Railway line inspection system based on high-speed pulse imaging

Abstract

The invention discloses a railway line inspection system based on high-speed pulse imaging, which relates to the technical field of railway line inspection and comprises an imaging module, a regulation and control module, a positioning module, a reconstruction module, an identification module, a storage module and a communication module. The imaging module is used for synchronously acquiring a multi-view image sequence of a track area through a plurality of groups of high-speed pulse cameras and outputting original image data; the regulation and control module is used for dynamically adjusting the exposure parameters of the next frame according to the regional brightness deviation of the previous frame of image output by the imaging module, and feeding back the exposure parameters to the imaging module for execution, so as to realize illumination self-adaption of continuous imaging; the positioning module is used for fusing GNSS, IMU and odometer data, generating high-precision position information by adopting a filtering algorithm, performing correction in combination with orbit characteristics when satellite signals are missing, and outputting geographic coordinates corresponding to each frame of image; and the reconstruction module is used for correcting and matching the multi-view image output by the imaging module to generate a disparity map.

Description

Technical Field

[0001] This invention relates to the field of railway line inspection technology, and in particular to a railway line inspection system based on high-speed pulse imaging. Background Technology

[0002] Railway line inspection technology is a technical system designed to ensure safe train operation. It comprehensively utilizes manual labor, mobile equipment, and fixed monitoring methods to regularly inspect, monitor, and evaluate the condition of railway infrastructure such as rails, roadbed, and overhead contact lines. Its core has evolved from the traditional manual "walking and visual inspection" model to a modern model that primarily relies on dynamic detection by mobile equipment such as integrated inspection vehicles and flaw detectors, supplemented by satellite remote sensing, drone inspections, and intelligent perception through ground sensor networks. This enables more efficient, accurate, and intelligent detection and early warning of line defects, making it a lifeline for ensuring safe and efficient railway transportation. Therefore, how to utilize advanced technologies to improve the intelligence level and safety of railway line inspection has become one of the most pressing issues to be addressed.

[0003] In the field of railway line inspection, existing visual inspection systems are prone to overexposure or underexposure of images in scenarios such as trains passing through tunnels, entering and exiting platforms, and alternating sunny and cloudy weather due to drastic changes in lighting. This results in the loss of details in key areas, affecting the accuracy of subsequent identification. Furthermore, GNSS signals are easily interrupted under tunnels and viaducts, and traditional inertial navigation systems suffer from integral drift, leading to the accumulation of dead reckoning errors over a long period of time, making it impossible to accurately mark the geographical coordinates of defects. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a railway line inspection system based on high-speed pulse imaging to solve the problems of existing visual inspection systems, which are prone to overexposure or underexposure of images due to drastic changes in lighting conditions such as trains passing through tunnels, entering and exiting platforms, and alternating sunny and cloudy weather. This leads to the loss of details in key areas, affecting the accuracy of subsequent identification. Furthermore, GNSS signals are easily interrupted under tunnels and viaducts, and traditional inertial navigation systems suffer from integral drift, resulting in the accumulation of dead reckoning errors over a long period of time, making it impossible to accurately mark the geographical coordinates of defects.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a railway line inspection system based on high-speed pulse imaging, comprising:

[0008] Imaging module, control module, positioning module, reconstruction module, recognition module, storage module, and communication module;

[0009] The imaging module is used to simultaneously acquire multi-view image sequences of the orbital area using multiple sets of high-speed pulse cameras and output raw image data.

[0010] The control module is used to dynamically adjust the exposure parameters of the next frame based on the regional brightness deviation of the previous frame image output by the imaging module, and feed the adjustment back to the imaging module for execution, so as to achieve continuous imaging with adaptive illumination.

[0011] The positioning module is used to fuse GNSS, IMU and odometry data, use filtering algorithms to generate high-precision location information, and combine orbital features to correct for missing satellite signals, and output the geographic coordinates corresponding to each frame of image;

[0012] The reconstruction module is used to correct and match the multi-view images output by the imaging module, generate a disparity map, calculate the depth based on the binocular geometric relationship, and construct a three-dimensional point cloud model of the track component.

[0013] The identification module is used to compare the three-dimensional point cloud output by the reconstruction module with the standard model to obtain deformation parameters. At the same time, it performs target detection on the original image and cross-validates the two-dimensional results with the three-dimensional deformation to confirm the disease type and location.

[0014] The storage module is used to bind the disease information output by the identification module with the coordinates output by the positioning module to generate a structured record with location tags. Normal data is stored with high compression, while abnormal data retains the original image and point cloud.

[0015] The communication module is used to upload the inspection data in the storage module to the ground center system when the train stops.

[0016] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the control module dynamically adjusts the exposure parameters of the next frame according to the regional brightness deviation of the previous frame image output by the imaging module. The specific steps are as follows:

[0017] The previous frame image, acquired and output by the imaging module, is divided into a set of M rows and N columns of region blocks according to the spatial layout. Each region block corresponds to a local imaging area in the image. This division method ensures that the image is uniformly covered in space and avoids the failure of overall exposure control due to sudden changes in local illumination.

[0018] For each divided region block, extract the gray values ​​of all pixels within it, and perform an arithmetic mean operation on these gray values ​​to obtain the average gray value of the region block, which is used as the brightness characterization value of the region.

[0019] The brightness characterization value of each region block is compared with the target brightness value preset by the system to calculate the difference, and the brightness deviation of each region is obtained.

[0020] To improve the spatial adaptability of exposure adjustment, the brightness deviation of all regions is assigned different weights according to their position in the image. Regions closer to the image center have higher weights, while edge regions have lower weights, reflecting the central position of the railway track in the field of view. After weighting, a summation operation is performed to generate a comprehensive error vector.

[0021] The comprehensive error vector is input into the nonlinear feedback function to dynamically generate the exposure adjustment amount Δt, expressed as:

[0022]

[0023] Among them, E v E is the magnitude of the comprehensive error vector calculated for the current period. v0 σ represents the comprehensive error amplitude of the previous period, used to characterize the error change trend. E α is the standard deviation of the error over several recent periods, reflecting the stability of error fluctuations; α is the proportional gain coefficient, which controls the response intensity; β is the nonlinear enhancement factor, used to improve the adjustment sensitivity when the error changes abruptly, and is a very small positive number to prevent the denominator from being zero and causing calculation abnormalities.

[0024] The calculated exposure adjustment is added to the currently used exposure time to obtain a new exposure time value, which is used as the exposure control parameter for the next frame image acquisition.

[0025] The exposure time is fed back to the high-speed pulse camera in the imaging module via the control bus, which then performs the exposure parameter update operation to achieve adaptive control of illumination during the continuous imaging process.

[0026] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the positioning module fuses GNSS, IMU, and odometer data to generate high-precision location information, and the specific steps are as follows:

[0027] In each data acquisition cycle, the absolute geographic coordinates output by the GNSS module, the three-dimensional angular velocity and three-dimensional acceleration output by the IMU module, and the vehicle displacement increment output by the odometer are obtained. All data are timestamped.

[0028] A time synchronization algorithm is used to align the data from the above multi-source sensors to ensure that all data items correspond to the same time and eliminate time misalignment caused by different sampling frequencies.

[0029] Using the optimal state estimate from the previous moment, combined with the angular velocity and acceleration measured by the IMU at the current moment, the state is predicted using the inertial propagation equation, and the prior state at the current moment k is calculated.

[0030] The aligned GNSS position and odometer displacement increment are used as observation inputs to form a multi-source observation vector, which is then substituted into the extended Kalman filter (EKF) framework.

[0031] In the Extended Kalman Filter (EKF) framework, the Kalman gain is calculated, and its value depends on the ratio of the system prior covariance to the observation noise covariance, which is used to balance the confidence of the predicted value and the observed value.

[0032] By using Kalman gain to correct the prior state, the optimal state estimate for the current time step is obtained. The update process satisfies the following expression:

[0033]

[0034] Among them, C(e) img (e) represents the position error vector e generated by matching the orbital feature points extracted by the imaging module with the map database. img The constructed compensation term, γ, is the visual aid weight coefficient, used to enhance the contribution of visual information when the GNSS signal is weak or interrupted, and H is the observation mapping matrix, which maps the state vector to the observation space.

[0035] During the long-term interruption of GNSS signals, the system switches to dead reckoning mode, using the current best state estimate as the initial value, continuously integrating the IMU angular velocity to update the attitude, and doubly integrating the acceleration to update the velocity and position. At the same time, it integrates the displacement increment output by the odometry to constrain the velocity and prevent the integral drift from being too fast.

[0036] To suppress the cumulative error of dead reckoning, images acquired by the imaging module are used to detect kilometer markers, switches, or fixed landmarks along the track using image recognition algorithms, and their known geographical coordinates are obtained.

[0037] The known coordinates are injected into the filter as pseudo-observations, forming an observation residual with the current calculated position. The state is then corrected through the filter update equation, thereby actively suppressing the positioning drift.

[0038] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the reconstruction module corrects and matches the multi-view images output by the imaging module to construct a three-dimensional point cloud model. The specific steps are as follows:

[0039] Acquire the raw images synchronously output by the left and right cameras in the imaging module;

[0040] Using a pre-calibrated camera intrinsic parameter matrix and distortion coefficient vector, distortion correction is performed on the left and right images. The distorted pixels are mapped to the ideal imaging plane through a pixel coordinate mapping function to obtain a geometrically distortion-free corrected image.

[0041] Select a pixel to be matched in the left corrected image, and find its corresponding matching point on the same horizontal scan line in the right corrected image;

[0042] Calculate the sum of squared gray-level differences within the local window of two pixels as the basic matching cost, expressing... Regions with rich textures have high weights, while flat regions have low weights. An exponential decay term is introduced to suppress edge mismatches, generating the final aggregation cost, which is calculated according to the following expression:

[0043] Calculate the corresponding aggregation cost S(p,d) for all candidate disparities d, and select the d that minimizes the cost as the optimal disparity;

[0044] Based on the geometric model of a binocular camera, the parallax is converted into object depth Z = (f·B) / d(x,y) using the focal length and baseline length;

[0045] By combining each pixel (x, y) in the image with its corresponding depth Z, its 3D coordinates X = (x, y) are calculated using the camera projection model. x )·Z / f,Y=(yc y )·Z / f, where c x c y Principal point coordinates;

[0046] The initial point cloud is formed by collecting all three-dimensional coordinate points.

[0047] Using a pre-calibrated camera extrinsic matrix, the point cloud is transformed from the camera coordinate system to the vehicle coordinate system, resulting in a 3D point cloud model in a unified space.

[0048] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the identification module compares the three-dimensional point cloud output by the reconstruction module with a standard model and combines it with the detection results of the original image to confirm defects. The specific steps are as follows:

[0049] Obtain the 3D point cloud output by the reconstruction module and the 3D model of the standard fastener or track component pre-existing in the database;

[0050] The iterative nearest point algorithm is used to spatially register the 3D point cloud or the 3D model of the track component. The optimal rigid body transformation matrix is ​​calculated through iterative optimization to minimize the sum of squared point-to-point distances between the transformed point cloud and the model.

[0051] The 3D point cloud is transmitted via T align Transform to the standard model coordinate system to achieve spatial alignment;

[0052] In the aligned state, for each point in the point cloud, find the nearest corresponding point on the standard model and calculate the Euclidean distance δ between them. i=∥p i -m i ∥;

[0053] From all δ i Extract the maximum value from the data and use it as the deformation evaluation index for the component.

[0054] A deep learning-based object detection algorithm is executed on the raw image output by the imaging module to identify whether there are target parts in the image, and the detection result and confidence level C are output. det ;

[0055] δ max With C det Perform joint discrimination δ max >θ and C det If the value is less than τ, then the component is considered missing.

[0056] If δ max >θ and C det If the value is ≥τ, it is determined that the component is loose or misaligned.

[0057] The output includes structured recognition results containing disease type, geographical location, and maximum offset. The judgment logic is controlled by the following fusion criterion, expressed as:

[0058]

[0059] Where δ0 is the baseline deformation variable used for normalization, λ is the fusion weight, which balances the contributions of three-dimensional deformation and two-dimensional detection, and when D>κ, it is confirmed as a valid disease.

[0060] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the storage module binds and classifies the defect information output by the identification module with the coordinates output by the positioning module, and stores them accordingly. The specific steps are as follows:

[0061] Each disease record output by the identification module is bound to the corresponding geographic coordinates provided by the positioning module at that time, generating a structured data item with a location label;

[0062] Classify and determine the type of disease based on the structured data items with location labels;

[0063] If the wear is normal or slight, it will enter the high-compression storage process;

[0064] If the data is missing, broken, or severely loose, then proceed with the complete data retention process.

[0065] In the compression and storage process, the compression encoder is invoked to perform JPEG2000 lossy compression on the associated original image. The compression ratio is dynamically set according to the importance of the track segment, and its value is determined by the following model, expressed as:

[0066]

[0067] in, The base compression ratio is G0, which is the center coordinate of key sections such as bridges and tunnels. μ and ν are adjustment parameters. The closer to G0, the lower the compression ratio and the more details are preserved.

[0068] In the complete storage process, the original image, 3D point cloud, recognition results and their coordinates are written to the abnormal data area in their original format without lossy processing;

[0069] Establish a dual index table based on mileage markers and timestamps to enable fast querying and backtracking by segment, time, and disease type.

[0070] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the communication module uploads the inspection data in the storage module to the ground center system when the train stops. The specific steps are as follows:

[0071] After the train enters the platform or designated stopping area, the communication module activates the network detection mechanism to periodically scan the surrounding available wireless communication resources, including 4G, 5G cellular networks and Wi-Fi access points.

[0072] Obtain signal strength, channel quality, and available bandwidth information for each access point, and assess connection stability;

[0073] The transmission strategy is determined based on the available bandwidth and the total amount of data to be uploaded: when the available bandwidth is greater than or equal to 10Mbps, it is determined to be a high-bandwidth environment, and the high-speed direct transmission mode is activated to directly establish an encrypted connection with the ground center system and start transmission; when the available bandwidth is less than 10Mbps, it is determined to be a low-bandwidth or unstable environment, and the segmented caching upload mode is activated to package the data in batches and upload them segment by segment.

[0074] The inspection data organized by the operation route in the storage module is divided into multiple logical data packets, and each data packet corresponds to a complete inspection record of a continuous segment.

[0075] A metadata header is constructed for each data packet, containing management information such as the packet's start mileage, end mileage, start and end times of data collection, total data volume, device number, and checksum.

[0076] The entire data packet is encrypted using AES-256 encryption, employing a dynamic key provided by the security key management system.

[0077] A secure communication channel is established with the ground center system receiving interface through the TLS secure transport layer protocol. After verifying the legality of the server's digital certificate, the encrypted data packets are pushed to the designated data receiving server one by one.

[0078] During the upload process, the network status is continuously monitored. If a signal interruption or connection abnormality is detected, the transmission is paused and enters a waiting-to-recovery state, while retaining the current transmission progress information.

[0079] Once the next train stops and a valid connection is re-established, the resume transmission mechanism will be automatically triggered to seamlessly continue uploading the incomplete data.

[0080] As a preferred embodiment of the railway line inspection system based on high-speed pulse imaging described in this invention, the communication module executes a breakpoint resumption mechanism when a network interruption occurs during the upload process. The specific steps are as follows:

[0081] The communication link status is monitored in real time during the data upload process. Once a network connection interruption or data transmission timeout is detected, the current transmission task is stopped immediately.

[0082] Record the length of the successfully sent data as the amount of data transmitted, and at the same time calculate the starting offset address of the remaining data to be transmitted.

[0083] The amount of data already transmitted and the starting address of the remaining data are combined into status information and written into a dedicated status file in the local non-volatile storage unit;

[0084] When the train stops again and detects an available wireless network again, the communication module prioritizes reading the breakpoint information in the local status file.

[0085] Locate the data segment in the storage module that has not yet been transmitted based on the starting address read;

[0086] The untransmitted data segment is repackaged into a sub-data packet, a new metadata header is constructed to indicate that it is a continuation segment, and the unique identifier of the original data packet is retained to ensure data consistency.

[0087] Perform the same encryption process on the sub-data packets as the initial transmission to generate encrypted ciphertext;

[0088] The encrypted sub-data packets were re-uploaded to the ground center system via a TLS secure channel;

[0089] After receiving the resumed data, the ground system decrypts and splices the data, and returns an acknowledgment response containing a data integrity verification value.

[0090] After receiving the confirmation response, the communication module extracts the verification value and compares it with the data fingerprint generated from the local original data;

[0091] If the two match, the transmission is considered successful, and the temporary data and status file in the local cache are cleared.

[0092] If there is a discrepancy, the data is considered abnormal, and the upload request for that data segment is re-initiated until the verification passes.

[0093] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the railway line inspection system based on high-speed pulse imaging as described in the first aspect of the present invention.

[0094] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the railway line inspection system based on high-speed pulse imaging as described in the first aspect of the present invention.

[0095] The beneficial effects of this invention are as follows: By integrating multiple modules such as imaging, control, positioning, reconstruction, recognition, storage, and communication, high-precision, all-weather, and automated inspection of railway lines is achieved. The system uses a high-speed pulse camera to achieve multi-view synchronous imaging. Combined with a dynamic exposure control mechanism based on regional brightness deviation, it effectively overcomes the problem of overexposure or underexposure of images under complex lighting conditions, ensuring stable image quality. By fusing GNSS, IMU, and odometer data and introducing a filtering algorithm that corrects visual features, it can still maintain high-precision positioning under conditions of missing satellite signals, ensuring that the location of defects can be traced. By using binocular stereo matching and three-dimensional point cloud reconstruction technology, combined with cross-validation of standard model comparison and two-dimensional image detection results, the accuracy and reliability of track component deformation recognition are significantly improved. Attached Figure Description

[0096] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0097] Figure 1 This is a schematic diagram of the railway line inspection system based on high-speed pulse imaging in Example 1. Detailed Implementation

[0098] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0099] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0100] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0101] Example, refer to Figure 1 This embodiment of the invention provides a railway line inspection system based on high-speed pulse imaging, comprising:

[0102] Imaging module, control module, positioning module, reconstruction module, recognition module, storage module, and communication module;

[0103] The imaging module is used to simultaneously acquire multi-view image sequences of the orbital area using multiple sets of high-speed pulse cameras, and output raw image data.

[0104] Furthermore, the imaging module uses a high frame rate camera to achieve an image acquisition capability of 22,000 frames per second, ensuring the synchronous acquisition of multi-view image sequences in the orbital area;

[0105] The camera outputs raw image data and supports high-speed acquisition mode to meet the needs of real-time monitoring and rapid response.

[0106] By optimizing the image transmission path, latency is reduced, ensuring efficient processing of the data stream;

[0107] It should be noted that the high-speed pulse camera in the imaging module adopts a global shutter design and has microsecond-level exposure control capability. It can effectively freeze track images and suppress motion blur under high-speed train operation conditions. The camera array is arranged at multiple points along the transverse and longitudinal directions of the vehicle body, covering key areas of the track structure, including the rail surface, fastening system, sleepers and track bed. It also supports a synchronous triggering mechanism to ensure that multi-view images are strictly aligned in time, providing a high-quality data foundation for subsequent stereo matching and 3D reconstruction.

[0108] The control module is used to dynamically adjust the exposure parameters of the next frame based on the regional brightness deviation of the previous frame image output by the imaging module, and feed the results back to the imaging module for execution, so as to achieve continuous imaging with adaptive illumination.

[0109] Furthermore, the previous frame image acquired and output by the imaging module is divided into a set of M rows and N columns of region blocks according to the spatial layout. Each region block corresponds to a local imaging area in the image. This division method ensures that the image is uniformly covered in space and avoids the failure of overall exposure control due to sudden changes in local illumination.

[0110] For each divided region block, extract the gray values ​​of all pixels within it, and perform an arithmetic mean operation on these gray values ​​to obtain the average gray value of the region block, which is used as the brightness characterization value of the region.

[0111] The brightness characterization value of each region block is compared with the target brightness value preset by the system to calculate the difference, and the brightness deviation of each region is obtained.

[0112] To improve the spatial adaptability of exposure adjustment, the brightness deviation of all regions is assigned different weights according to their position in the image. Regions closer to the image center have higher weights, while edge regions have lower weights, reflecting the central position of the railway track in the field of view. After weighting, a summation operation is performed to generate a comprehensive error vector.

[0113] The comprehensive error vector is input into the nonlinear feedback function to dynamically generate the exposure adjustment amount Δt, expressed as:

[0114]

[0115] Among them, E v E is the magnitude of the comprehensive error vector calculated for the current period. v0 σ represents the comprehensive error amplitude of the previous period, used to characterize the error change trend. E α is the standard deviation of the error over several recent periods, reflecting the stability of error fluctuations; α is the proportional gain coefficient, which controls the response intensity; β is the nonlinear enhancement factor, used to improve the adjustment sensitivity when the error changes abruptly, and is a very small positive number to prevent the denominator from being zero and causing calculation abnormalities.

[0116] The calculated exposure adjustment is added to the currently used exposure time to obtain a new exposure time value, which is used as the exposure control parameter for the next frame image acquisition.

[0117] The exposure time is fed back to the high-speed pulse camera in the imaging module via the control bus, which then performs the exposure parameter update operation to achieve adaptive control of illumination during the continuous imaging process.

[0118] It should be noted that by dividing the previous frame image output by the imaging module into M rows and N columns of region blocks and calculating the brightness deviation of each region, and dynamically adjusting the exposure parameters by combining spatial weighting and nonlinear feedback mechanisms, it can effectively cope with complex scenes such as uneven lighting and sudden changes in brightness along the track, avoid overall overexposure or underexposure, significantly improve the contrast and detail visibility of continuous image sequences, provide high-quality visual input for subsequent image processing, and reduce system energy consumption and data redundancy caused by repeated acquisition.

[0119] The positioning module is used to fuse GNSS, IMU and odometry data, use filtering algorithms to generate high-precision location information, and combine orbital features to correct for missing satellite signals, and output the geographic coordinates corresponding to each frame of image.

[0120] Furthermore, in each data acquisition cycle, the absolute geographic coordinates output by the GNSS module, the three-dimensional angular velocity and three-dimensional acceleration output by the IMU module, and the vehicle displacement increment output by the odometer are obtained, and all data are timestamped.

[0121] A time synchronization algorithm is used to align the data from the above multi-source sensors to ensure that all data items correspond to the same time and eliminate time misalignment caused by different sampling frequencies.

[0122] Using the optimal state estimate from the previous moment, combined with the angular velocity and acceleration measured by the IMU at the current moment, the state is predicted using the inertial propagation equation, and the prior state at the current moment k is calculated.

[0123] The aligned GNSS position and odometer displacement increment are used as observation inputs to form a multi-source observation vector, which is then substituted into the extended Kalman filter (EKF) framework.

[0124] In the Extended Kalman Filter (EKF) framework, the Kalman gain is calculated, and its value depends on the ratio of the system prior covariance to the observation noise covariance, which is used to balance the confidence of the predicted value and the observed value.

[0125] By using Kalman gain to correct the prior state, the optimal state estimate for the current time step is obtained. The update process satisfies the following expression:

[0126]

[0127] Among them, C(e) img (e) represents the position error vector e generated by matching the orbital feature points extracted by the imaging module with the map database. img The constructed compensation term, γ, is the visual aid weight coefficient, used to enhance the contribution of visual information when the GNSS signal is weak or interrupted, and H is the observation mapping matrix, which maps the state vector to the observation space.

[0128] During the long-term interruption of GNSS signals, the system switches to dead reckoning mode, using the current best state estimate as the initial value, continuously integrating the IMU angular velocity to update the attitude, and doubly integrating the acceleration to update the velocity and position. At the same time, it integrates the displacement increment output by the odometry to constrain the velocity and prevent the integral drift from being too fast.

[0129] To suppress the cumulative error of dead reckoning, images acquired by the imaging module are used to detect kilometer markers, switches, or fixed landmarks along the track using image recognition algorithms, and their known geographical coordinates are obtained.

[0130] The known coordinates are injected into the filter as pseudo-observations, forming observation residuals with the current calculated position. The state is then corrected through the filter update equation, thereby actively suppressing positioning drift.

[0131] It should be noted that by fusing GNSS, IMU, and odometer data, and introducing a visual compensation term generated by track feature matching into the filtering algorithm, high-precision position information can be continuously output in areas where satellite signals are blocked or weakened. This effectively suppresses the cumulative error in dead reckoning, ensures that each frame of image has reliable spatial coordinates, improves the accuracy and traceability of defect location, and meets the stringent geographical accuracy requirements of railway inspection.

[0132] The reconstruction module is used to correct and match the multi-view images output by the imaging module, generate a disparity map, calculate the depth based on the binocular geometric relationship, and construct a three-dimensional point cloud model of the track component.

[0133] Furthermore, it acquires the raw images synchronously output by the left and right cameras in the imaging module;

[0134] Using a pre-calibrated camera intrinsic parameter matrix and distortion coefficient vector, distortion correction is performed on the left and right images. The distorted pixels are mapped to the ideal imaging plane through a pixel coordinate mapping function to obtain a geometrically distortion-free corrected image.

[0135] Select a pixel to be matched in the left corrected image, and find its corresponding matching point on the same horizontal scan line in the right corrected image;

[0136] Calculate the sum of squared gray-level differences within the local window of two pixels as the basic matching cost, expressing...

[0137] Regions with rich textures have high weights, while flat regions have low weights. An exponential decay term is introduced to suppress edge mismatches, generating the final aggregation cost, which is calculated according to the following expression:

[0138]

[0139] Calculate the corresponding aggregation cost S(p,d) for all candidate disparities d, and select the d that minimizes the cost as the optimal disparity;

[0140] Based on the geometric model of a binocular camera, the parallax is converted into object depth Z = (f·B) / d(x,y) using the focal length and baseline length;

[0141] By combining each pixel (x, y) in the image with its corresponding depth Z, its 3D coordinates X = (x, y) are calculated using the camera projection model. x )·Z / f,Y=(yc y )·Z / f, where c x c y Principal point coordinates;

[0142] The initial point cloud is formed by collecting all three-dimensional coordinate points.

[0143] Using a pre-calibrated camera extrinsic matrix, the point cloud is transformed from the camera coordinate system to the vehicle coordinate system to obtain a 3D point cloud model in a unified space;

[0144] It should be noted that after distortion correction of multi-view images, a neighborhood gradient-weighted aggregation cost function is used for stereo matching, which improves the robustness of matching weak textures and edge regions and generates more accurate disparity maps. By combining the depth calculation of the binocular geometric model and constructing a three-dimensional point cloud, high-fidelity reconstruction of the surface morphology of the track components is achieved, providing a reliable three-dimensional geometric basis for subsequent millimeter-level deformation detection and enhancing the system's ability to identify subtle defects.

[0145] The identification module is used to compare the 3D point cloud output by the reconstruction module with the standard model to obtain deformation parameters. At the same time, it performs target detection on the original image and cross-validates the 2D results with the 3D deformation to confirm the type and location of the disease.

[0146] Furthermore, by utilizing AI intelligent systems to analyze image and point cloud data, the system can automatically identify defect types and provide accurate classification results.

[0147] Automatically mark the location of defects, and combine the geographic coordinates provided by the positioning module to accurately mark the specific kilometer marker location where the disease occurred;

[0148] By combining deep learning algorithms, we can improve recognition accuracy, reduce manual intervention, and increase work efficiency.

[0149] Obtain the 3D point cloud output by the reconstruction module and the 3D model of the standard fastener or track component pre-existing in the database;

[0150] The iterative nearest point algorithm is used to spatially register the 3D point cloud or the 3D model of the track component. The optimal rigid body transformation matrix is ​​calculated through iterative optimization to minimize the sum of squared point-to-point distances between the transformed point cloud and the model.

[0151] The 3D point cloud is transmitted via T align Transform to the standard model coordinate system to achieve spatial alignment;

[0152] In the aligned state, for each point in the point cloud, find the nearest corresponding point on the standard model and calculate the Euclidean distance δ between them. i =∥p i -m i ∥;

[0153] From all δ i Extract the maximum value from the data and use it as the deformation evaluation index for the component.

[0154] A deep learning-based object detection algorithm is executed on the raw image output by the imaging module to identify whether there are target parts in the image, and the detection result and confidence level C are output. det ;

[0155] δ max With C det Perform joint discrimination δ max >θ and C det If the value is less than τ, then the component is considered missing.

[0156] If δ max >θ and C det If the value is ≥τ, it is determined that the component is loose or misaligned.

[0157] The output includes structured recognition results containing disease type, geographical location, and maximum offset. The judgment logic is controlled by the following fusion criterion, expressed as:

[0158]

[0159] Where δ0 is the baseline deformation, used for normalization, λ is the fusion weight, which balances the contributions of three-dimensional deformation and two-dimensional detection, and when D>κ, it is confirmed as a valid disease;

[0160] It should be noted that by comparing the 3D point cloud with the standard model to obtain the maximum offset, and combining it with the target detection results of the original image for cross-validation, the type of disease is comprehensively judged through fusion criteria. This effectively reduces the false alarm and false negative rates caused by single-modal detection. This dual-modal fusion mechanism improves the accuracy of identifying typical diseases such as missing or loose parts, enhances the diagnostic reliability of the system in complex interference environments, and improves the credibility and practicality of inspection results.

[0161] The storage module is used to bind the disease information output by the identification module with the coordinates output by the positioning module, generate a structured record with location tags, store normal data with high compression, and retain the original image and point cloud for abnormal data.

[0162] Furthermore, the storage unit has high-speed storage capabilities, enabling it to quickly save a large number of raw images and analysis results generated by the imaging and recognition modules;

[0163] It supports high-speed video export, which facilitates post-event review and further analysis, ensuring the complete recording and rapid sharing of key information;

[0164] A hierarchical storage mechanism is introduced to distinguish between normal and abnormal data for storage, thereby improving data retrieval efficiency;

[0165] Each disease record output by the identification module is bound to the corresponding geographic coordinates provided by the positioning module at that time, generating a structured data item with a location label;

[0166] Classify and determine the type of disease based on the structured data items with location labels;

[0167] If the wear is normal or slight, it will enter the high-compression storage process;

[0168] If the data is missing, broken, or severely loose, then proceed with the complete data retention process.

[0169] In the compression and storage process, the compression encoder is invoked to perform JPEG2000 lossy compression on the associated original image. The compression ratio is dynamically set according to the importance of the track segment, and its value is determined by the following model, expressed as:

[0170]

[0171] in, The base compression ratio is G0, which is the center coordinate of key sections such as bridges and tunnels. μ and ν are adjustment parameters. The closer to G0, the lower the compression ratio and the more details are preserved.

[0172] In the complete storage process, the original image, 3D point cloud, recognition results and their coordinates are written to the abnormal data area in their original format without lossy processing;

[0173] Establish a dual index table based on mileage markers and timestamps to enable fast querying and backtracking by segment, time, and disease type;

[0174] It should be noted that a hierarchical storage strategy is implemented for data based on the severity of the disease. For normal data, an adaptive compression model based on geographic location sensitivity is used, retaining more details in key areas and increasing the compression rate in non-key areas. This significantly reduces the overall storage space usage while ensuring the integrity of key information. For abnormal data, the original images and point clouds are retained to ensure retrospective analysis, balancing storage efficiency and diagnostic needs, and extending the system's continuous operation capability without external storage support.

[0175] The communication module is used to upload the inspection data in the storage module to the ground center system when the train stops.

[0176] Furthermore, when the train stops, the communication module uploads the inspection data to the ground center system, including the results of automatic identification and the marked location information;

[0177] Implement the breakpoint resume function to ensure that unfinished data transmission tasks can be seamlessly resumed after a network interruption;

[0178] After the train enters the platform or designated stopping area, the communication module activates the network detection mechanism to periodically scan the surrounding available wireless communication resources, including 4G, 5G cellular networks and Wi-Fi access points.

[0179] Obtain signal strength, channel quality, and available bandwidth information for each access point, and assess connection stability;

[0180] The transmission strategy is determined based on the available bandwidth and the total amount of data to be uploaded: when the available bandwidth is greater than or equal to 10Mbps, it is determined to be a high-bandwidth environment, and the high-speed direct transmission mode is activated to directly establish an encrypted connection with the ground center system and start transmission; when the available bandwidth is less than 10Mbps, it is determined to be a low-bandwidth or unstable environment, and the segmented caching upload mode is activated to package the data in batches and upload them segment by segment.

[0181] The inspection data organized by the operation route in the storage module is divided into multiple logical data packets, and each data packet corresponds to a complete inspection record of a continuous segment.

[0182] A metadata header is constructed for each data packet, containing management information such as the packet's start mileage, end mileage, start and end times of data collection, total data volume, device number, and checksum.

[0183] The entire data packet is encrypted using AES-256 encryption, employing a dynamic key provided by the security key management system.

[0184] A secure communication channel is established with the ground center system receiving interface through the TLS secure transport layer protocol. After verifying the legality of the server's digital certificate, the encrypted data packets are pushed to the designated data receiving server one by one.

[0185] During the upload process, the network status is continuously monitored. If a signal interruption or connection abnormality is detected, the transmission is paused and enters a waiting-to-recovery state, while retaining the current transmission progress information.

[0186] Once the next train stops and a valid connection is re-established, the resume mechanism will be automatically triggered to enable seamless uploading of incomplete data.

[0187] When a network interruption occurs during the upload process, the communication module executes a breakpoint resume mechanism. The specific steps are as follows:

[0188] The communication link status is monitored in real time during the data upload process. Once a network connection interruption or data transmission timeout is detected, the current transmission task is stopped immediately.

[0189] Record the length of the successfully sent data as the amount of data transmitted, and at the same time calculate the starting offset address of the remaining data to be transmitted.

[0190] The amount of data already transmitted and the starting address of the remaining data are combined into status information and written into a dedicated status file in the local non-volatile storage unit;

[0191] When the train stops again and detects an available wireless network again, the communication module prioritizes reading the breakpoint information in the local status file.

[0192] Locate the data segment in the storage module that has not yet been transmitted based on the starting address read;

[0193] The untransmitted data segment is repackaged into a sub-data packet, a new metadata header is constructed to indicate that it is a continuation segment, and the unique identifier of the original data packet is retained to ensure data consistency.

[0194] Perform the same encryption process on the sub-data packets as the initial transmission to generate encrypted ciphertext;

[0195] The encrypted sub-data packets were re-uploaded to the ground center system via a TLS secure channel;

[0196] After receiving the resumed data, the ground system decrypts and splices the data, and returns an acknowledgment response containing a data integrity verification value.

[0197] After receiving the confirmation response, the communication module extracts the verification value and compares it with the data fingerprint generated from the local original data;

[0198] If the two match, the transmission is considered successful, and the temporary data and status file in the local cache are cleared.

[0199] If there is a discrepancy, it is determined to be a data anomaly, and the upload request for that data segment is re-initiated until the verification passes.

[0200] It should be noted that the communication module intelligently selects direct transmission or segmented buffer upload mode based on available bandwidth, and performs breakpoint resume transmission after interruption, ensuring that inspection data can still be transmitted completely and securely to the ground center in an unstable wireless environment. This mechanism significantly improves the success rate and efficiency of data upload, avoids data loss or duplicate transmission caused by network fluctuations, and ensures the data closed loop of inspection business and the timeliness of operation and maintenance response.

[0201] This embodiment also provides a computer device applicable to a railway line inspection system based on high-speed pulse imaging, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the railway line inspection system based on high-speed pulse imaging as proposed in the above embodiment.

[0202] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0203] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the railway line inspection system based on high-speed pulse imaging as proposed in the above embodiments. The storage medium 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 Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0204] In summary, this invention achieves high-precision, all-weather, and automated inspection of railway lines by integrating multiple modules such as imaging, control, positioning, reconstruction, recognition, storage, and communication. The system uses a high-speed pulse camera to achieve multi-view synchronous imaging, and combined with a dynamic exposure control mechanism based on regional brightness deviation, it effectively overcomes the problems of overexposure or underexposure of images under complex lighting conditions, ensuring stable image quality. By fusing GNSS, IMU, and odometer data and introducing a filtering algorithm that corrects visual features, it can still maintain high-precision positioning under conditions of missing satellite signals, ensuring that the location of defects can be traced. By using binocular stereo matching and three-dimensional point cloud reconstruction technology, combined with cross-validation of standard model comparison and two-dimensional image detection results, the accuracy and reliability of track component deformation recognition are significantly improved.

[0205] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A railway line inspection system based on high-speed pulse imaging, characterized in that: include: Imaging module, control module, positioning module, reconstruction module, recognition module, storage module, and communication module; The imaging module is used to simultaneously acquire multi-view image sequences of the orbital area using multiple sets of high-speed pulse cameras and output raw image data. The control module is used to dynamically adjust the exposure parameters of the next frame based on the regional brightness deviation of the previous frame image output by the imaging module, and feed the adjustment back to the imaging module for execution, so as to achieve continuous imaging with adaptive illumination. The positioning module is used to fuse GNSS, IMU and odometry data, use filtering algorithms to generate high-precision location information, and combine orbital features to correct for missing satellite signals, and output the geographic coordinates corresponding to each frame of image; The reconstruction module is used to correct and match the multi-view images output by the imaging module, generate a disparity map, calculate the depth based on the binocular geometric relationship, and construct a three-dimensional point cloud model of the track component. The identification module is used to compare the three-dimensional point cloud output by the reconstruction module with the standard model to obtain deformation parameters. At the same time, it performs target detection on the original image and cross-validates the two-dimensional results with the three-dimensional deformation to confirm the disease type and location. The storage module is used to bind the disease information output by the identification module with the coordinates output by the positioning module to generate a structured record with location tags. Normal data is stored with high compression, while abnormal data retains the original image and point cloud. The communication module is used to upload the inspection data in the storage module to the ground center system when the train stops.

2. The railway line inspection system based on high-speed pulse imaging as described in claim 1, characterized in that: The control module dynamically adjusts the exposure parameters of the next frame based on the regional brightness deviation of the previous frame image output by the imaging module. The specific steps are as follows: The previous frame image, acquired and output by the imaging module, is divided into a set of M rows and N columns of region blocks according to the spatial layout. Each region block corresponds to a local imaging area in the image. This division method ensures that the image is uniformly covered in space and avoids the failure of overall exposure control due to sudden changes in local illumination. For each divided region block, extract the gray values ​​of all pixels within it, and perform an arithmetic mean operation on these gray values ​​to obtain the average gray value of the region block, which is used as the brightness characterization value of the region. The brightness characterization value of each region block is compared with the target brightness value preset by the system to calculate the difference, and the brightness deviation of each region is obtained. To improve the spatial adaptability of exposure adjustment, the brightness deviation of all regions is assigned different weights according to their position in the image. Regions closer to the image center have higher weights, while edge regions have lower weights, reflecting the central position of the railway track in the field of view. After weighting, a summation operation is performed to generate a comprehensive error vector. The comprehensive error vector is input into the nonlinear feedback function to dynamically generate the exposure adjustment amount Δt, expressed as: Among them, E v E is the magnitude of the comprehensive error vector calculated for the current period. v0 The comprehensive error amplitude of the previous period is used to characterize the error change trend, σ. E α is the standard deviation of the error over several recent periods, reflecting the stability of error fluctuations; α is the proportional gain coefficient, which controls the response intensity; β is the nonlinear enhancement factor, used to improve the adjustment sensitivity when the error changes abruptly, and is a very small positive number to prevent the denominator from being zero and causing calculation abnormalities. The calculated exposure adjustment is added to the currently used exposure time to obtain a new exposure time value, which is used as the exposure control parameter for the next frame image acquisition. The exposure time is fed back to the high-speed pulse camera in the imaging module via the control bus, which then performs the exposure parameter update operation to achieve adaptive control of illumination during the continuous imaging process.

3. The railway line inspection system based on high-speed pulse imaging as described in claim 2, characterized in that: The positioning module fuses GNSS, IMU, and odometer data to generate high-precision location information. The specific steps are as follows: In each data acquisition cycle, the absolute geographic coordinates output by the GNSS module, the three-dimensional angular velocity and three-dimensional acceleration output by the IMU module, and the vehicle displacement increment output by the odometer are obtained. All data are timestamped. A time synchronization algorithm is used to align the data from the above multi-source sensors to ensure that all data items correspond to the same time and eliminate time misalignment caused by different sampling frequencies. Using the optimal state estimate from the previous moment, combined with the angular velocity and acceleration measured by the IMU at the current moment, the state is predicted using the inertial propagation equation, and the prior state at the current moment k is calculated. The aligned GNSS position and odometer displacement increment are used as observation inputs to form a multi-source observation vector, which is then substituted into the extended Kalman filter (EKF) framework. In the Extended Kalman Filter (EKF) framework, the Kalman gain is calculated, and its value depends on the ratio of the system prior covariance to the observation noise covariance, which is used to balance the confidence of the predicted value and the observed value. By using Kalman gain to correct the prior state, the optimal state estimate for the current time step is obtained. The update process satisfies the following expression: Among them, C(e) img (e) represents the position error vector e generated by matching the orbital feature points extracted by the imaging module with the map database. img The constructed compensation term, γ, is the visual aid weight coefficient, used to enhance the contribution of visual information when the GNSS signal is weak or interrupted, and H is the observation mapping matrix, which maps the state vector to the observation space. During the long-term interruption of GNSS signals, the system switches to dead reckoning mode, using the current best state estimate as the initial value, continuously integrating the IMU angular velocity to update the attitude, and doubly integrating the acceleration to update the velocity and position. At the same time, it integrates the displacement increment output by the odometry to constrain the velocity and prevent the integral drift from being too fast. To suppress the cumulative error of dead reckoning, images acquired by the imaging module are used to detect kilometer markers, switches, or fixed landmarks along the track using image recognition algorithms, and their known geographical coordinates are obtained. The known coordinates are injected into the filter as pseudo-observations, forming an observation residual with the current calculated position. The state is then corrected through the filter update equation, thereby actively suppressing the positioning drift.

4. The railway line inspection system based on high-speed pulse imaging as described in claim 3, characterized in that: The reconstruction module corrects and matches the multi-view images output by the imaging module to construct a three-dimensional point cloud model. The specific steps are as follows: Acquire the raw images synchronously output by the left and right cameras in the imaging module; Using a pre-calibrated camera intrinsic parameter matrix and distortion coefficient vector, distortion correction is performed on the left and right images. The distorted pixels are mapped to the ideal imaging plane through a pixel coordinate mapping function to obtain a geometrically distortion-free corrected image. Select a pixel to be matched in the left corrected image, and find its corresponding matching point on the same horizontal scan line in the right corrected image; The sum of squared grayscale differences within the local window containing two pixels is used as the basic matching cost, expressed as follows: The basic matching cost is weighted and aggregated. The weights ω(q) are adaptively adjusted according to the gradient characteristics of the neighborhood, with higher weights for textured regions and lower weights for flat regions. An exponential decay term is also introduced to suppress edge mismatches, generating the final aggregated cost, which is calculated according to the following expression: Where G(q) is the gradient magnitude at the neighborhood point q, η is the edge protection coefficient, which controls the attenuation intensity; calculate the corresponding aggregation cost S(p,d) for all candidate disparities d, and select the d that minimizes the cost as the optimal disparity; Based on the geometric model of a binocular camera, the parallax is converted into object depth Z = (f·B) / d(x,y) using the focal length and baseline length; By combining each pixel (x, y) in the image with its corresponding depth Z, its 3D coordinates X = (x, y) are calculated using the camera projection model. x )·Z / f,Y=(yc y )·Z / f, where c x c y Principal point coordinates; The initial point cloud is formed by collecting all three-dimensional coordinate points. Using a pre-calibrated camera extrinsic matrix, the point cloud is transformed from the camera coordinate system to the vehicle coordinate system, resulting in a 3D point cloud model in a unified space.

5. The railway line inspection system based on high-speed pulse imaging as described in claim 4, characterized in that: The identification module compares the 3D point cloud output by the reconstruction module with the standard model and combines it with the detection results of the original image to confirm the disease. The specific steps are as follows: Obtain the 3D point cloud output by the reconstruction module and the 3D model of the standard fastener or track component pre-existing in the database; The iterative nearest point algorithm is used to spatially register the 3D point cloud or the 3D model of the track component. The optimal rigid body transformation matrix is ​​calculated through iterative optimization to minimize the sum of squared point-to-point distances between the transformed point cloud and the model. The 3D point cloud is transmitted via T align Transform to the standard model coordinate system to achieve spatial alignment; In the aligned state, for each point in the point cloud, find the nearest corresponding point on the standard model and calculate the Euclidean distance δ between them. i =∥p i -m i ∥; From all δ i Extract the maximum value from the data and use it as the deformation evaluation index for the component. A deep learning-based object detection algorithm is executed on the raw image output by the imaging module to identify whether there are target parts in the image, and the detection result and confidence level C are output. det ; δ max With C det Perform joint discrimination δ max >θ and C det If the value is less than τ, then the component is considered missing. If δ max >θ and C det If the value is ≥τ, it is determined that the component is loose or misaligned. The output includes structured recognition results containing disease type, geographical location, and maximum offset. The judgment logic is controlled by the following fusion criterion, expressed as: Where δ0 is the baseline deformation variable used for normalization, λ is the fusion weight, which balances the contributions of three-dimensional deformation and two-dimensional detection, and when D>κ, it is confirmed as a valid disease.

6. The railway line inspection system based on high-speed pulse imaging as described in claim 5, characterized in that: The storage module binds and categorizes the disease information output by the identification module with the coordinates output by the positioning module. The specific steps are as follows: Each disease record output by the identification module is bound to the corresponding geographic coordinates provided by the positioning module at that time, generating a structured data item with a location label; Classify and determine the type of disease based on the structured data items with location labels; If the wear is normal or slight, it will enter the high-compression storage process; If the data is missing, broken, or severely loose, then proceed with the complete data retention process. In the compression and storage process, the compression encoder is invoked to perform JPEG2000 lossy compression on the associated original image. The compression ratio is dynamically set according to the importance of the track segment, and its value is determined by the following model, expressed as: in, The base compression ratio is G0, which is the center coordinate of key sections such as bridges and tunnels. μ and ν are adjustment parameters. The closer to G0, the lower the compression ratio and the more details are preserved. In the complete storage process, the original image, 3D point cloud, recognition results and their coordinates are written to the abnormal data area in their original format without lossy processing; Establish a dual index table based on mileage markers and timestamps to enable fast querying and backtracking by segment, time, and disease type.

7. The railway line inspection system based on high-speed pulse imaging as described in claim 6, characterized in that: The communication module uploads the inspection data stored in the storage module to the ground center system when the train stops. The specific steps are as follows: After the train enters the platform or designated stopping area, the communication module activates the network detection mechanism to periodically scan the surrounding available wireless communication resources, including 4G, 5G cellular networks and Wi-Fi access points. Obtain signal strength, channel quality, and available bandwidth information for each access point, and assess connection stability; The transmission strategy is determined based on the available bandwidth and the total amount of data to be uploaded: when the available bandwidth is greater than or equal to 10Mbps, it is determined to be a high-bandwidth environment, and the high-speed direct transmission mode is activated to directly establish an encrypted connection with the ground center system and start transmission; when the available bandwidth is less than 10Mbps, it is determined to be a low-bandwidth or unstable environment, and the segmented caching upload mode is activated to package the data in batches and upload them segment by segment. The inspection data organized by the operation route in the storage module is divided into multiple logical data packets, and each data packet corresponds to a complete inspection record of a continuous segment. A metadata header is constructed for each data packet, containing management information such as the packet's start mileage, end mileage, start and end times of data collection, total data volume, device number, and checksum. The entire data packet is encrypted using AES-256 encryption, employing a dynamic key provided by the security key management system. A secure communication channel is established with the ground center system receiving interface through the TLS secure transport layer protocol. After verifying the legality of the server's digital certificate, the encrypted data packets are pushed to the designated data receiving server one by one. During the upload process, the network status is continuously monitored. If a signal interruption or connection abnormality is detected, the transmission is paused and enters a waiting-to-recovery state, while retaining the current transmission progress information. Once the next train stops and a valid connection is re-established, the resume transmission mechanism will be automatically triggered to seamlessly continue uploading the incomplete data.

8. The railway line inspection system based on high-speed pulse imaging as described in claim 7, characterized in that: When a network interruption occurs during the upload process, the communication module executes a breakpoint resume mechanism. The specific steps are as follows: The communication link status is monitored in real time during the data upload process. Once a network connection interruption or data transmission timeout is detected, the current transmission task is stopped immediately. Record the length of the successfully sent data as the amount of data transmitted, and at the same time calculate the starting offset address of the remaining data to be transmitted. The amount of data already transmitted and the starting address of the remaining data are combined into status information and written into a dedicated status file in the local non-volatile storage unit; When the train stops again and detects an available wireless network again, the communication module prioritizes reading the breakpoint information in the local status file. Locate the data segment in the storage module that has not yet been transmitted based on the starting address read; The untransmitted data segment is repackaged into a sub-data packet, a new metadata header is constructed to indicate that it is a continuation segment, and the unique identifier of the original data packet is retained to ensure data consistency. Perform the same encryption process on the sub-data packets as the initial transmission to generate encrypted ciphertext; The encrypted sub-data packets were re-uploaded to the ground center system via a TLS secure channel; After receiving the resumed data, the ground system decrypts and splices the data, and returns an acknowledgment response containing a data integrity verification value. After receiving the confirmation response, the communication module extracts the verification value and compares it with the data fingerprint generated from the local original data; If the two match, the transmission is considered successful, and the temporary data and status file in the local cache are cleared. If there is a discrepancy, the data is considered abnormal, and the upload request for that data segment is re-initiated until the verification passes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the railway line inspection system based on high-speed pulse imaging as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the railway line inspection system based on high-speed pulse imaging as described in any one of claims 1 to 8.

Images