. The railway subgrade disease detection method according to claim 3, characterized in that: The railway subgrade disease detection model further includes an RNN layer connected to the CNN layer, and the RNN layer includes a plurality of GRU layers connected in sequence and an output layer connected to the last GRU layer.

5. A railway subgrade disease detection device based on deep learning, characterized in that: include: An acquisition module, used for acquiring detection radar images through a vehicle-mounted roadbed detection radar system; A preprocessing module, used for preprocessing the detection radar image to obtain a preprocessed image; A disease detection module, used for inputting the preprocessed image into a pre-trained railway roadbed disease detection model to perform disease detection, and obtain railway roadbed disease detection results; wherein the railway roadbed disease detection results include classification results for indicating waveform types and regression results for indicating upper and lower positions of abnormal waveforms, and the waveform types include normal waveforms and abnormal waveforms; The post-processing module is used to post-process the railway subgrade disease detection results to determine the abnormal location.

6. The railway subgrade disease detection device according to claim 5, characterized in that: The preprocessing includes but is not limited to discrete sampling processing, zero line correction processing, gain setting processing, bandpass filtering processing and autoregressive sliding average processing.

7. The railway subgrade disease detection device according to claim 5, characterized in that: The railway subgrade disease detection model includes a CNN layer, and the CNN layer includes a first convolutional layer, 9 target layers and a second residual layer connected in sequence.

8. The railway subgrade disease detection device according to claim 7, characterized in that: The railway subgrade disease detection model further includes an RNN layer connected to the CNN layer, and the RNN layer includes a plurality of GRU layers connected in sequence and an output layer connected to the last GRU layer.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the railway subgrade disease detection method based on deep learning as described in any one of claims 1 to 4 is executed.

10. An electronic device comprising a processor, a memory and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the railway subgrade disease detection method based on deep learning as described in any one of claims 1-4.

Images