Steel rail surface state detection method and system based on deep learning

Abstract

The invention relates to the technical field of rail surface detection, and relates to a steel rail surface state detection method and system based on deep learning, the method comprises the following steps: first information and second information are obtained, the first information comprises steel rail image information of a complete road section, and the second information comprises steel rail dynamics performance information corresponding to the complete road section; constructing a training set according to the first information, and obtaining a semantic segmentation model of the trained steel rail light band; extracting a light band information data set according to the trained semantic segmentation model of the steel rail light band; and detecting the surface state of the steel rail according to the light band information data set and the second information, and establishing the data mapping relation according to the light band information data set and the second information, so that the dynamic performance is accurately predicted according to the avoidance state of the steel rail.

Description

Technical Field

[0001] The present invention relates to the technical field of track surface detection, and in particular, to a method and system for detecting the surface state of a steel rail based on deep learning. Background Art

[0002] In the prior art, the method for measuring the dynamic performance corresponding to the surface state of a steel rail usually calculates by establishing a steel rail model and a turnout model using the steel rail profile, so as to measure the dynamic performance. However, the method for predicting the dynamic performance based on the surface contact state of the steel rail is still blank. Therefore, there is an urgent need for a method for detecting the surface state of a steel rail that can predict its dynamic performance according to the surface state of the steel rail. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for detecting the surface state of a steel rail based on deep learning to improve the above problems.

[0004] To achieve the above purpose, the embodiments of the present application provide the following technical solutions:

[0005] On the one hand, the embodiments of the present application provide a method for detecting the surface state of a steel rail based on deep learning, and the method includes:

[0006] Obtain first information and second information, where the first information includes the steel rail image information of a complete section, and the second information includes the steel rail dynamic performance information corresponding to the complete section;

[0007] Construct a training set according to the first information to obtain a trained semantic segmentation model of the steel rail light band;

[0008] Extract a light band information data set according to the trained semantic segmentation model of the steel rail light band;

[0009] Detect the surface state of the steel rail according to the light band information data set and the second information.

[0010] On the second hand, the embodiments of the present application provide a system for detecting the surface state of a steel rail based on deep learning, and the system includes:

[0011] An acquisition module, configured to acquire first information and second information, where the first information includes the steel rail image information of a complete section, and the second information includes the steel rail dynamic performance information corresponding to the complete section;

[0012] A first processing module, configured to construct a training set according to the first information to obtain a trained semantic segmentation model of the steel rail light band;

[0013] A second processing module, configured to extract a light band information data set according to the trained semantic segmentation model of the steel rail light band;

[0014] A third processing module, configured to detect the surface state of the rail according to the light band information data set and the second information.

[0015] In a third aspect, an embodiment of the present application provides a device, which includes a memory and a processor. The memory is used to store a computer program; the processor is used to implement the steps of the above-mentioned rail surface state detection method based on deep learning when executing the computer program.

[0016] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned rail surface state detection method based on deep learning are implemented.

[0017] The beneficial effects of the present invention are as follows:

[0018] The present invention constructs a training set through the first information and uses the training set to construct a semantic segmentation model of the trained rail light band to extract the light band information data set from the rail image information, and then establishes a data mapping relationship according to the light band information data set and the second information, so as to realize the accurate prediction of the dynamic performance according to the rail avoidance state.

[0019] Other features and advantages of the present invention will be described in the subsequent description, and part of them will become obvious from the description, or be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the structures specifically pointed out in the written description, claims, and drawings. Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It is a schematic flowchart of the rail surface state detection method based on deep learning described in the embodiments of the present invention.

[0022] Figure 2 It is a schematic structural diagram of the rail surface state detection system described in the embodiments of the present invention.

[0023] Figure 3 It is a schematic structural diagram of the rail surface state detection device described in the embodiments of the present invention.

[0024] Labels in the figure: 901, acquisition module; 902, first processing module; 903, second processing module; 904, third processing module; 9021, first processing unit; 9022, second processing unit; 9023, training unit; 90221, first preprocessing unit; 90222, second preprocessing unit; 9041, third processing unit; 9042, fourth processing unit; 9043, fifth processing unit; 9044, sixth processing unit; 90411, first calculation unit; 90412, second calculation unit; 90413, third calculation unit; 9031, acquisition unit; 9032, seventh processing unit; 9033, eighth processing unit; 9034, ninth processing unit; 9035, tenth processing unit; 9036, eleventh processing unit; 800, rail surface condition detection device based on deep learning; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed implementation manners

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and illustrated herein generally can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive distinction and cannot be understood as indicating or implying relative importance.

[0027] Embodiment 1:

[0028] This embodiment provides a method for detecting the surface condition of a rail based on deep learning. It can be understood that in this embodiment, a scenario can be laid, for example:

[0029] See Figure 1 , the figure shows that this method includes steps S1, S2, S3, and S4.

[0030] Step S1: Obtain the first information and the second information. The first information includes the rail image information of the complete section, and the second information includes the rail dynamic performance information corresponding to the complete section.

[0031] In this step, use an orbital inspection trolley to collect the rail image information of the complete section, and use a dynamic inspection vehicle to collect the rail dynamic performance information corresponding to the complete section.

[0032] Step S2: Construct a training set according to the first information to obtain a trained semantic segmentation model of the rail light band.

[0033] In step S2, it also includes step S21, step S22, and step S23, which specifically include:

[0034] Step S21: Construct the rail light band coordinate position information according to the first information to obtain the rail surface ground truth image information.

[0035] In this step, divide the rail image information into two categories, namely the rail light band and the rail, and manually annotate the rail image information to obtain the rail surface ground truth image information.

[0036] Step S22: Preprocess the rail surface ground truth image information to obtain the preprocessed rail surface ground truth image information.

[0037] In step S23, it also includes step S231 and step S232, which specifically include:

[0038] Step S231: Perform a first preprocessing on the rail surface ground truth image information to obtain the preprocessed rail surface ground truth image information. The first preprocessing is used to perform a mapping process on the rail surface ground truth image information.

[0039] In this step, the first preprocessing is specifically:

[0040] f[P(i,j,channel)] = C(i,j,channel)

[0041] In the above formula, i and j respectively represent the row number and column number in the rail surface ground truth image; channel represents the channel in the rail surface ground truth image, and the value is 0, 1, 2; f represents a specific mapping function, P(i,j,channel) represents the pixel value of a certain channel of the rail surface ground truth image; C(i,j,channel) represents the pixel value of the processed image on a certain channel. It should be noted that in this step, the grid with a pixel value of 255 needs to be mapped to 1.

[0042] Step S232: Perform second preprocessing on the first information to obtain the preprocessed rail image information after the second preprocessing. The second preprocessing is used to reduce the pixel values of the first information.

[0043] In this step, the second preprocessing is specifically as follows:

[0044] P(i, j, channel) = [C(i, j, channel) / k]

[0045] Where i and j respectively represent the row number and column number of the rail image; channel represents the channel of the rail image, taking values of 0, 1, 2; k represents the processing parameter determined according to segmentation classification; P(i, j, channel) represents the pixel value of a certain channel of the rail image; C(i, j, channel) represents the pixel value of the processed image on a certain channel. It should be noted that using the above formula can reduce the pixel values of the rail image, accelerate the convergence of the neural network, and enable the neural network to better search for digital signals of specific targets from the image big data, thereby improving the detection speed.

[0046] Step S23: Construct a training set based on the rail surface ground truth image information and the first information, and train the semantic segmentation model of the rail light band to obtain the trained semantic segmentation model of the rail light band.

[0047] Step S3: Extract the light band information dataset according to the trained semantic segmentation model of the rail light band;

[0048] In step S3, there are also steps S31, S32, S33, S34, S35, and S36, which specifically include:

[0049] Step S31: Obtain the rail image information in the training set sample;

[0050] In this step, a specific implementation is to fix the size of the rail image information in the training set sample to 240X320X1 after obtaining it.

[0051] Step S32: Send the rail image information in the training set sample to the deep learning network model for downsampling to obtain the first feature image information, which includes feature images of different sizes;

[0052] In this step, the feature images of different sizes are specifically 1 / 2, 1 / 4, 1 / 8, and 1 / 16 of the original size.

[0053] Step S33: Upsample the feature image with the smallest size in the first feature image information to obtain the second feature image information;

[0054] In this step, the feature image with an original size of 1 / 16 is upsampled to obtain a feature image with an original size of 1 / 8, i.e., the second feature image.

[0055] Step S34: After fusing the feature images with the same size in the second feature image information and the first feature image information, perform upsampling to obtain the third feature image information;

[0056] In this step, after combining the feature maps with an original size of 1 / 8 in the second feature image and the first feature image, perform upsampling when necessary to obtain the third feature image information. It should be noted that the size of the third feature image information is 1 / 4 of the original size.

[0057] Step S35: Repeatedly perform upsampling on the third feature image information to obtain the fourth feature image information, and the size of the fourth feature image information is the same as the rail image information in the training set samples;

[0058] In this step, repeat the above upsampling process until the fourth feature image information is obtained.

[0059] Step S36: Establish the light band information dataset according to the fourth feature image information.

[0060] In this step, traverse each row of pixels in the connected regions of the fourth feature image information, calculate the position coordinates of the rail light band on the rail, the width of the rail light band, and the change magnitude of the rail light band width along the extension direction of the rail, and establish the light band information dataset;

[0061] Step S4: Detect the surface state of the rail according to the light band information dataset and the second information.

[0062] In step S4, there are also steps S41, S42, S43, and S44, which specifically include:

[0063] Step S41: Perform normalization processing on the second information to obtain the second information after normalization processing;

[0064] In step S41, there are also steps S411, S412, and S413, which specifically include:

[0065] Step S411: Use a sliding window to calculate the average value of the data in the second information to obtain the first calculation result;

[0066] In this step, the specific process of using a sliding window to calculate the average value of the data in the second information is as follows:

[0067]

[0068] In the above formula, x i ‘ is the data value after the i-th moving average; x i is the i-th data value; n is the window length of the moving average filter.

[0069] Step S412: Normalize the first calculation result to obtain a second calculation result;

[0070] In this step, normalizing the data is a well-known technical solution to those skilled in the art, so it will not be elaborated here.

[0071] Step S413: Perform a logarithmic transformation on the second calculation result to standardize the second information.

[0072] In this step, the specific process of performing a logarithmic transformation on the second calculation result is as follows:

[0073] x log = ln(x2)

[0074] In the above formula, x log represents the third calculation result, x2 is the second calculation result. In this step, the logarithmic transformation converts data within one numerical range to data within another numerical range, and realizing the compression and expansion of the data fluctuation range can effectively reduce the volatility of the data and make it smoother.

[0075] Step S42: Send the second information after the standardization process to the embedding layer to obtain a first feature vector;

[0076] Step S43: Send the first feature vector to the multi-head self-attention model to obtain a second feature vector;

[0077] In this step, three matrix feature vectors of the feature layer are obtained by using three fully connected layers for the first feature vector respectively, where the first matrix feature vector is Query (Q), the second matrix feature vector is Key (K), and the third matrix feature vector is Value (V); perform matrix multiplication operations on the obtained first matrix feature vector and the second matrix feature vector, and multiply the calculation result by a scaling factor to obtain a sub-feature vector; use the SoftMax activation function to convert all elements in the sub-feature vector matrix into relative probabilities between different elements to obtain a matrix feature layer with a probability distribution; multiply the matrix feature layer with the probability distribution and the third matrix feature vector to obtain a second feature vector.

[0078] Step S44: Send the second feature vector to the neural network model to obtain a third feature vector, and the third feature vector is the prediction result of the dynamic performance of the rail surface.

[0079] In this step, the neural network model is a transformer block. A data mapping relationship is established through the transformer network. It should be noted that Transformer does not rely on fixed position information. Position information is introduced into the model through position encoding, enabling it to process input data of different lengths and orders, and having good versatility.

[0080] Embodiment 2:

[0081] As Figure 2 shown, this embodiment provides a rail surface condition detection system based on deep learning. The system includes an acquisition module 901, a first processing module 902, a second processing module 903, and a third processing module 904, which specifically include:

[0082] The acquisition module 901 is used to acquire the first information and the second information. The first information includes the rail image information of the complete section, and the second information includes the rail dynamic performance information corresponding to the complete section;

[0083] The first processing module 902 is used to construct a training set based on the first information to obtain a trained semantic segmentation model of the rail light band;

[0084] The second processing module 903 is used to extract the light band information dataset according to the trained semantic segmentation model of the rail light band;

[0085] The third processing module 904 is used to detect the rail surface condition according to the light band information dataset and the second information.

[0086] In a specific implementation manner of the present disclosure, the first processing module 902 further includes a first processing unit 9021, a second processing unit 9022, and a training unit 9023, which specifically include:

[0087] The first processing unit 9021 is used to construct the rail light band coordinate position information based on the first information to obtain the true value image information of the rail surface;

[0088] The second processing unit 9022 is used to preprocess the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface;

[0089] The training unit 9023 is used to construct a training set based on the true value image information of the rail surface and the first information and train the semantic segmentation model of the rail light band to obtain a trained semantic segmentation model of the rail light band.

[0090] In a specific embodiment of the present disclosure, the second processing unit 9022 further includes a first preprocessing unit 90221 and a second preprocessing unit 90222, which specifically include:

[0091] The first preprocessing unit 90221 is configured to perform a first preprocessing on the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface, and the first preprocessing is used to perform a mapping process on the true value image information of the rail surface;

[0092] The second preprocessing unit 90222 is configured to perform a second preprocessing on the first information to obtain the second preprocessed rail image information, and the second preprocessing is used to reduce the pixel values of the first information.

[0093] In a specific embodiment of the present disclosure, the third processing module 904 further includes a third processing unit 9041, a fourth processing unit 9042, a fifth processing unit 9043, and a sixth processing unit 9044, which specifically include:

[0094] The third processing unit 9041 is configured to perform a normalization process on the second information to obtain the second information after the normalization process;

[0095] The fourth processing unit 9042 is configured to send the second information after the normalization process to the embedding layer to obtain a first feature vector;

[0096] The fifth processing unit 9043 is configured to send the first feature vector to the multi-head self-attention model to obtain a second feature vector;

[0097] The sixth processing unit 9044 is configured to send the second feature vector to the neural network model to obtain a third feature vector, and the third feature vector is the prediction result of the dynamic performance of the rail surface.

[0098] In a specific embodiment of the present disclosure, the third processing unit 9041 further includes a first calculation unit 90411, a second calculation unit 90412, and a third calculation unit 90413, which specifically include:

[0099] The first calculation unit 90411 is configured to calculate the average value of the data in the second information by using a sliding window to obtain a first calculation result;

[0100] The second calculation unit 90412 is configured to perform a normalization process on the first calculation result to obtain a second calculation result;

[0101] The third calculation unit 90413 is configured to perform a logarithmic transformation on the second calculation result to obtain the second information after the normalization process.

[0102] In a specific embodiment of the present disclosure, the second processing module 903 further includes an acquisition unit 9031, a seventh processing unit 9032, an eighth processing unit 9033, a ninth processing unit 9034, a tenth processing unit 9035, and an eleventh processing unit 9036, which specifically include:

[0103] The acquisition unit 9031 is configured to acquire rail image information in the training set samples;

[0104] The seventh processing unit 9032 is configured to send the rail image information in the training set samples to a deep learning network model for downsampling to obtain first feature image information, where the first feature image information includes feature images of different sizes;

[0105] The eighth processing unit 9033 is configured to upsample the feature image with the smallest size in the first feature image information to obtain second feature image information;

[0106] The ninth processing unit 9034 is configured to fuse the second feature image information with the feature images of the same size in the first feature image information and then perform upsampling to obtain third feature image information;

[0107] The tenth processing unit 9035 is configured to repeatedly perform upsampling on the third feature image information to obtain fourth feature image information, where the size of the fourth feature image information is the same as the rail image information in the training set samples;

[0108] The eleventh processing unit 9036 is configured to establish the light band information data set according to the fourth feature image information.

[0109] It should be noted that regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0110] Embodiment 3:

[0111] Corresponding to the above method embodiment, a rail surface state detection device based on deep learning is further provided in this embodiment. A rail surface state detection device based on deep learning described below can be correspondingly referred to the above-described method for detecting the rail surface state based on deep learning.

[0112] Figure 3 It is a block diagram of a rail surface state detection device 800 based on deep learning shown according to an exemplary embodiment. As Figure 3As shown, the deep learning-based rail surface condition detection device 800 may include: a processor 801 and a memory 802. The deep learning-based rail surface condition detection device 800 may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0113] Among them, the processor 801 is used to control the overall operation of the deep learning-based rail surface condition detection device 800 to complete all or part of the steps in the above-mentioned deep learning-based rail surface condition detection method. The memory 802 is used to store various types of data to support the operation of the deep learning-based rail surface condition detection device 800. These data may include, for example, instructions for any application or method operating on the deep learning-based rail surface condition detection device 800, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, and so on. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The multimedia component 803 may include a screen and an audio component. The screen can be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the deep learning-based rail surface condition detection device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G, or 4G, or a combination of one or more of them. Accordingly, the communication component 805 may include: a Wi-Fi module, a Bluetooth module, and an NFC module.

[0114] In an exemplary embodiment, the deep learning-based rail surface condition detection device 800 can be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, and is used to execute the above-mentioned deep learning-based rail surface condition detection method.

[0115] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided. When the program instructions are executed by a processor, the steps of the above-mentioned deep learning-based rail surface condition detection method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above program instructions can be executed by the processor 801 of the deep learning-based rail surface condition detection device 800 to complete the above-mentioned deep learning-based rail surface condition detection method.

[0116] Embodiment 4:

[0117] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. A readable storage medium described below can be correspondingly referred to with a deep learning-based rail surface condition detection method described above.

[0118] A readable storage medium has a computer program stored thereon. When the computer program is executed by a processor, the steps of the deep learning-based rail surface condition detection method in the above method embodiment are implemented.

[0119] The readable storage medium can specifically be various readable storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0120] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0121] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for detecting the surface state of rail based on deep learning, characterized in that Including: Obtain the first information and the second information, where the first information includes the rail image information of the complete section, and the second information includes the rail dynamic performance information corresponding to the complete section; Construct a training set according to the first information to obtain a trained semantic segmentation model of the rail light band; Extract the light band information dataset according to the trained semantic segmentation model of the rail light band; Detect the surface state of the rail according to the light band information dataset and the second information.

2. The method for detecting the surface state of a steel rail based on deep learning according to claim 1, wherein Constructing a training set according to the first information includes: Construct the rail light band coordinate position information according to the first information to obtain the true value image information of the rail surface; Preprocess the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface; Construct a training set according to the true value image information of the rail surface and the first information, and train the semantic segmentation model of the rail light band to obtain a trained semantic segmentation model of the rail light band.

3. The method for detecting the surface state of a steel rail based on deep learning according to claim 2, characterized in that, Preprocessing the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface includes: Perform the first preprocessing on the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface, where the first preprocessing is used to perform mapping processing on the true value image information of the rail surface; Perform the second preprocessing on the first information to obtain the second preprocessed rail image information, where the second preprocessing is used to reduce the pixel value of the first information.

4. The method for detecting the surface state of steel rails based on deep learning according to claim 1, characterized in that Detecting the surface state of the rail according to the light band information dataset and the second information includes: Perform standardization processing on the second information to obtain the standardized second information; Send the standardized second information to the embedding layer to obtain the first feature vector; Send the first feature vector to the multi-head self-attention model to obtain the second feature vector; Send the second feature vector to the neural network model to obtain the third feature vector, where the third feature vector is the prediction result of the rail surface dynamic performance.

5. The method for detecting the surface state of a steel rail based on deep learning according to claim 4, wherein Performing standardization processing on the second information to obtain the standardized second information includes: Use a sliding window to calculate the average value of the data in the second information to obtain the first calculation result; Perform normalization processing on the first calculation result to obtain the second calculation result; Perform logarithmic transformation on the second calculation result to obtain the standardized second information.

6. A rail surface condition detection system based on deep learning, characterized in that, Including: An acquisition module for acquiring the first information and the second information, where the first information includes the rail image information of the complete section, and the second information includes the rail dynamic performance information corresponding to the complete section; A first processing module for constructing a training set according to the first information to obtain a trained semantic segmentation model of the rail light band; A second processing module for extracting the light band information dataset according to the trained semantic segmentation model of the rail light band; A third processing module for detecting the surface state of the rail according to the light band information dataset and the second information.

7. The rail surface condition detection system based on deep learning according to claim 6, wherein, The first processing module includes: A first processing unit for constructing the rail light band coordinate position information according to the first information to obtain the true value image information of the rail surface; A second processing unit for preprocessing the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface; A training unit for constructing a training set based on the true value image information of the rail surface and the first information and training a semantic segmentation model of the rail light band to obtain a trained semantic segmentation model of the rail light band.

8. The deep learning-based rail surface condition detection system according to claim 7, wherein The second processing unit includes: A first preprocessing unit for performing a first preprocessing on the true value image information of the rail surface to obtain the preprocessed true value image information of the rail surface, where the first preprocessing is used to perform a mapping process on the true value image information of the rail surface; A second preprocessing unit for performing a second preprocessing on the first information to obtain the second preprocessed rail image information, where the second preprocessing is used to reduce the pixel values of the first information.

9. The rail surface condition detection system based on deep learning according to claim 6, wherein, The third processing module includes: A third processing unit for performing a normalization process on the second information to obtain the normalized second information; A fourth processing unit for sending the normalized second information to an embedding layer to obtain a first feature vector; A fifth processing unit for sending the first feature vector to a multi-head self-attention model to obtain a second feature vector; A sixth processing unit for sending the second feature vector to a neural network model to obtain a third feature vector, where the third feature vector is the prediction result of the dynamic performance of the rail surface.

10. The rail surface condition detection system based on deep learning according to claim 9, characterized in that, The third processing unit includes: A first calculation unit for calculating the average value of the data in the second information using a sliding window to obtain a first calculation result; A second calculation unit for performing a normalization process on the first calculation result to obtain a second calculation result; A third calculation unit for performing a logarithmic transformation on the second calculation result to obtain the normalized second information.

Images