A signal processing method, apparatus and electronic device

Abstract

The application provides a signal processing method, device and electronic equipment. The application collects a rail audio signal in a train operation process, predicts a rail state of a high-frequency component in the rail audio signal, obtains the rail state, combines track information of the train, determines a section with a running risk in a section through which the train passes, realizes automatic detection and analysis of a running risk of the rail, improves detection efficiency of high-speed railway rail detection, and ensures safe operation of the high-speed railway.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of train safety detection, and in particular to a signal processing method and device and electronic equipment. BACKGROUND

[0002] With the rapid development of high-speed railway, the safety detection of steel rails is particularly important. In the process of use, the steel rails will be broken, cracked and other damage forms that affect and limit the performance of the steel rails, i.e. rail damage. There are many types of rail damage, common ones include wear, peeling, rail head core damage, rail waist butterfly, hole cracks, etc. Rail damage is the main cause of rail breakage and an important hidden danger affecting train safety. Train derailment accidents are mainly caused by rail breakage. During acceleration and braking, and when passing through rail joints, curves and switches, the steel rails are subjected to strong friction, extrusion, bending and impact. Their long-term repeated effects make the steel rails prone to fatigue cracks. Once the cracks occur, they are easy to quickly expand, causing major and severe accidents such as rail breakage. The friction, extrusion, bending and impact of high-speed trains on the rails are more prominent, so high-speed railway rails are more prone to cracking, and the speed of developing from cracks to rail breakage is faster. In order to ensure the safe operation of high-speed railways, the detection cycle of high-speed railways must be shortened. At the same time, due to the high density of high-speed railway traffic and high speed, the traditional detection speed is difficult to meet the needs of high-speed railways.

[0003] Currently, the detection of high-speed railway rails in China mainly relies on manual inspection and rail inspection vehicle detection. Manual inspection is time-consuming, labor-intensive and has poor reliability. The time for on-site inspection is insufficient, and manual inspection cannot meet the needs of high-speed railway safety inspection. The detection of track spatial geometric size information, rail crack degree index and rail stress index mainly relies on the combination of rail inspection vehicles and manual inspection. Rail inspection vehicles have complete detection items, but they have long detection cycles, high costs, limited detection positions and angles, and poor flexibility. The current rail detection has low detection efficiency, and high-speed railways have safety hazards. SUMMARY

[0004] The present application provides a signal processing method, device and electronic equipment, which can improve the detection efficiency of high-speed railway rail detection and ensure the safe operation of high-speed railways.

[0005] In a first aspect, the present application provides a signal processing method, comprising: acquiring a rail audio signal collected during train operation and track information of the train; performing spectral analysis and sliding window segmentation on the rail audio signal to separate normal components and high-frequency components of the rail audio signal, and obtaining high-frequency components of multiple time periods; obtaining a rail state corresponding to the high-frequency components of each time period based on the high-frequency components of each time period and a preset rail state prediction model; the rail state includes a normal state, a stress state and a hidden damage state; determining a section with a running risk in a route section of the train based on the rail state corresponding to the high-frequency components of each time period and the track information of the train.

[0006] In a possible implementation, the rail state corresponding to the high-frequency components of each time period is obtained based on the high-frequency components of each time period and the preset rail state prediction model, comprising: performing spectral analysis and grayscale processing on the high-frequency components of each time period to obtain multiple grayscale images; performing wavelet transform and inverse wavelet transform on the multiple grayscale images to enhance the high-frequency part of each grayscale image and obtain multiple high-frequency images; inputting the multiple high-frequency images into the preset rail state prediction model to obtain the rail state corresponding to the high-frequency components of each time period.

[0007] In a possible implementation, before the rail state corresponding to the high-frequency components of each time period is obtained based on the high-frequency components of each time period and the preset rail state prediction model, the method further comprises: acquiring multiple rail audio signals with known rail states; generating training samples based on the multiple rail audio signals with known rail states; and performing neural network training based on the training samples to obtain the preset rail state prediction model.

[0008] In a possible implementation, each training sample in the training samples takes a high-frequency image as input and takes a rail state corresponding to the high-frequency image as output; the preset rail state prediction model is obtained by performing neural network training based on the training samples, comprising: performing neural network training based on each training sample taking a high-frequency image as input and taking a rail state as output to obtain a first prediction model; constructing a teacher-student model taking the first prediction model as a teacher model; wherein the teacher model and the student model in the teacher-student model are neural network models with the same depth; performing multi-scale knowledge distillation training and pruning quantization on the student model based on the detection accuracy of the teacher model and the student model, and determining the trained student model as the preset rail state prediction model.

[0009] In a possible implementation, based on the detection accuracy of the teacher model and the student model, the student model is subjected to multi-scale knowledge distillation training and pruning quantization, and the trained student model is determined as the preset rail state prediction model, including: based on the detection accuracy of the teacher model and the student model, the student model is subjected to multi-scale knowledge distillation training; the number of channels of each normalization layer in the trained student model and the weight values of each channel are calculated; the channels with weight values less than a threshold in each normalization layer are deleted to obtain a simplified student model, wherein at least one channel is reserved for each normalization layer; based on each training sample, the simplified student model is retrained, and the trained student model is determined as the preset rail state prediction model.

[0010] In a possible implementation, after the high-frequency component of each period is input into the preset rail state prediction model to obtain the rail state corresponding to the high-frequency component of each period, the method further includes: if the rail state corresponding to the high-frequency component of a period is a stress state, inputting the high-frequency component of the period into a preset stress prediction model to obtain the stress corresponding to the high-frequency component of the period; calculating the stress level corresponding to the high-frequency component of the period; determining the position of the rail with stress based on the period in which the high-frequency component is located and the track information of the train; generating a first instruction based on the stress level corresponding to the high-frequency component of the period and the position of the rail with stress, the first instruction being used to instruct an operation and maintenance personnel to detect the position of the rail with stress through a rail inspection vehicle; and sending the first instruction to a user.

[0011] In a possible implementation, after the high-frequency component of each period is input into the preset rail state prediction model to obtain the rail state corresponding to the high-frequency component of each period, the method further includes: obtaining a driving video shot during driving of the train, extracting a video clip corresponding to the position of the rail with stress, and analyzing to obtain a tie laying condition and a rail deformation condition; determining the operation risk level of the position of the rail with stress based on the tie laying condition and the rail deformation condition; and determining the emergency level of the first instruction based on the operation risk level of the position of the rail with stress.

[0012] In a possible implementation, after the first instruction is sent to the user, the method further includes: receiving a first response, the first response including a detection result of the rail inspection vehicle on the position of the rail with stress; if the detection result is that the rail is normal, retraining the rail state prediction model based on the high-frequency component of the period and the normal state corresponding to the high-frequency component of the period to obtain a retrained rail state prediction model; and performing rail state detection based on the retrained rail state prediction model; and if the detection result is a stress state or a hidden damage state, generating a maintenance instruction to instruct the user to maintain the position of the rail with stress.

[0013] In a second aspect, an embodiment of the present application provides a signal processing apparatus, comprising: a communication unit configured to acquire a rail audio signal collected during train operation and track information of the train; and a processing unit configured to perform spectral analysis and sliding window segmentation on the rail audio signal, separate normal components and high-frequency components of the rail audio signal, and obtain high-frequency components of multiple time periods; obtain rail states corresponding to the high-frequency components of the multiple time periods based on the high-frequency components of the multiple time periods and a preset rail state prediction model; the rail states include a normal state, a stress state, and a state of existing hidden damage; and determine a section with a running risk in a route section of the train based on the rail states corresponding to the high-frequency components of the multiple time periods and the track information of the train.

[0014] In a possible implementation, the processing unit is specifically configured to perform spectral analysis and grayscale processing on the high-frequency components of the multiple time periods to obtain multiple grayscale images; perform wavelet transform and inverse wavelet transform on the multiple grayscale images to enhance high-frequency parts in the grayscale images and obtain multiple high-frequency images; and input the multiple high-frequency images into the preset rail state prediction model to obtain the rail states corresponding to the high-frequency components of the multiple time periods.

[0015] In a possible implementation, the processing unit is further configured to acquire multiple rail audio signals with known rail states; generate training samples based on the multiple rail audio signals with the known rail states; and perform neural network training based on the training samples to obtain the preset rail state prediction model.

[0016] In a possible implementation, each training sample in the training samples takes a high-frequency image as input and takes a rail state corresponding to the high-frequency image as output; the processing unit is specifically configured to perform neural network training based on the training samples, take the high-frequency image as input, and take the rail state as output to obtain a first prediction model; construct a teacher-student model by taking the first prediction model as a teacher model; the teacher model and a student model in the teacher-student model are neural network models with the same depth; perform multi-scale knowledge distillation training and pruning quantization on the student model based on detection accuracies of the teacher model and the student model, and determine a trained student model as the preset rail state prediction model.

[0017] In a possible implementation, the processing unit is specifically configured to perform multi-scale knowledge distillation training on the student model based on the detection accuracies of the teacher model and the student model; calculate a number of channels of each normalization layer and weight values of each channel in the trained student model; delete channels with weight values less than a threshold value in each normalization layer to obtain a simplified student model, wherein at least one channel is reserved in each normalization layer; retrain the simplified student model based on the training samples, and determine a trained student model as the preset rail state prediction model.

[0018] In a possible implementation, the processing unit is further configured to: if the high-frequency component of the time period corresponds to a stressed state of the rail, input the high-frequency component of the time period into a preset stress prediction model to obtain a stress corresponding to the high-frequency component of the time period; calculate a stress level corresponding to the high-frequency component of the time period; determine a position of the rail with stress based on a time period in which the high-frequency component is located and trajectory information of the train; generate a first instruction based on the stress level corresponding to the high-frequency component of the time period and the position of the rail with stress, the first instruction being used to instruct a track inspection vehicle to detect the position of the rail with stress; and send the first instruction to a user.

[0019] In a possible implementation, the processing unit is further configured to: obtain a driving video captured during driving of the train, extract a video clip corresponding to the position of the rail with stress, and analyze a tie laying condition and a rail deformation condition; determine a running risk level of the position of the rail with stress based on the tie laying condition and the rail deformation condition; and determine an emergency level of the first instruction based on the running risk level of the position of the rail with stress.

[0020] In a possible implementation, the processing unit is further configured to: receive a first response, the first response including a detection result of the track inspection vehicle on the position of the rail with stress; if the detection result is normal, retrain a rail state prediction model based on the high-frequency component of the time period and the normal state corresponding to the high-frequency component of the time period, to obtain a retrained rail state prediction model; and perform rail state detection based on the retrained rail state prediction model; if the detection result is a stressed state or a dark injury state, generate a maintenance instruction to instruct a user to maintain the position of the rail with stress.

[0021] In a third aspect, an electronic device is provided, including a memory and a processor, the memory storing a computer program, and the processor being configured to invoke and run the computer program stored in the memory to perform the steps of the method in the first aspect and any possible implementation manner of the first aspect.

[0022] In a fourth aspect, a computer readable storage medium is provided, the computer readable storage medium storing a computer program, and the computer program being configured to perform the steps of the method in the first aspect and any possible implementation manner of the first aspect when executed by a processor.

[0023] The application provides a signal processing method, device and electronic equipment, and the application collects rail audio signals in the running process of a train, predicts the rail state of high-frequency components in the rail audio signals, combines track information of the train after obtaining the rail state, determines a section with a running risk in a section passed by the train, realizes automatic detection and analysis of the running risk of the rail, improves the detection efficiency of high-speed rail detection, and ensures safe operation of the high-speed rail. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0025] Figure 1 is a flowchart of a signal processing method provided by the embodiment of the present application;

[0026] Figure 2 is a structural schematic diagram of a signal processing device provided by the embodiment of the present application;

[0027] Figure 3 is a structural schematic diagram of an electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0028] In the following description, specific details are set forth in order to provide a thorough understanding of the embodiments of the present application. However, persons skilled in the art should understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary details.

[0029] In the description of the present application, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this paper is only a description of the relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, "at least one" "multiple" means two or more. "First", "second", etc. are not limited in quantity and execution order, and "first", "second", etc. are not necessarily different.

[0030] In the embodiments of the present application, the word "exemplary" or "for example" is used to mean serving as an example, instance, or illustration. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, use of the word "exemplary" or "for example" is intended to present concepts in a concrete manner in order to facilitate understanding of the application.

[0031] In addition, the terms "comprising" and "having," and any variations thereof, as used in the description of the present application, are intended to cover both the inclusive and exclusive cases. For example, a process, method, system, product, or apparatus that comprises a list of steps or modules is not limited to the listed steps or modules, but can optionally further include other steps or modules not listed, or can optionally further include other steps or modules inherent to such processes, methods, products, or apparatus.

[0032] In order to make the objects, technical solutions, and advantages of the present application clearer, the following will be described with reference to the accompanying drawings of the present application through specific embodiments.

[0033] As shown in Figure 1 The present embodiment provides a signal processing method. The execution subject is a signal processing device. The method comprises steps S101-S104.

[0034] S101, acquiring a rail audio signal collected in a train running process, and track information of the train.

[0035] In some embodiments, the rail audio signal is a sound emitted by the rail collected in the train running process; or the rail audio signal is a signal reflected by the rail after the train emits an infrared signal to the rail in the train running process; or the rail audio signal can also be a signal reflected by the rail after the train emits an ultrasonic wave to the rail in the train running process.

[0036] As a possible implementation manner, the train can collect the audio signal of the rail along the line by using an audio receiver installed at the bottom of the train in the train running process.

[0037] As another possible implementation manner, the train can collect the rail audio signal by emitting an infrared signal to the rail and receiving the echo signal reflected by the rail in the train running process.

[0038] As another possible implementation manner, the train can collect the rail audio signal by emitting an ultrasonic wave to the rail and receiving the ultrasonic wave signal reflected by the rail in the train running process.

[0039] In some embodiments, the track information of the train comprises position information at each time and rail information running at each time in the train running process.

[0040] As one possible approach, the train can use positioning devices to determine its location at different times during its journey.

[0041] As one possible approach, the train can obtain a travel log and extract information about the rails traveling at different times from the log.

[0042] S102. Perform spectrum analysis and sliding window segmentation on the rail audio signal to separate the normal component and high-frequency component of the rail audio signal, and obtain the high-frequency component of multiple time periods.

[0043] As one possible implementation, embodiments of the present invention can perform Fourier transform on the rail audio signal and perform spectral analysis to obtain the frequency and amplitude of the rail audio signal; based on the frequency and amplitude, the rail audio signal can be separated to obtain normal components and high-frequency components. Embodiments of the present invention can use a sliding window to segment the normal components and high-frequency components to obtain normal components and high-frequency components for multiple time periods.

[0044] S103. Based on the high-frequency components of each time period and the preset rail condition prediction model, the rail condition corresponding to the high-frequency components of each time period is obtained.

[0045] In this embodiment of the application, the rail condition includes normal condition, stressed condition, and condition with hidden damage.

[0046] As one possible implementation, embodiments of the present invention can analyze and obtain the rail condition corresponding to the high-frequency components of each time period based on steps S1031-S1033.

[0047] S1031. Perform spectral analysis and grayscale processing on the high-frequency components of each time period to obtain multiple grayscale images.

[0048] As one possible implementation, for any high-frequency component within a given time period, this embodiment of the invention can perform a decimal conversion based on the amplitude of the high-frequency component at each moment within that time period to obtain an image matrix. Each element in the image matrix represents the amplitude value at each moment. Then, using the element values ​​in the image matrix as pixel values ​​for a grayscale image, a grayscale image is generated, thus achieving the grayscale image conversion of the high-frequency component within that time period.

[0049] S1032. Perform wavelet transform and inverse wavelet transform on multiple grayscale images to enhance the high-frequency components in each grayscale image, thereby obtaining multiple high-frequency images.

[0050] As one possible implementation, for any grayscale image, embodiments of the present invention can perform wavelet transform on the grayscale image to obtain multiple sub-images. For example, embodiments of the present invention can perform convolution operations on the grayscale image in the horizontal and vertical directions based on a low-pass filter or a high-pass filter to obtain each sub-image.

[0051] The plurality of sub-images can include an LL sub-image, an LH sub-image, an HL sub-image, and an HH sub-image. The LL sub-image represents low-frequency information of the gray-scale image, the LH sub-image represents a horizontal detail component of high-frequency information of the gray-scale image, the HL sub-image represents a vertical detail component of high-frequency information of the gray-scale image, and the HH sub-image represents a diagonal detail component of high-frequency information of the gray-scale image.

[0052] Afterwards, the embodiment of the present application can perform denoising processing on the plurality of sub-images based on a preset threshold value to obtain a plurality of denoised sub-images. The plurality of denoised sub-images are subjected to inverse wavelet transform to obtain a high-frequency image.

[0053] Illustratively, the embodiment of the present application can perform inverse wavelet transform on the plurality of denoised sub-images based on inverse wavelet transform coefficients to obtain the high-frequency image. It should be noted that the embodiment of the present application can perform threshold quantization processing on the wavelet transform coefficients to obtain the inverse wavelet transform coefficients. The inverse wavelet transform coefficients include low-frequency coefficients and high-frequency coefficients subjected to threshold quantization processing.

[0054] Illustratively, the embodiment of the present application can determine the high-frequency coefficients subjected to threshold quantization processing based on the following formula.

[0055]

[0056] wherein η(ω) is the high-frequency coefficient subjected to threshold quantization processing, ω is the high-frequency coefficient, T is the denoising threshold value, and k is the element order in the sub-image.

[0057] In the embodiment of the present application, the high-frequency coefficient ω can be a horizontal detail coefficient cH j+1 (m,n) in the high-frequency coefficient, a vertical detail coefficient cV j+1 (m,n) in the high-frequency coefficient, and a diagonal detail coefficient cD j+1 (m,n) in the high-frequency coefficient.

[0058] As a possible implementation manner, the embodiment of the present application can reconstruct the high-frequency coefficients subjected to threshold quantization processing and the low-frequency coefficients to obtain a denoised image. Illustratively, the embodiment of the present application can input the high-frequency coefficients subjected to threshold quantization processing and the low-frequency coefficients into a decomposition formula of the gray-scale image to inversely deduce an expression of the denoised image to obtain the high-frequency image.

[0059] S1033, inputting the plurality of high-frequency images into a preset rail state prediction model to obtain a rail state corresponding to a high-frequency component of each time period.

[0060] Therefore, before prediction by the rail state prediction model, the embodiment of the present application can perform denoising by wavelet processing, so that the high-frequency part in the high-frequency component of each time period is clearer, and the accuracy of the prediction result is improved.

[0061] In S104, based on the rail state corresponding to the high-frequency component of each time period and the track information of the train, a section with a running risk in the route of the train is determined.

[0062] As a possible implementation manner, if the rail state corresponding to the high-frequency component of a time period is a stress state, the embodiment of the present application can extract the position information of the train in the time period and the rail information traveled by the train in the time period based on the track information of the train, determine the route of the train in the time period based on the position information of the train in the time period and the rail information traveled by the train in the time period, and determine the route of the train in the time period as the section with a running risk in the route of the train.

[0063] The present application provides a signal processing method, which collects rail audio signals in the running process of a train, predicts the rail state of the high-frequency component in the rail audio signals, determines a section with a running risk in the route of the train after obtaining the rail state in combination with the track information of the train, realizes automatic detection and analysis of the running risk of the rail, and improves the detection efficiency of the detection of the rail of the high-speed railway, thereby ensuring the safe operation of the high-speed railway.

[0064] Optionally, the signal processing method provided by the embodiment of the present application further comprises steps S201-S203 before step S103.

[0065] In S201, a plurality of rail audio signals with known rail states are obtained.

[0066] In S202, training samples are generated based on the plurality of rail audio signals with known rail states.

[0067] In some embodiments, each training sample in the training samples takes a high-frequency image as input and takes the rail state corresponding to the high-frequency image as output.

[0068] As a possible implementation manner, the embodiment of the present application can perform frequency spectrum analysis and sliding window segmentation on the plurality of rail audio signals with known rail states, separate the normal component and the high-frequency component of the rail audio signals, and obtain the high-frequency components of a plurality of time periods. Then, frequency spectrum analysis and grayscale processing are performed on the high-frequency components of the plurality of time periods to obtain a plurality of grayscale images. Wavelet transformation and inverse wavelet transformation are performed on the plurality of grayscale images to enhance the high-frequency part in each grayscale image, and a plurality of high-frequency images are obtained.

[0069] Therefore, the present application can determine the training samples by taking the high-frequency image as input and taking the rail state corresponding to the high-frequency image as output.

[0070] S203, based on the training samples, performing neural network training to obtain a preset rail state prediction model.

[0071] As a possible implementation manner, the embodiment of the present application can obtain the preset rail state prediction model based on steps S2031-S2033.

[0072] S2031, based on each training sample, taking the high-frequency image as input and taking the rail state as output, performing neural network training to obtain a first prediction model.

[0073] S2032, taking the first prediction model as a teacher model, constructing a teacher-student model. The teacher model and the student model in the teacher-student model are neural network models of the same depth.

[0074] S2033, based on the detection accuracy of the teacher model and the student model, performing multi-scale knowledge distillation training and pruning quantization on the student model, and determining the trained student model as the preset rail state prediction model.

[0075] For example, step S2033 can be specifically implemented as steps A1-A4.

[0076] A1, based on the detection accuracy of the teacher model and the student model, performing multi-scale knowledge distillation training on the student model.

[0077] A2, calculating the number of channels and the weight values of each channel in each normalization layer of the trained student model.

[0078] A3, deleting the channels with weight values less than a threshold in each normalization layer to obtain a simplified student model. Each normalization layer retains at least one channel.

[0079] A4, based on each training sample, retraining the simplified student model, and determining the trained student model as the preset rail state prediction model.

[0080] In this way, the embodiment of the present application can perform knowledge distillation through the teacher-student model, simplify the structure of the rail state prediction model, improve the model calculation efficiency on the premise of ensuring the prediction accuracy of the rail state prediction model, and improve the rail state prediction efficiency.

[0081] Optionally, the signal processing method provided by the embodiment of the present application further includes steps S301-S308 after step S103.

[0082] S301, if the rail state corresponding to the high-frequency component of a certain period is a stress state, inputting the high-frequency component of the period into a preset stress prediction model to obtain the stress corresponding to the high-frequency component of the period.

[0083] S302, determine a stress level corresponding to the high-frequency component of the period.

[0084] S303, determine a position of the stressed rail based on the period in which the high-frequency component is located and the trajectory information of the train.

[0085] S304, generate a first instruction based on the stress level corresponding to the high-frequency component of the period and the position of the stressed rail.

[0086] The first instruction is used to instruct the maintenance personnel to detect the position of the stressed rail by the track inspection vehicle.

[0087] S305, send the first instruction to the user.

[0088] In this way, the embodiment of the present application can determine the stress level and the position of the stressed rail after determining that the rail is in a stressed state, and instruct the maintenance personnel to detect and confirm the stress level and the position of the stressed rail, thereby ensuring the accuracy of rail detection.

[0089] S306, receive a first response.

[0090] In some embodiments, the first response includes the detection result of the track inspection vehicle on the position of the stressed rail.

[0091] S307, if the detection result is that the rail is normal, retrain the rail state prediction model based on the high-frequency component of the period and the normal state corresponding to the high-frequency component of the period, to obtain a retrained rail state prediction model; and perform rail state detection based on the retrained rail state prediction model.

[0092] In this way, when the model prediction is wrong, the model is retrained to further improve the prediction accuracy of the rail state prediction model.

[0093] S308, if the detection result is a stressed state or a dark injury state, generate a maintenance instruction to instruct the user to maintain the position of the stressed rail.

[0094] In this way, the maintenance instruction is sent to the maintenance personnel to ensure the safety of high-speed train operation.

[0095] Optionally, the signal processing method provided by the embodiment of the present application further includes steps S401-S403 after step S103.

[0096] S401, obtain a driving video taken during train driving.

[0097] S402, extract a video clip corresponding to the position of the stressed rail and analyze to obtain a tie laying condition and a rail deformation condition.

[0098] S403, determine a running risk level of the position of the rail with stress based on the tie laying condition and the rail deformation condition.

[0099] S404, determine an emergency level of the first instruction based on the running risk level of the position of the rail with stress.

[0100] Therefore, when generating the first instruction, the embodiment of the present application can determine the running risk level and the emergency level by judging the tie laying condition and the rail deformation condition through the driving video. When the running risk is high, the operation and maintenance personnel are instructed to prioritize the processing of the first instruction, thereby ensuring the driving safety of the high-speed train.

[0101] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution. The execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0102] The following is a device embodiment of the present application. For details not described in detail, reference can be made to the corresponding method embodiments described above.

[0103] Figure 2 A structure schematic diagram of a signal processing device provided by an embodiment of the present application is shown. The signal processing device 500 includes a communication unit 501 and a processing unit 502.

[0104] The communication unit 501 is configured to acquire the rail audio signal collected in the train running process and the track information of the train.

[0105] The processing unit 502 is configured to perform spectral analysis and sliding window segmentation on the rail audio signal, separate the normal component and the high-frequency component of the rail audio signal, and obtain the high-frequency component of a plurality of time periods; based on the high-frequency component of each time period and a preset rail state prediction model, obtain the rail state corresponding to the high-frequency component of each time period; the rail state includes a normal state, a stress state and a hidden damage state; based on the rail state corresponding to the high-frequency component of each time period and the track information of the train, determine the section with running risk in the route section of the train.

[0106] In a possible implementation, the processing unit 502 is specifically configured to perform spectral analysis and grayscale processing on the high-frequency component of each time period to obtain a plurality of grayscale images; perform wavelet transform and inverse wavelet transform on the plurality of grayscale images to enhance the high-frequency part in each grayscale image to obtain a plurality of high-frequency images; input the plurality of high-frequency images into the preset rail state prediction model to obtain the rail state corresponding to the high-frequency component of each time period.

[0107] In a possible implementation, the processing unit 502 is further configured to acquire a plurality of rail audio signals of a known rail state; generate training samples based on the plurality of rail audio signals of the known rail state; and perform neural network training based on the training samples to obtain the preset rail state prediction model.

[0108] In a possible implementation, each training sample in the training samples takes a high-frequency image as input and takes a rail state corresponding to the high-frequency image as output; the processing unit 502 is specifically configured to perform neural network training based on each training sample, take the high-frequency image as input, and take the rail state as output, to obtain a first prediction model; construct a teacher-student model by taking the first prediction model as a teacher model, wherein the teacher model and the student model in the teacher-student model are neural network models of the same depth; and perform multi-scale knowledge distillation training and pruning quantization on the student model based on detection accuracy of the teacher model and the student model, and determine the trained student model as the preset rail state prediction model.

[0109] In a possible implementation, the processing unit 502 is specifically configured to perform multi-scale knowledge distillation training on the student model based on detection accuracy of the teacher model and the student model; calculate a number of channels of each normalization layer and weight values of each channel in the trained student model; delete a channel with a weight value less than a threshold value in each normalization layer to obtain a simplified student model, wherein at least one channel is reserved in each normalization layer; retrain the simplified student model based on each training sample, and determine the trained student model as the preset rail state prediction model.

[0110] In a possible implementation, the processing unit 502 is further configured to, if the rail state corresponding to the high-frequency component of a time period is a stress state, input the high-frequency component of the time period into a preset stress prediction model to obtain stress corresponding to the high-frequency component of the time period; calculate a stress level corresponding to the high-frequency component of the time period; determine a rail position with stress based on a time period in which the high-frequency component is located and track information of the train; generate a first instruction based on the stress level corresponding to the high-frequency component of the time period and the rail position with stress, the first instruction being used to instruct an operation and maintenance personnel to detect the rail position with stress through a track inspection vehicle; and send the first instruction to a user.

[0111] In a possible implementation, the processing unit 502 is further configured to acquire a driving video shot during train driving, extract a video clip corresponding to the rail position with stress, and analyze to obtain a tie laying condition and a rail deformation condition; determine a running risk level of the rail position with stress based on the tie laying condition and the rail deformation condition; and determine an emergency level of the first instruction based on the running risk level of the rail position with stress.

[0112] In a possible implementation, the processing unit 502 is further configured to receive a first response, the first response comprising a detection result of the rail inspection vehicle on the position of the steel rail under stress; if the detection result is that the steel rail is normal, retrain a steel rail state prediction model based on the high frequency component of the time period and the normal state corresponding to the high frequency component of the time period, to obtain a retrained steel rail state prediction model; and perform steel rail state detection based on the retrained steel rail state prediction model; if the detection result is that the steel rail is under stress or has a hidden damage state, generate a maintenance instruction to instruct a user to maintain the position of the steel rail under stress.

[0113] Figure 3 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 3 the electronic device 600 of this embodiment includes a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. The processor 601 implements the steps in each of the method embodiments described above when executing the computer program 603, for example, steps S101-S104 as shown in Figure 1 Alternatively, the processor 601 implements the functions of each module / unit in each of the device embodiments described above when executing the computer program 603, for example, the functions of the communication unit 501 and the processing unit 502 as shown in Figure 2

[0114] For example, the computer program 603 can be divided into one or more modules / units, which are stored in the memory 602 and executed by the processor 601 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 603 in the electronic device 600. For example, the computer program 603 can be divided into Figure 2 the communication unit 501 and the processing unit 502 as shown in

[0115] ​The processor 601 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0116] The memory 602 can be an internal storage unit of the electronic device 600, for example, a hard disk or a memory of the electronic device 600. The memory 602 can also be an external storage device of the electronic device 600, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. provided on the electronic device 600. Further, the memory 602 can include both the internal storage unit and the external storage device of the electronic device 600. The memory 602 is used to store the computer program and other programs and data required by the terminal. The memory 602 can also be used to temporarily store data that has been output or will be output.

[0117] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit, and the integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0118] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0119] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0120] In the embodiments provided by the present application, it should be understood that the disclosed apparatus / terminal and method can be implemented in other ways. For example, the apparatus / terminal embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling or direct coupling or communication connection between each other can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or in other forms.

[0121] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0122] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0123] The integrated module / unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0124] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A signal processing method, characterized by, The method comprises the following steps: obtaining rail audio signals collected during train operation and track information of the train; performing spectral analysis and sliding window segmentation on the rail audio signals to separate normal components and high-frequency components of the rail audio signals, and obtaining high-frequency components of multiple time periods; based on the high-frequency components of each time period and a preset rail state prediction model, obtaining rail states corresponding to the high-frequency components of each time period; the rail state includes normal state, stress state and existence of hidden injury state; based on the rail states corresponding to the high-frequency components of each time period and the track information of the train, determining a section with operation risk in the train route; the method based on the high-frequency components of each time period and the preset rail state prediction model to obtain the rail states corresponding to the high-frequency components of each time period comprises the following steps:

2. The signal processing method of claim 1, wherein, performing spectral analysis and grayscale processing on the high-frequency components of each time period to obtain multiple grayscale images; performing wavelet transform and inverse wavelet transform on the multiple grayscale images to enhance the high-frequency part of each grayscale image and obtain multiple high-frequency images; inputting the multiple high-frequency images into the preset rail state prediction model to obtain the rail states corresponding to the high-frequency components of each time period. Before the method based on the high-frequency components of each time period and the preset rail state prediction model to obtain the rail states corresponding to the high-frequency components of each time period, the method further comprises the following steps: obtaining multiple rail audio signals with known rail states; generating training samples based on the multiple rail audio signals with known rail states; 3. The signal processing method of claim 2, wherein, based on the training samples, performing neural network training to obtain a preset rail state prediction model. each training sample in the training samples takes a high-frequency image as input and takes a rail state corresponding to the high-frequency image as output; the method based on the training samples and performing neural network training to obtain a preset rail state prediction model comprises the following steps: based on each training sample, taking a high-frequency image as input and taking a rail state as output, performing neural network training to obtain a first prediction model; taking the first prediction model as a teacher model to construct a teacher-student model: the teacher model and the student model in the teacher-student model are neural network models with the same depth; 4. The signal processing method of claim 3, wherein, based on the detection accuracy of the teacher model and the student model, performing multi-scale knowledge distillation training and pruning quantization on the student model, and determining the trained student model as the preset rail state prediction model. the method based on the detection accuracy of the teacher model and the student model, performing multi-scale knowledge distillation training and pruning quantization on the student model, and determining the trained student model as the preset rail state prediction model comprises the following steps: based on the detection accuracy of the teacher model and the student model, performing multi-scale knowledge distillation training on the student model; calculating the number of channels and the weight values of each normalization layer in the trained student model; deleting the channels with weight values less than a threshold in each normalization layer to obtain a simplified student model, wherein at least one channel is reserved in each normalization layer; based on the training samples, retraining the simplified student model, and determining the trained student model as the preset rail state prediction model.

5. The signal processing method of claim 1, wherein, The method further comprises: if the rail state corresponding to the high-frequency component of the time period is a stress state, inputting the high-frequency component of the time period into a preset stress prediction model to obtain a stress corresponding to the high-frequency component of the time period; calculating a stress level corresponding to the high-frequency component of the time period; determining a rail position with stress based on the time period in which the high-frequency component is located and the track information of the train; generating a first instruction based on the stress level corresponding to the high-frequency component of the time period and the rail position with stress, the first instruction being used to instruct an operation and maintenance personnel to detect the rail position with stress by using a track inspection vehicle; sending the first instruction to a user.

6. The signal processing method of claim 5, wherein, The method further comprises: obtaining a driving video captured during driving of the train; extracting a video clip corresponding to the rail position with stress and analyzing to obtain a tie laying condition and a rail deformation condition; determining a running risk level of the rail position with stress based on the tie laying condition and the rail deformation condition; determining an emergency level of the first instruction based on the running risk level of the rail position with stress.

7. The signal processing method of claim 5, wherein, The method further comprises: receiving a first response, the first response including a detection result of the rail position with stress by the track inspection vehicle; if the detection result is normal, retraining the rail state prediction model based on the high-frequency component of the time period and the normal state corresponding to the high-frequency component of the time period to obtain a retrained rail state prediction model, and detecting a rail state based on the retrained rail state prediction model; if the detection result is a stress state or a dark injury state, generating a maintenance instruction to instruct the user to maintain the rail position with stress.

8. A signal processing device, characterized by The method comprises: a communication unit configured to obtain a rail audio signal collected during driving of a train and track information of the train; a processing unit configured to perform spectral analysis and sliding window segmentation on the rail audio signal, separate normal components and high-frequency components of the rail audio signal, and obtain high-frequency components of multiple time periods; obtaining a rail state corresponding to the high-frequency components of the multiple time periods based on the high-frequency components of the multiple time periods and a preset rail state prediction model, the rail state including a normal state, a stress state, and a dark injury state; determining a section with a running risk in a route of the train based on the rail state corresponding to the high-frequency components of the multiple time periods and the track information of the train; the processing unit is specifically configured to perform spectral analysis and grayscale processing on the high-frequency components of the multiple time periods to obtain multiple grayscale images, perform wavelet transform and inverse wavelet transform on the multiple grayscale images to enhance high-frequency parts of the grayscale images, and obtain multiple high-frequency images; inputting the multiple high-frequency images into the preset rail state prediction model to obtain the rail state corresponding to the high-frequency components of the multiple time periods.

9. An electronic device, comprising: A computer program product comprising a memory storing a computer program and a processor configured to invoke and run the computer program stored in the memory to perform the steps of the method according to any one of claims 1 to 7.

Images